A tale of two Quads—followup

Given yesterdays’s conflicting Photoshop test results, I rebooted my PowerMac Quad (with 8GB ECC memory), and re-ran the tests. I quit and restarted Photoshop CS2 each time. This time I also tested with History = 1, in part to give the MacBook Pro a chance of having enough memory. Here are the figures:

PowerMac Quad, History =  1, cache levels = 6: 49/49 seconds (3 runs)
PowerMac Quad, History = 20, cache levels = 6: 80/85/88/90 seconds (4 runs)
MacBook Pro,   History =  1, cache levels = 6: 198/200/201 seconds (3 runs)

Using Photoshop’s History mechanism has a large impact. The variability using History=20 is disturbing, but no other programs were running, all sharing was off, and I didn’t even touch the mouse! Unfortunately, by not running Activity Monitor, it's impossible to see if any other system activity kicked in, but I heard no disk activity, and there were ample leftover CPU cycles.

I can’t explain yesterday’s slower timings other than a mistake (I’m not sure how), or that the system reboot fixed something. Still, I am unable to match the 38 second figure recorded by barefeats.com, although the figure with History=20 is only slightly higher than the barefeats History=20 result.

A tale of two Quads

On August 28, barefeats.com posted some performance numbers comparing the PowerMac G5 Quad to the Intel-based Mac Pro on a new “Retouch Artists” Photoshop benchmark.

Barefeats.com obtained a PowerMac G5 Quad timing (54 seconds, then 36 seconds in a revision) that is vastly different than mine. I’m not saying their figure is wrong—but it is completely different from my result, and there must be an explanation. My results were as follows:

run #1: 166 seconds, processor performance = auto
run #2: 165 seconds, processor performance = auto
run #3: 162 seconds, processor performance = high
run #4: 163 seconds, processor performance= high

PowerMac G5 Quad, 8GB RAM, 4-way striped RAID, CS2 history = 20, CS2 Cache Levels = 6, CS2 memory usage = 100% (3072MB), dual monitors, nVIDIA GeForce 6600 with 256MB memory. No other applications running except Activity Monitor. Time measured is clock (real) time.

I’m hoping barefeats.com is right and my copy of Photoshop will double in speed by reinstalling it! (I’m half serious, given the scrolling problems I’ve had with Photoshop). More than likely, the numbers reflect some Photoshop CS2 configuration. My feeling is that running a benchmark with artificial settings is silly, so I’ve stuck to the settings I normally use in everyday work.

Before we delve into details, let me state some principles to be borne in mind when reading benchmark results:

First caveat: always mistrust new, unproven benchmarks. Good benchmarks are extremely difficult to write. I know—I wrote benchmarks as a professional software developer for a time.

Second caveat: Take performance claims with a grain of salt until you see a consensus emerge.

Third caveat: Configuration and preferences can affect timings considerably.

Fourth caveat: Validate the results by testing yourself, if possible.

I trust barefeats.com, and I regularly correspond with Rob-Art, the site author. I’m working with him to determine why our results are so different.

Autofocus accuracy with the Canon EOS 1Ds Mark II and 85mm/f1.2L II USM

I’ve posted the next article in my series on focus accuracy: Canon EOS 1Ds Mark II / 85mm/f1.2L II USM.

Digital Infrared Class—October 25 at Keeble & Shucat

If you’re interested in shooting in infrared using a digital camera, don’t miss my Oct 25, 2006 class at Keeble & Shucat Photography, in Palo Alto, CA. For more information, please click the flyer below to see it at full size.

Digital Infrared article updated

I’ve improved the look of some photos and added a few more examples to my Digital Infrared article. I’ve also added additional links to other sites with information on infrared

You heard it here first—Surefire flashlights

The Wall Street Journal ran an article today (August 24, 2006) in its “Personal Journal” section entitled ‘Tactical’ Flashlights Emerge Into the Spotlight. I’ve previously extolled the merits of such lights for good reasons—they’re capable of lasting a lifetime, and produce a bright, smooth beam unlike ordinary household flashlights. Please see my Recommended Products and Vendors page for details, and don’t forget about the special discount of 5% at batteryjunction.com, exclusively through diglloyd.com!

Discount at batteryjunction.com

Diglloyd.com is pleased to offer diglloyd readers a special discount at batteryjunction.com. Please see the Recommended Products and Vendors page for details. As of this writing, diglloyd.com has no financial interest in this discount—it’s simply a fringe benefit to diglloyd.com readers!

BatteryJunction.com offers lithium ion batteries of all sorts, NiMH batteries, Alkaline batteries, SUREFILE and Streamlight and INOVA flashlights, as as various battery chargers, battery packs, camera and camcorder batteries, iPod accessories, inverters, converters, and other related items.

More on Mac Pro memory

AnandTech has an interesting article “Understanding FB DIMMS”, as part of its discussion of the new Apple Mac Pro (FB-DIMMS are the technical name for the type of memory module used by the Mac Pro).

It’s really an article for nerds, and I confess I’m not nerdy enough to really appreciate it all, but there are a few choice points made which are worthwhile to understand:

• Filling all 8 memory slots will in theory reduce memory performance over using just 4 slots due to the nature of the serial data transfer in an FB-DIMM—increased latency of 3-5 nanoseconds. Of course, the slots must be filled appropriately in both cases. So the ideal configurations are 4 X 512GB, 4 X 1GB, or 4 X 2GB. For most users, the price of buying 2GB FB-DIMMS is prohibitive. Whether any of this matters in practice is another matter—and the effects are likely to be very small.
  
  
• AnandTech states that each FB-DIMM consumes between 3 and 6 watts resulting in “~40W” with all 8 slots populated. The math varies because, according to AnandTech, “there’s a range is because how active the AMB is depends on how close it is to the memory controller”.

Since heat is a good way to get a system crash or other failure, it seems unwise to buy cheap memory that doesn’t have heat sinks—40 watts is a lot of heat in the small space occupied by the memory modules. I won’t be buying Mac Pro memory without the heat sinks.

Bottom line: stick to 4 or 8 modules, installed correctly, unless you have an unlimited budget. If you really only need 4GB, buy 4 X 1GB and be happy. If you need 8GB, buy 8 X 1GB and be happy. If you need 16GB (first verify that you need 16GB), be happy until the bill comes due.

Correction on MacPro memory prices at satech.com

In my August 18 entry, I described my past experiences with satech.com memory. Their original quoted price was apparently for compatible memory they’d previously sold for Dell, Acer, etc, without the Apple-approved heat sinks. After multiple back and forth messages with satech.com, the story is now clear. I apologize to any readers who relied on the original price to make a decision, as I do my best to provide reliable information. This was my mistake in not double-verifying the price and exact specifications.

The satech.com (aka ramfinder) price for 2GB (1 X 1GB) kit of memory for the MacPro is $439, or $220/GB That’s $1756 for 8GB as compared with $2300 for Apple memory (30% more). They describe it as “factory approved memory”:

This is the same memory product you would buy from your system manufacturer (Apple, Compaq, Dell, HP, IBM, Gateway, Sony and others) directly when you buy the system.

Please see the MacPro memory page at satech.com. Whether the $544 price difference (for 8 X 1GB) is worth it will depend on the personal situation; it is certainly easier to pre-order and have no further work to do.

In such situations, I like to look at the overall system cost to determine if it’s worth the hassle. Assuming you were to buy a new MacPro with 8GB of RAM, here are the system costs, and the price difference:

A price difference of 10-16% is nothing to snivel at, but for high-end users whose time is valuable, the convenience and assurance of Apple memory might well override that “FUD” factor.

Memory speed on the new MacPro

Memory in the Mac Pro is server-quality memory. In particular, it is ECC memory.

According to actual testing by barefeats.com (and an Apple tech note), full memory speed is not achieved unless 4 modules are installed correctly (“quad channel”). Please see “SPECIAL REPORT: What Everyone Should Know about Mac Pro Memory Upgrades” at barefeats.com for details.

According to barefeats.com, the speed gain ranges from nothing at all, to 25% faster on the 3.0 GHz model (slower CPU speeds will show less of a difference). Since the Mac pro is a rather expensive workstation suitable for users who need the performance, slowing its speed with inappropriate memory selection and/or installation is self-defeating.

The Mac Pro uses two memory “risers”. For quad-channel memory performance, the memory needs to be installed on both risers in pairs (eg 2 X 1GB on one riser and 2 X 1GB on the other riser, and in matching slots).

The following memory configurations (installed appropriately) allow peak speed:

• 4 X 512MB
• 8 X 512MB
  
  
• 4 X 1GB
• 4 X 1GB + 4 X 512MB
• 8 X 1GB
  
  
• 4 X 2GB
• 4 X 2GB + 4 X 512MB
• 4 X 2GB + 4 X 1GB
• 8 X 2GB

Note that the base memory configuration offered by Apple will not allow top performance:

• 1GB (2 X 512MB)

In other words, anything but the default 1GB setup is OK.

How much memory to buy?

Bottom line: don’t waste your money on more than 8GB unless you determine in advance that your particular application can actually benefit.

One (1) gigabyte isn’t enough for serious work with digital photography, so buy 2GB (4 X 512MB) as a minimum configuration with the Mac Pro (or PowerMac G5 Quad). Four (4) gigabytes should be enough for many users, though Photoshop CS2 will scarf up 3GB just for itself when hard-pressed. Add a raw converter, mail, web browser, etc, and heavy Photoshop users should plan on buying 6GB (as 4 X 1GB + 4 X 512MB) or 8GB (as 8 X 1GB). In practice, 8GB can rarely be fully utilized.

Although the new Mac Pro physically allows up to 16GB RAM, this is nothing new—so did the PowerMac G5 series. Mac OS 10.4 (Tiger) is still mostly 32-bit (32 bits = 2^32 = 4 gigabytes). So no conventional Mac OS X application can use more than 4GB of memory without being written as a 64-bit application using a restricted API set, thus gaining access to all the memory available. But this is not feasible for most applications, because they must use Mac OS X APIs that are still 32-bit.

In short , more than 8GB of memory is of little or no value for most users. While Mac OS X puts unused memory to use as a disk cache, there is marginal benefit from having multiple gigabytes of disk cache. While it is true that Mac OS X Activity Viewer might show more than 8GB of physical RAM in use, it is largely a “feel good” effect.

Adobe Photoshop CS2 could benefit from the ability to use more than 3GB. With large files (400MB film scans with multiple layers, Photoshop can spend a lot of time reading and writing to the disk on my PowerMac G5 Quad, even though I have 8GB of RAM, and have set Photoshop to use the maximum it allows—3 GB:

Adobe Photoshop CS2 memory usage setting

Mac OS X 10.5 (Leopard), allegedly moves down the 64-bit path. Assuming the 10.5 libraries have all been upgraded to be 64-bit, it will then become possible to write a 64-bit application which can use whatever memory is available. Of course, such applications won’t run on 10.4, so many vendors might well stick with 32-bit applications, rather than test and supply both 32 and 64-bit versions.

I regularly see 2-3GB of free memory while running Photoshop, even as it slows down reading and writing from the disk. To test this theory, I loaded Photoshop up with six huge files (200-400MB each). Photoshop “Scratch” was displayed as (lower left corner of the image window). I also started many other programs, and ran batches of images through Nikon Capture and Digital Photo Professional.

I then ran a long Photoshop action on two of the files. There were no Page ins/outs, and only a little disk activity. The memory display in Activity Monitor was as follows afterwards:

Interpreting the above, we can see that there is over 1GB free and 5GB of “Inactive” (aka largely wasted). What this really means is that there is up to 6GB of memory that could be used if needed. Click to see the whole Activity Monitor view—it’s rather outrageous—I was running Photoshop, Adobe Help Center, Dreamweaver 8, Safari, Mail, Adobe Reader, Nikon Capture, Digital Photo Professional, Terminal, Excel, Word, file sharing, SpamSieve, Activity Monitor.

Still, I wasn’t satisfied. So I started a batch via Digital Photo Professional and Nikon Capture at the same time. I cycled through the massive files in Photoshop and drag-scrolled at actual pixels. During that activity, the “Active” memory did go up, but the total of “Free” and “Inactive” was still about 3.5GB, with only about 4.5GB of memory actually being used (“Wired” + “Active”).

Bottom line: 6GB is plenty for nearly all users, even heavy Photoshop users—8GB is a “hedge your bet” amount.

No 85mm/f1.2L double image in out-of-focus background when shooting infrared

In my August 2 entry , I confirmed a double-image with the Canon EF 85mm/f1.2L II USM beginning at about 60% away from the optical center, worsening as the image becomes increasingly out of focus.

I have since verified that this problem does not occur when shooting in infrared, thus establishing that the problem is a form of chromatic aberration, and an excellent example of how real and theoretical depth of field are often at odds when chromatic aberrations are present. Though the color (visible light) image was shot with the 1Ds Mark II, and the infrared image was shot with the 5D, I think the results can be trusted (besides, I’m not going to convert a 1Ds Mark II to infrared!).

More than likely what’s going on is that by shooting in infrared, a narrower spectral band is used, and thus the color fringing seen in visible light (due to the broad spectral range) simply doesn’t occur. A common misconception is that infrared images are not sharp as sharp as visible light images; this is simply untrue and will be explored in a future article. Both images below are well out of focus at f1.2.

EF 85mm/1.2L II USM on EOS 1Ds Mark II f1.2 @ 1/160 sec 65mm behind plane of focus 80% of the way from center to right edge	EF 85mm/1.2L II USM on Canon 5D f1.2@ 1/100 sec 80mm behind plane of focus 85% of the way from center to right edge

Mac Pro memory, at a very appealing price

Satech has revised their pricing. Please see the 20 August entry instead. The original blog entry as written is below.

The 3GHz MacPro

The testing website barefeats.com kindly ran a hash benchmark for me (see below), the same algorithm used in IntegrityChecker. The results (which I consider quite indicative of the “true” performance of the machine for tasks involving lots of CPU and memory) show about a 44% advantage of the MacPro 3.0GHz quad-core vs the PowerMac G5 Quad. See also diglloyd PowerMac G5 Quad and MacBook Pro Experience Report.

llcG5:~ lloyd$ /ic test-speed
IntegrityChecker(tm) Version 1.0.2
(c)2003-2006 diglloyd, Inc. All Rights Reserved.

Test size = 32MB
Testing...

Chunk Size K  MB/sec
           1    97.9
           2   101.6
           4   103.5
           8   104.6
          16   105.1
          32   105.3
          64   105.4
         128   105.5
         256   105.4
Best chunk size: 256K

This computer has 4 processing units.
Testing using 4 threads simultaneously...
thread 0: 103.9...105.3
thread 1: 102.3...105.2
thread 2: 101.1...105.3
thread 3: 101.4...105.6
Aggregate rate: 421MB/sec

llcG5:~ lloyd$ /ic test-speed
IntegrityChecker(tm) Version 1.0.2
(c)2003-2006 diglloyd, Inc. All Rights Reserved.

Test size = 32MB
Testing...

Chunk Size K  MB/sec
           1   142.3
           2   148.0
           4   151.2
           8   152.8
          16   153.7
          32   154.0
          64   154.6
         128   154.8
         256   154.8
Best chunk size: 256K

This computer has 4 processing units.
Testing using 4 threads simultaneously...
thread 0: 153.6...154.1
thread 1: 153.6...154.2
thread 2: 153.5...154.1
thread 3: 153.7...154.2
Aggregate rate: 617MB/sec

These results indicate that for tasks that are CPU and memory intensive, the MacPro 3GHz has a 46% performance (throughput) advantage over the 2.5GHz PowerMac G5 Quad.

Scaling the 3GHz down to 2.5GHz would yield roughly 514 MB/sec, or 22% performance gain at the same clock speed as the PowerMac G5 Quad. But high-end users will opt for the 3GHz (or faster future models), so the 46% number is of more relevance.

The cost of MacPro memory is still a major burden in owning a MacPro, at least if you need 4GB or more. My advice remains unchanged—wait for 3-6 months for the price of memory to come down, the CPU speed to go up, and any “version 1.0” bugs in the MacPro to be worked out.

The 3GHz MacPro

My committment to my assertions in my August 8 entry is wavering slightly in that the 3GHz model of the Apple MacPro has proved to be very fast indeed at some Adobe Photoshop CS2 tasks, and is faster on Adobe After Effects, even though both are emulated.

However, I stick to my original advice that buying a MacPro now is too early if you also need lots of memory—it’s just too expensive as of yet.

I paid about $800 for 8GB of ECC RAM for my PowerMac G5 Quad (non-Apple, 3rd-party). Compare that to about $2500 (from Apple) or perhaps $2000 (3rd-party). If price is not a significant consideration, the MacPro looks like an awesome machine, even for Photoshop users. But verify performance with the specific features you use before committing.

Of course, if you are buying new, my advice remains to go with the MacPro—any extra money you spend on MacPro ECC memory will be no worse than the depreciation on a PowerMac G5 Quad.

Please see barefeats.com for SHOOTOUT: Quad Xeon 3.0GHz Mac Pro versus Quad G5 2.5GHz Power Mac on the MacPro 3GHz and Photoshop benchmarks.

More on the Fuji S3 UVIR

As mentioned previously in this blog, I’ll be presenting a two-hour class on digital infrared photography at Keeble & Shucat in Palo Alto, CA. The date now appears semi-firm as Wednesday October 25.

It is very likely that I’ll be purchasing the new Fuji S3 UVIR camera and if so, I’ll be comparing it with several other infrared cameras (converted from visible light), such as the Nikon D70, Nikon D200 and Canon EOS 5D.

ArsTechnica article on Apple MacPro

Article: “The End of one era, the beginning of another” is worth a read. See also my comments on the new MacPro in the August 10 entry.

My new Recommended Products and Vendors page

I have added a Recommended Products and Vendors page which lists select products and vendors that I have personally used and meet my highest expectations.

My Introduction to Digital Infrared shooting class

If you’re in the San Francisco Bay Area and have an interest in infrared photography, you might want to attend my class which I’ll be offering at Keeble & Shucat in Palo Alto. The final date is being worked out, but it will likely be a Wednesday evening the last week of October or the first week of November. By then I hope to have a good feel for the new Fuji S3 Pro UVIR, in addition to my Nikon D70-IR and Canon EOS 5D-IR.

Caution traveling to the UK

If you’re planning on traveling to or through the UK, be sure to read about the new “no carry-on” rules brought brought about by a plot to detonate explosive devices. If you’re rich enough to afford a private airplane (like the google boys), get your own private plane now.

Potential theft and breakage of expensive camera equipment and computers is now a very serious issue. Got a $30K digital back?—big problem. Knowing the unlimited capacity for government incompetence, we’ll see if the rules actually get lifted anytime soon. I’d advise anyone with really expensive stuff to ship it insured instead via a reliable carrier.

More on the new Apple MacPro

[See also my MacBook Pro Experience Report]

I’m sticking with my assessments in my August 8 entry, but barefeats.com tests show that the Apple MacPro 2.66 GHz is so fast that it is in spitting range of the PowerMac G5 Quad with Photoshop (see also the barefeats.com Armchair Analysis). This is true even though the MacPro has to run Photoshop in emulated mode. A 3 GHz model will close that gap by up to 13% (3.0 vs 2.66 GHz)—but the real-world improvement remains to be seen, since it depends on more than just clock speed. But $800 is better invested in memory than a piddly 13% clock-speed increase.

So for users who are casual or moderate users of Photoshop, the MacPro might be a reasonable investment now, provided that the extremely expensive memory is palatable. If you need 8GB or 16GB, it’s going to hurt. Were I starting with an all-new machine, I’d go with the MacPro and live with the emulated performance for a time.

On the other hand, fully loaded PowerMac Quad G5 machines might quickly become bargains. If someone offered me $3500 for my Quad with its 8GB ECC memory (6 months old), I’d quickly accept the offer. But selling it for less than that to get a performance decrease with Photoshop (which I personally could live with) makes no sense at all, so I’m just going to keep the G5 Quad for another 6-9 months—still the best Mac I’ve ever used.

It also appears that the MacPro 2.66 GHz offers up to 50% performance increases when using Universal Binary applications that make use of all the cores. Whether this is also true of single-threaded applications will become clear within the next week or so. I will also be posting test results for my Integrity Checker program, courtesy of barefeats.com. Integrity Checker makes the most efficient use of a quad-core (or any number of cores) of any program I’ve yet come across. Well, I’m an expert software developer, so what do you expect?

See also the macintouch.com preliminary report and, for geeks and developers, the Apple hardware note.

The New Nikon D80—little brother to the Nikon D200

The Nikon D80 is now officially announced. DPReview has a short preview and robgalbraith.com has one as well . With the same resolution as the Nikon D200, it might make an excellent backup body. Too bad I just purchased a second D200 a few weeks ago at $1699, since the D80 is priced at $999 (US). Still, the D200 is presumably a better-built camera.

Nikon also announced a 70-300/f4.5-5.6G ED-IF AFS VR zoom. Note that is is not a DX lens (it covers the full 36 X 24mm frame), which suggests that Nikon has something up their sleeve in the full-frame category.

An 18-135/f3.5-5.6G ED-IF AF-S DX zoom was also announced. It remains to be seen if either of these lenses are better than the 18-200/f3.5-5.6G ED-IF AF-S VR in terms of optical performance.

A rant on web design

Let’s hope Nikon doesn’t switch over from its old lens page to this new-style, slow-to-load, and really annoying Nikon lens page, which doesn’t make it possible to direct-link to a specific lens.

Instead you are greeted with useless animations and tiny type scrollable in a tiny area. Information of practical use is not provided, such as flare, distortion or MTF charts. Have fun reading the specifications in 8 point type confined to a 260 X 130 pixel area of the page. Idiotic best describes this type of web design. Shown below is my desktop (just the main screen—I have a 2nd monitor also). The area reserved for lens specifications is circled in red:

Does Nikon really think photographers all use 800 X 600 screens?

The Fujifilm S3 Pro UVIR camera

Fuji is about to offer a variant of its Finepix S3 Pro digital SLR, the Finepix S3 PRO UVIR, designed to enhance image capture of infrared and ultraviolet images. This represents the first step in the right direction—expanding the coverage of digital photography from color alone to other speciality areas, such as infrared and ultraviolet and monochrome (no monochrome digital SLR yet exists).

Fuji is targeting this camera at the forensic and crime markets, but it will likely make a terrific camera for those of us who enjoy shooting in ultraviolet and infrared, at least for landscape and similar shooting. It won’t be very interesting for any shooting involving moving subjects.

It’s not clear whether simply modifying a regular S3 Pro and installing clear glass over the sensor would achieve essentially the same results, albeit without the Live CCD Preview Mode. (Such conversions are performed by MaxMax and LifePixel). Fuji doesn’t mention anything about different autofocus, sensor design, etc. If this is the case, then the *only* advantage to the UVIR model would be the live preview. However, the S3 Pro UVIR is still a good deal, since converting a conventional camera costs somewhere in the $350-$500 range, plus the loss of warranty coverage.

Here’s my take on the good and bad points of the Fuji Finepix S3 PRO UVIR—

• Good—Live CCD preview mode. This means locking the mirror up, and perhaps a tripod also.
  
  
• Good—use of clear glass over the sensor so that the full spectrum is transmitted to the spectrum. Bad—special filters will be needed to block visible light and/or other undesired areas of the spectrum.
  
  
• Good—low noise. Or so Fuji claims, but I’m inclined to accept their claim.
  
  
• Bad—no autofocus support for ultraviolet or infrared; manual focus is required. This is probably why Fuji chose to offer the “live CCD preview mode”, though it had better allow a high zoom level, or getting sharp images is going to be difficult. It is possible (possibly through internal filtration over the focus sensor), to autofocus accurately in UV or IR. Perhaps we’ll see this someday.
  
  
• Bad—manual exposure. Fuji made no attempt to offer a UV or IR mode, which would properly meter for that spectral band. This is not a big deal, but it does slow working speed down.
  
  
• Bad—the 25MB image size for a “full range” image is simply too large, squandering storage precious space on the in-camera card as well as on a computer hard disk once downloaded, not to mention the increased demands on card speed and disk speed. Even the 16.7 megapixel Canon EOS 1Ds Mark II manages to keep its files around 16MB most of the time—and the Fuji is only a 6 + 6 megapixel camera.
  
  
• Bad—apparently the sensor is still a color sensor. Shooting in infrared or ultraviolet is inherently monochromatic. Though a color sensor has some advantages for “false color” shooting (eg mixing visible light with infrared or manipulating red/green/blue color channels, which vary in their response to UV and IR), the dyes used over the photosites not only require interpolation to produce the “color” but degrade the potential ISO sensitivity by as much as 2 stops.
  
  
• Bad—I don’t want to fiddle around with 4 AA batteries. I want a dedicated lithium ion battery pack such as used in the Nikon D200 or Canon EOS 5D.

Do all the “bad” points mean that the S3 Pro UVIR camera isn’t a contender? Of course not! I’m delighted to see steps in the direction of specialized digital photography, and Fuji deserves praise for heading that way. Incidentally, Canon offered a specialized 20D for astrophotography about 18 months ago, but it was never available for purchase in the USA (to my knowledge). Nor was it necessarily superior for infrared or ultraviolet shooting.

I might buy an S3 Pro if the image quality is really as superior as Fuji claims, though I’m already getting outstanding results with my Canon EOS 5D, which I converted for pure infrared use. I will be testing (at least informally) the S3 Pro UVIR as soon as I can lay my hands on one.

So what would be the ideal infrared and/or UV camera? Here’s what I hope Nikon or Canon will offer sometime soon (since those are the lens lines I own):

• a spectrum-specific body: pure infrared or pure ultraviolet, with autofocus and exposure systems adjusted for a defined spectral band (eg infrared from 750-900 nanometers). If this means buying one body for infrared, and one for ultraviolet, it’s worth it if exposure and autofocus work well. An swappable internal filter would also do the trick nicely, with appropriate settings to switch “modes”.
  
  
• a high-resolution monochrome sensor sensitive from 300-1200 nanometers so that monochromatic images could be taken with any section of the spectral band. A monochrome sensor would likely allow a base ISO of 400 or 800, and probably allow ISO 12,800. File sizes would be much smaller, true 16-bit data would be possible, interpolation would not be needed, aliasing would be reduced, etc. Image quality could be phenomenal.

Let’s hope Canon and/or Nikon also have something up their sleeves.

Why I won’t be buying a MacPro for at least 9 months

Apple’s new Intel Xeon “Woodcrest” dual-CPU (quad core) MacPro looks to be a formidable workstation, well beyond the needs of anyone not doing heavy-duty Photoshop, video editing, sound production, etc. In other words, a complete waste of money for most users, who would do just as well with an iMac Intel Duo. Still, the MacPro looks to be the most elegantly designed workstation I’ve yet seen, well worth lusting over...until you consider the price. But about 18 months from now, a $1499 iMac will probably offer similar performance!

Though you can buy an el-cheapo stripped down (and not very useful) MacPro 2GHz model for $2499, by the time you go to 3 GHz, configure it with 8GB RAM, a better video card, a 30" display, and a terabyte of storage, you’re looking at $8748 (plus tax). But Apple is feeling generous—free shipping is included.

You get the following improvements, which indeed are useful to high-definition video, photography and sound production houses with bottomless budgets:

• Faster performance, assuming your programs are Universal Binaries to make use of them. And assuming they’re not clueless about making use of all 4 cores, which most programs are. I still don’t have *any* programs that can make use of all 4 cores on my PowerMac Quad (except my own self-written Integrity Checker). Even Photoshop CS2 is lame in this regard, barely using 1.5 cores on most tasks.
  
  
• Space for 4 internal hard drives, though I already have stuffed 5 drives into my PowerMac G5 Quad. So it’s nice to have more space internally, but whoop-de-doo. And if Apple's SATA implementation is as inferior and incompatible with some drives as on the current PowerMac G5 Quad, you'll be better off buying your own card and external enclosure.
  
  
• Faster memory throughput. I’ve heard this one before. It’s a scam to eke out another 3%-10% performance at 3X the memory price. I keep having to buy a new set if 8GB of memory each time I buy a new desktop Mac!@@#! At least it’s ECC, but you could get ECC with the PowerMac G5 Quad, and at reasonable prices too.
  
  
• Three (3) 16X PCI slots. Any card needing such a slot is likely to cost well over $1000. However, unlike the previous G5 Quad, a high-end dual-width video card does not occupy one of the other 3 slots. Note the on-board fans on these cards—they'll make noise—no thanks.
  
  
• Two firewire 800 ports (oh my gawd, twice as many!). Firewire 800 is dog-slow compared to SATA, and also incompetently implemented on prior G5 PowerMacs (slow write performance). Maybe Apple has fixed this on the MacPro, but who cares? External SATA is far faster.
  
  
• Optional 2nd Superdrive. I'm sure someone has a use for this, but it’s not so super to me. They should have added two more drive bays instead, or a tray-loading popcorn popper or toaster oven.

What do you get the raw end of the deal on? A lot:

• First and foremost: you’ll get slower performance than with the PowerMac G5 Quad with certain critical applications. Photoshop CS2 will likely run as emulated PowerPC code until sometime next spring. Nikon Capture, Digital Photo Professional, Dreamweaver, Microsoft Office, etc will all run slower than the G5 Quad. By the time all these applications are updated, the price for your no-longer-“hot” new system will have dropped by $2000. Don’t be a sucker—wait till next March to buy when speeds will beyond 3 GHz and memory will be cheaper and Photoshop will be a Universal Binary.
  
  
• “Expansive memory” eg the same amount as in the prior PowerMac G5 Quad—16 GB of outrageously expensive ECC RAM. How does an extra $5700 sound for 16GB of RAM? That’s $356/MB. Ouch. Ouch. Ouch.
  
  
• A basic, slow video card that can only support one dual-link screen. You have to pay more for two video cards and/or a higher end card.
  
  
• $800 more to get 13% more megahertz (2.66 to 3 GHz). Or choose the “I’m feeling wimpy and have no good reason for this machine, so I’ll be a cheapskate” 2GHz model. Then tack on the cost of more memory, a screen and more storage, and kick yourself for buying slow processors.
  
  
• Overpriced hard drives: $400 each for additional 500GB SATA drives that were in today's Fry's Electronics advertisement for $189.
  
  
• Sluggish “Airport Extreme” when other vendors are supporting much faster protocols (eg 300 megabits/sec ala Netgear RangeMax). C'mon, I'm going to use a wireless 54 megabit/sec connection when I can use two 1000 Mbps (gigabit) ethernet ports? Remember Star Trek’s Scotty trying to get the warp drive on the Enterprise to go faster...c'mon, just plug in the ethernet cable, pulleeease. Or add some dilithium crystals. And knowing Apple’s marketing-hype-over-performance approach, the two gigabit ports still probably suck up a full core each if you actually try to use them at maximum speed, jumbo frame support notwithstanding.
  
  
• You have to pay for iWork at $79, Aperture at $299, etc. A $9000 machine and they want more money so you can actually run applications without Rosetta (PowerMac emulation)? These programs should be included, at least once you hit the $5000 mark.
  
  
• Apple wants $29 for bluetooth when I configure a $9000 machine?
  
  
• Why isn’t Mac OS X Server included once I pass the $5K threshold? It’s priced like a server, so include the server version of the OS.

In short, even my usual lust for the latest and greatest finds the new MacPro outrageously expensive for promised performance that might not materialize (depending on your applications) for 6-12 months from now—when prices will have dropped considerably.

Double image in out-of-focus background with 85mm/f1.2L II USM confirmed

In my 31 July entry, I reported a double-image with the Canon EF 85mm/f1.2L II USM about 60% away from the optical center. I have confirmed that this phenomenon exists with another brand-new sample of the lens, so presumably it is due to the lens design, and not peculiar to any particular sample of the lens. Focusing accuracy (using a custom target with “depth”) on both lenses shows an error of about 5 millimeters (front/back).

Canon EF 85mm/1.2L II USM on Canon EOS 1Ds Mark II f1.2, 1/160 sec, mirror lockup, ISO 50 65mm behind plane of focus 80% of the way from center to right edge
Lens Sample 1	Lens Sample 2

Double image in out-of-focus background (Canon EF 85mm/f1.2L II USM)

In researching a new article, I came across an interesting form of chromatic aberration with my new Canon EF 85mm/f1.2L II USM. By the way, I consider the upgrade well worth it. Though the focusing speed remains relatively sluggish and difficult to use for action, it’s a welcome speedup over the older model.

At f1.2, a magenta/green double image begins to appear about 60% of the way from the optical center to the edge of the frame. The effect is symmetric to both the left and right sides (my test was also about 10% below the vertical center of the frame).

The target patches shown below were behind the plane of focus as indicated, and also away from the optical center as indicated.

Canon EF 85mm/1.2L II USM on Canon EOS 1Ds Mark II f1.2, 1/125 sec, mirror lockup
45mm behind plane of focus 65% from center to right edge	65mm behind plane of focus 80% from center to right edge

Whether this is normal, or whether it indicates a problem with the lens is unclear. I suspect it is normal behavior related to the type of optical focusing the lens employs. Until I can test another sample, I won’t know for sure.

The double image seems to be a form of chromatic aberration, probably induced by the optical formula outside the plane of focus, since I haven’t observed any color fringing on in-focus subject matter. The color fringing is plain to see in other out-of-focus areas of the image as well.

It is disappointing that Canon didn’t completely eliminate chromatic aberration with this new version of the venerable 85mm/f1.2L. Canon’s terse press release notes only faster focusing, though sites like robgalbraith.com and dpreview.com state that the new version is “the same optically as its predecessor”. This claim appears to be contradicted by Canon’s own lens specifications. However, the apparent contradiction might simply be incomplete or poorly-explained specifications, rather than an actual difference.

Canon’s web site symbols indicates that the new version contains 3 aspherical elements and 1 “Super UD” element, whereas its predecessor indicates just one aspherical element. [See the online lens brochure, page 38, for an explanation of the symbols that Canon displays for its lenses]. Additional data was gleaned from Canon’s EF Lens Work III (well worth having in your photographic library, even if you’re not a Canon shooter).

Canon’s book “EF Lens Work III—The Eyes of EOS”, page 59 indicates “one large diameter ground and polished aspherical element” and “two high-refraction glass elements” for the original 85mm/f1.2L. Canon’s current (July 2006) online lens brochure says nothing about UD or Super UD lens elements in the original 85mm/f1.2L. There is also an Eyes of EOS CD-ROM, which I haven’t used.

I suppose it’s hard to complain much, since it’s already a one-of-a-kind lens with overall outstanding performance!

Infrared backfocus

I’m still sorting out which lenses work well and which ones don’t on my converted Canon EOS 5D. As noted in an earlier entry, I had it converted for pure infrared use by maxmax.com.

Although MaxMax adjusts their converted cameras for infrared focusing, there is enough lens-to-lens variation (between different lens models eg 50mm vs 85mm) that this adjustment might not be very helpful in many cases. I will be compiling a table of lenses and how they perform with infrared on both the 5D and D200 (no converted D200 just as yet however).

Below is an example of the focusing error with the Canon EF 50mm/f1.4 at f1.4, taken using a custom target I built this evening with stepped bars 5mm apart (front to back):

Target*	Four (4) centimeters behind target

There is about a 40mm focusing error at a distance of around 3-4 feet. Stopping down at least 3 stops would be needed to render crisp results at the chosen focus area. This is why I shoot much of my infrared work at f11—to minimize the number of ruined shots.

I will be exploring this issue in great detail going forward, as well as double-checking for accurate visible light focus with non-modified camera bodies.

MacBook Pro and Serial ATA (SATA)—SeriTek

I’ve updated my MacBook Pro Experience Report with tests using the SeriTek/2ENSM2-E External Bundle with a 2-drive striped RAID 0 setup using 2 Maxtor 7V300F0 drives, previously discussed in this blog.

The speed is dramatically faster than anything the internal hard drive or Firewire has to offer, and will go a long way towards eliminating the bottleneck of disk speed on the MacBook Pro. The sole remaining serious limitation of the MacBook Pro for serious photography work is the limitation to only 2GB of system memory.

Infrared flare with the 85/f1.2L II USM

The Canon EF 85mm/f1.2L II USM shows pronounced flare in infrared when there is a very bright light source just outside the frame (in this case a 50 watt MR-16 Solux bulb). I haven’t yet tested for this issue in visible light shooting. The full-frame example below used no filter, and the Canon-supplied lens hood was attached:

More on infrared hot spots

Yesterday’s discussion of infrared hot spots did not demonstrate the unpredictible nature of infrared flare, which has little regard for the prime versus zoom debate. Consider the two examples below. Contrast is better with a prime lens, right?

Wrong! Both examples were taken at f11. The top one was taken with the Canon EF 100-400/f4.5-5.6L IS USM and the bottom one with the Canon EF 300/f4L IS USM. The MTF charts suggest that the prime lens should offer notably better contrast than the zoom overall. However, MTF at the center of the image is better with the zoom (at least at 400mm), and in infrared the prime clearly shows the infrared flare problem seen above. It’s no “small” problem—the letter “A” above is about 1300 pixels wide in the original frame! [Thanks to Keeble & Shucat of Palo Alto, CA for loaning me the lenses for this test—a terrific store if you live in the area].

Depth of field and megapixels

Sometimes it’s worth reminding ourselves how little megapixels matter with some subjects—limited depth of field can eliminate nearly all of the potential resolution of an image. This example was shot with the Canon EOS 5D-IR (infrared conversion) using the EF 100-400/f4.5-5.6L IS USM at f11. Color as opposed to infrared would make little difference. Click to see the whole frame from which the crops below were taken.

(To conserve viewing space, the railroad tracks were rotated to a horizontal position). Even at a greatly-reduced size, the foreground and background blur are evident—and this is at f11! Results are worse at f16 due to diffraction (I tried both apertures). Click for actual-pixels, or see below.

Below is an actual-pixels crop. Only a very narrow band of the entire frame is actually in acceptably sharp focus, no more than 300 pixels (out of 2912), probably no more than a 100 pixel band is at peak sharpness. A 6-megapixel camera (instead of 12) wouldn’t produce an appreciably different result.

Photos like the example make a good case for a reduced-frame sensor, such as found on the Nikon D200, where the smaller sensor allows use of a shorter focal length lens at the same subject-to-camera distance . See Depth of Field—The Shorter Focal-Length Advantage.

Canon EF 85mm/f1.2L II initial impressions

I sold my EF 85mm/f1.2L prior to buying the improved 85mm/f1.2L II. Therefore, I won’t be doing a side-by-side comparison, though I might offer more comments from time to time. Please see the thorough review at www.wlcastleman.com.

I like to shoot the 85mm/f1.2L wide-open to f2.8. The new version shows pleasing contrast wide open, with an impression of great sharpness. The only problem is that pinpoint focus is even more important, since a “hit” looks terrific, and a “miss” is a throwaway. At 85mm and f1.2, depth of field is essentially non-existent on a Canon EOS 1Ds Mark II, and so you had better get it right. I can’t yet say whether the autofocus accuracy is always up to the task.

Below is a real-world example which used autofocus, ISO 1600, f2 @ 1/250 sec and “a monkey on speed” (4 year old child). I “cheated” and used f2; whether focus at f1.2 would have been adequate can’t be said with certainty.

85mm/f1.2L II, handheld f2 @ 1/250 sec, ISO 1600 processed using Digital Photo Professional
Unsharpened	Sharpness = 2

The 85mm/f1.2L II can’t be characterized as a high contrast lens, but it offers high resolving power with a desirable contrast for a beautiful effect at wider apertures. The contrast in the above example is awfully good for such a fast and unique lens. Or at least it satisfies me, and I’m one picky customer. If you shoot portraits using Canon EOS, you simply must own this lens (sorry).

Infrared “hot spots”

Sooner or later when processing infrared images (probably sooner), you’ll discover that some lenses are poorly suited to the task because their coatings produce an objectionable flare or “hot spot” in the center of the image. This flare isn’t present in visible-light photographs, but it can be present with some lenses, even some of the very expensive ones, such as the otherwise outstanding Canon EF 70-200/f2.8L IS. Below is an example, sized down (but representing about 80% of the width of the original frame). Note how the grayish letter “A” is significantly lighter than the other letters.

The contrast loss is a serious problem for some images, though in some it can be harder to detect. It is also a problem in that it can “blow out” or “clip” the highlights in a key part of the image—the center. Of course, any uniform image area such as sky or water will show such hotspots readily. I have also observed hot spots in areas other than the center with some lenses.

Post-processing can greatly reduce the visibility of such hot spots, but any clipping of detail can of course not be recovered.

My search is on the for the most suitable lenses for infrared work on the Canon EOS 5D, and I already have a few favorites. In general, primes are better suited to the task, though some zooms are quite fine performers too.

Look for a diglloyd.com article on lens selection for Canon and Nikon in the late August/early September time frame, as I accumulate more experience with more lenses.

Faster diglloyd.com web site

The diglloyd.com website has been switched over to a high speed internet connection. Minimum throughput should now be about 3 megabits per second, and as fast as 8 megabits. Of course, lots of intervening servers and networks might be between you and diglloyd.com, so you’re unlikely to see speeds that high. Please let me know if you find the response to be faster (or slower).

Canon 135mm/f2L in infrared

I received a Canon EF 135mm/f2L USM today for use on my EOS 5D-IR and can sum up its performance in infrared quite simply: WOW! It offers the best infrared performance I’ve seen on the 5D yet, right out to the corners, with no flare hot spot. I’ll post a sample sometime soon.

Canon 35mm/f1.4L in infrared

I just started trying out the Canon EF 35mm/f1.4L in infrared with the EOS 5D-IR and can report that it’s excellent, excepting some corner softness, which is also typical in color (the “L” zooms outperform it in the corners, and the corner softness is quite consistent with its MTF chart). Here’s a snapshot, and an actual-pixels crop:

Full Frame	Actual-pixels crop

Optical misalignment with the Canon EOS 24-70/f2.8L

I previously discussed alignment issues with my Nikon D2X, which ultimately required a $400 repair to replace the lens mount on the camera body itself (and taking an outrageous 3 months to perform—the reason my D200 vs D2X review was delayed) .

The Canon EF 24-70/f2.8L apparently is not in my good graces either. My first copy, about 18 months ago, required 2 trips to Canon service to resolve decentering and other optical issues. That copy ultimately proved to be a very fine performer, once the optics were aligned properly. I sold that lens about a year ago.

About a year later, I took up the Canon EOS 1Ds Mark II again, and purchased a new 24-70/f2.8L (about last December). I used it a bit, but hadn't used it for serious shooting to date. However, recently I began to employ it for infrared shooting on my Canon EOS 5D, which I had converted to pure infrared use by MaxMax.com (quick turnaround, clean results).

I immediately noticed softness in infrared with the 24-70/f2.8L, which quickly led me to shoot some tests shots on an unmodified Canon EOS 1Ds Mark II in color, confirming the misalignment with normal color photos (to rule out any infrared weirdness). Click to see the left/center/right color crops, at 50% actual pixels; The left side of the frame is quite blurry.

Even at f5.6 at a distance of about 100 feet there is fairly severe sharpness degradation, with a blurry center and sharper edges, the left edge being the sharpest. Shown below are actual-pixels crops taken from the full frame.

Optically-misaligned Canon EF 24-70/f2.8L USM, 1/250 sec, f5.6
Left	Center	Right

Given that the 24-70/f2.8L saw little use, and absolutely never had any rough handling, it’s yet another case of the poor quality control which I’ve discussed before in some of my reviews. The lens is now on its way to Canon service. Isn’t it odd that it’s happened with both copies of the 24-70 that I’ve owned? My 24-105/f4L has no such issues.

Is the Nikon 18-200/f3.5-5.6G ED-IF AF-S VR DX zoom good enough?

I really like this flower shot below (D200 with 18-200VR), both for its gorgeous color palette, simple color composition and smooth background (click for larger version). It was a snapshot I worked for a minute or two, taking 3 or 4 frames, and this one was my favorite. It was shot handheld at 1/100 second @ f10 with the 18-200VR.

It demonstrates what gorgeous color Nikon bodies can produce without the fiddling around often required with Canon EOS bodies and their low-contrast, less saturated look. Some work was done to bring out the dragonfly, and some contrast and saturation was added, but the image remains true to the original, and matches my “minds eye” of what I remember shooting. I’ve already had several requests for prints, and it’s for sale if you’d like to buy a print.

I don’t normally like variable-aperture “prosumer” zooms, but the long zoom range, Vibration Reduction (VR), high color saturation, contrast and sharpness make the 18-200VR a no-brainer for walk-around shooting under daytime conditions. The lens does need stopping down to f8 or f11 for best results, and the corners stay somewhat soft, but it’s hard to quarrel with the great combination of features. And as a bonus, it performs quite well in infrared.

Not convinced? And rightly so—many times, reduced-size images look good, but can’t hold up beyond an 18 X 12 or so. However, I printed a section equivalent to a 46" wide print, and while more detail would be helpful, I’m going ahead with a 46" print, perhaps larger, with no qualms as to image quality. The only problem is that this particular image is out of gamut for every paper I’ve tried, with Epson Premium Glossy offering the best match. That purple/blue color as well as the saturated yellows are a challenge for the Epson printers as well as the Lightjet.

Infrared article updated

Some more examples of infrared renderings are now in my Digital Infrared article.

Nikon D200 vs D2x comparative review

New front page photo

I really have dozens, if not hundreds of images that I find quite appealing sitting in my celluloid and electronic files, so every so often I dig around and find one that still appeals to me months or years later. One such image is the ferris wheel image I shot with the Canon EOS 1Ds Mark II last year, using the 24mm/f1.4L at dusk.

I’ve also posted a new infrared shot that I find quite appealing (Nikon D70IR). Both photos are on the photos page.

Interested in infrared photography?

I now have both a Nikon D70 IR and a Canon EOS 5D IR. There are some interesting differences...and I’m thinking of getting a 2nd Nikon D200 and converting it for infrared use. To fund all that buying (the conversion alone costs $450 per camera), I’m considering writing an in-depth paid article on infrared shooting, using those three cameras together, and assessing which one(s) produce the best IR imagery. Email me if you would find such an article interesting. Meanwhile, check out the free article Digital Infrared here at diglloyd.com

Nikon D200 vs D2x comparative review

The D200 vs D2x review will be available tomorrow. I haven’t yet posted the article overview but will do so by early evening.

Nikon D200 vs D2x comparative review

I’m putting the finishing touches on my D200 vs D2x comparative review (paid review). It will be available within a day or two. While not as extensive as the D2x vs EOS review, it still weighs in at a comparatively hefty 74MB, and anyone considering whether to buy a D2x or a D200 should find it quite useful indeed. A full description of the review contents will be posted soon.