My Micro Photography Rig

I'd like to get into photomicroscopy.  Specifically, 10X to 20X magnification (but up to 100X) of abstract subjects, printed large on fine art paper and canvas.
Contents
  1. Image Gallery
  1. My current system
  1. My target system
Goals
  1. I'd like to use this as a learning exercise.  Both in the fundamentals of photomicroscopy.  And in 3D printing to close tolerances and implementing fine control when necessary.
  2. I hope that using high quality components will let me cook up a system capable of high quality images.
  3. I want to be able to make large prints.  Thus, "stack & stitch" capability using the Stackshot rails for X, Y and Z.
  4. I hope to find and adapt older, slightly used, high quality components from eBay, etc.
  5. I plan to speed up my whole image creation process by employing faster hardware and pipelining the process.
 My current system (a work in progress)

Lake Placid, 7 Jun 2019
Here is my current system.  Left to right: a new PC and monitor (for post processing), the old PC and monitor (for capture), the camera.

Regarding the camera (left to right):
  • Sony a7R IV (w/ Pixel Shift). The a7R IV's USB-C port is connected to Sony Imaging Edge Remote running on a PC, allowing tethered remote control of the camera. The a7R IV's Multi-terminal port is connected to the Stackshot controller which actuates the shutter under the control of Zerene running on the PC. You can setup Zerene to move the Stackshot rail to a desired start position, take a certain number of shots, moving the
camera forward by a certain distance between each shot. The tethered connection (over the USB-C cable) streams each image from the camera to the PC (saving them on the NAS drive). With Pixel Shift set on, after the 4th image has been streamed to the PC, Sony Imaging View automatically combines the 4 images together to create the pixel-shifted result image.  I have mapped the NAS drive to Windows 10 (as "W" drive) and all the image files are saved there.  I can then read in the pixel-shifted (ARQ) file from the shared drive to my faster Windows 10 system (PC2) to convert it to a 16-bit TIFF file for stacking, again on that fast Windows 10 system.  I just need to write a program that automatically deletes the intermediary files and kicks off the stacker (Zerene) at the appropriate times.
  • Thorlabs SM2 tubes and Slip Rings attached with Arca-Swiss clamps to an Arca-Swiss rail.  Credit to Robert OToole's very helpful 52MM STUDIO SETUP.
  • Thorlabs ITL200 tube lens
  • Filters: all mounted in 3D-printed holders allowing rotation
  • Mitutoyo 10X & 20X Apo objectives (my other objectives and tube lenses are listed here)
  • A cheap but effective X-Y stage mounted on a 3D-printed rotating disc
  • A 3D-printed holder for the Polarizer and filters.  Again, all rotatable.
  • Allowances for a condenser and collimeter
  • A 3D-printed holder for an LED light source
  • The X-Y stage, filters and light source are mounted on an Arca-Swiss rail to a pair of Stackshot rails arranged to move in the X-Y axes (up-down and left-right to support stitching, ie. panoramas)
  • a Stackshot rail in the Z axis (forward and back for focus stacking)
  • A 12" x 36" x 2.4" Thorlabs optical breadboard, resting on four AV5 Sorbothane feet on a fairly solid IKEA dresser on thick carpet on a concrete slab.  I love the optical breadboard.
     
    Model Diameter Load Range
    AV4 27 mm 3 to 14 lbs
    AV5 38 mm 3 to 33 lbs
    AV6 45 mm 3 to 50 lbs
    The Sorbothane feet vary in Diameter, Load Range and Isolator Efficiency - summarized here.  To select the best size, the weight each foot supports should not exceed the Load Range, but be as close as possible to the upper bounds. 
    • Weight each foot needs to support:
      The optical breadboard weighs 28kg or 62 lbs.  Say, the rest of my rig weighs no more than 20 lbs.  Then, each of 4 feet under the breadboard need to support say, (62+20)/4 = 21 lbs. 
    • The AV5 feet have an upper load rating of 33 lbs.  So, they look to be sufficient.
    • AV5 vs AV6:
      The AV6 feet have an upper load rating of 50 lbs.  The Isolator Efficiency curves for the AV5 show better isolation (down and to the left is better) than the AV6.  Credit: I thought I saw this pointed out first on Robert OToole's website but I can't find the reference now.

 
 
 
The X-Y rotating stage makes it easy to move the sample up & down and left & right.  Then to rotate the subject to pose it.
 
The optical breadboard is isolated from my tabletop on the left on which my monitors and keyboards rest.  Still, I leave the room when I'm taking a series of photos (which can be a couple hundred) of say a butterfly wing for focus stacking.
 
Lake Placid, 14 Aug 2019
At the 42MP resolution of the a7R III (7952 x 5304px) and 61MP resolution of the a7R IV (9504 x 6336px), the images would print at resolutions of:
Print Size a7R III
42MP Image
Print dpi		 a7R IV
61MP Image
Print dpi		 Notes
13" x 20" 400 475 Canvas: utilizing the full width of my 24" canvas for the width, with a 2" border for staples to the frame
18" x 27" 300 352 The often quoted, 300dpi
20" x 30" 265 317 Canvas: utilizing the full width of my 24" canvas for the height, with a 2" border for staples to the frame
22" x 33" 241 288 Paper: utilizing the full width of my 24" paper for the height, with a 1" border for handling and framing

I will update this section periodically as I develop the system.
References:
   - 52MM STUDIO SETUP (after which I modelled my setup)
   - Matching Camera to Microscope Resolution
which I've plotted here ‐‐>
The dotted horizontal line (at Pixel Size = 3.76 µm) corresponds to my a7R IV (not counting the presumed benefits of Pixel Shift).  My Mitutoyo objectives are highlighted in the ovals.
 
I think this plot shows that my a7R IV is capable of out-resolving my Mitutoyo objectives (but not by much for the 10X.. again, not counting the benefit of Pixel Shift). 
 
Note, this plot is based on traditional microscope design (the text refers to aligning the condenser and matching the condenser's NA to the objective's NA), so I may be mixing apples and oranges.
Camera
Lake Placid, 15 Oct 2019
I now have an a7R IV.  Here is a look at the a7R IV image quality from Digital Camera Review's website compared with the a7R III, Panasonic S1R and Pentax K-1 Mark II.
 
Click on the image to enlarge it or on the link above to use the tool.
 
 
 
Lake Placid, 21 Jul 2019
I have an a7R IV ordered - scheduled to ship Sep 12...
  • Sensor resolution has been increased from 42MP to 61MP.
  • Pixel pitch has decreased from 4.56 µm to 3.76 µm.
  • Pixel Shift has increased from full pixel to half pixel - with resolution of combined image increasing from 42MP to 240MP.
I suspect this may outresolve my microscope objectives.  But, I hope to see benefits at macro scale - say, 1:1 of my Sony 90mm Macro.  We'll see.  Here is an early comparison from Tony Northrup, using an 85mm f/1.4 GM.
Boynton Beach, 8 Sep 2019
Here is an interesting take on resolution:  The TRUTH: High Megapixels + BIG Prints are a WASTE? iPhone vs Sony a7R IV vs Sony a7 III.  I have sumarized his results here: I added the 20x30 column, interpolating from his results.  20x30 is where I plan to print for a while.
 
Here is another of his videos on 240MP pixel shift:
240 MEGAPIXELS BLEW MY MIND (Sony a7R IV vs iPhone X, a7 III, a7R III). 
Lake Placid, 20 Aug 2019
"To correctly reconstruct an image, it must be sampled at at least twice the highest spatial frequency contained within the image. Sampling it at a lower spatial frequency will result in ambiguity which will cause artifacts in the reconstructed image. Making the Pitch finer raises the sampling frequency. Making the pitch coarser lowers the sampling frequency."
Jim Kasson
Lake Placid, 20 Aug 2019
"Lot of handwringing over the R4 "outresolving" unworthy lenses.... any examples of this on cameras like the R2-3 or the 5DSR? What does sensor outresolving look like on photos?"
A DPR poster posed an interesting question: “What does sensor outresolving look like in photos?” In order to deal with this, I needed to define what “sensor outresolving” means. I’m taking it to mean that the sensor samples the projected image from the lens not just at, but more finely than the Nyquist frequency.
Lens outresolving sensor Sensor outresolving lens
Settings: Lightroom with a few contrast, white balance and exposure tweaks, and sharpening to strength = 20, radius = 1 and detail = 0. --Sensors outresolving lenses
Jim Kasson
Lake Placid, 2 Aug 2019
"Just about any modern FF E-mount prime, on axis, can benefit from the transitions from 42 to 61 to 100 to 125 to 168 MP. 168 MP is only twice the resolution of 42 MP."
Jim Kasson
Lake Placid, 24 Jul 2019
To have a look at diffraction at these high resolutions for say, my Sony 90mm Macro lens, I ran several scenarios through the MACRO DIFFRACTION CALULATOR at photopills.com with the following results:
Camera MP Focal Length (mm) Aperture (f/) Focal Distance (cm) Magnification Effective Aperture (f/) Dia. Airy Disk (um) 100% Crop Print
Pixel Size (um) Diffraction Limited for f/2.8 Diffraction May Become Visible (f/) Circle of Confusion (CoC) (um) Diffraction Limited for f/2.8 Diffraction May Become Visible (f/)
a7R III 42 90 2.8 36 1X 5.7 7.53 4.54 No 4.5 30.00 No 11
a7R III-PS4 84 90 2.8 36 1X 5.7 7.53 3.21 No 3.2 30.00 No 11
a7R IV 61 90 2.8 36 1X 5.7 7.53 3.76 No 3.5 30.00 No 11
a7R IV-PS4 122 90 2.8 36 1X 5.7 7.53 2.66 Yes 2.5 30.00 No 11
a7R IV-PS16 240 90 2.8 36 1X 5.7 7.53 1.90 Yes 1.8 30.00 No 11
So, even my 90mm Macro lens may become diffraction limited with 240MP Pixel Shift.  Ie. the sensor outresolves the lens.  I wonder though if the other benefits of pixel shift - reduced moiré, reduced noise and improved color accuracy - will be evident in that combination.
Lake Placid, 1 Aug 2019
Recently, I have read some commentaries on whether 61MP is warranted.
I think
  1. I need to make the measurements (as best I can) and see for myself.
  2. This may make me want to delve into techniques and subjects that will push the boundaries in ways I don't appreciate now.  Nothing wrong with that.
Lake Placid, 22 Aug 2019
Some reasons why the a7R IV and Pixel Shift make sense for me (yes, this is my rationalization for buying it, although it has a number or other appealing upgrades):
  1. Shooting with microscope objectives limits cropping when composing the shot.  There is only one distance at which the subject is in focus - it's not like you can recompose the shot by taking a few steps forward or back.  It's like having a manual focus lens and not being able to rotate the focus ring (there is no focus ring).  So, I will want to do some cropping in post processing, maybe much of the time.  Having more megapixels helps.
  2. I want to print large so high resolution and image quality are desireable.  See The TRUTH: High Megapixels + BIG Prints are a WASTE? .
  3. My rig is very stable.  Thus, I should not have problems with shake-induced blur possible with pixel-shift.
  4. My subjects are not moving.  They are microscope slides or mounted specemins in a studio.  Movement (people walking, wind in the trees) messes up pixel-shift.
  5. People talk about image resolution being limited by lens resolution.  I have two cases:
    1. For microscope magnifications (greater than 1:1):
      I think my microscope lenses are pretty good.  My Mitutoyo microscope objectives are well regarded on the photomacrography.net forum that I follow.  The lens that I'll likely use as my "tube lens", the Thorlabs ITL200, is also well regarded in tests.  For example, in this test of 18 lenses, it was rated the best.  And again, in his followup.
    2. For macro 1:1 magnification:
      I think my macro lens is pretty good.  My macro lens (a Sony 90mm Macro) is the 2nd sharpest lens ranked on DXOMark (scroll down to the start of the list of lenses, then click on the top of the Sharpness column (on where it says, "Sharpness") to sort the sharpest to the top). 
      In More PhotoGeekery: Finite Conjugate MTF Bench Tests for Macro Lenses Roger Cicala compared MTF curves for the Sony 90mm Macro and Canon 100mm f/2.8L Macro.  I've sumarized his results here.  It shows the Sony besting the Canon when focussed at 1:2 magnification.  It shows the Canon besting the Sony when focussed at infinity (which applies if I were to use it as a "tube lens" with a microscope objective). But, from what I've seen so far, my Thorlabs ITL200 appears to perform better  as a tube lens.  It sort of makes sense that the macro lenses that I've tested do not perform well as a tube lens compared to the ITL200.  The ITL200 was designed to operate focussed at infinity.
  6. People point out that this high-MP and Pixel Shift are demanding of compute resources.  I have been gearing up for that - a new 5GHz 8-core PC, a future high capacity NAS, upgrading to Cat 8 (40 Gbps) ethernet, a future 48-node Multiprocessor, plans to automate the process between the point of composing the shot (which is done manually, of course) to post-processing (again, done manually).
  7. A number of reviews point out that Pixel Shift has benefits besides increasing resolution (reduced moiré, reduced noise and improved color accuracy).  For example, see 240 MEGAPIXELS BLEW MY MIND. 
  8. It's fun to push the limits.  I hope it will challenge me in any number of ways that I don't even know about now.
Here is a look at the a7R III dynamic range from the Photons to Photos website.  It is the black line.  The Phase One IQ4 (blue line) bests it but that is a medium format $52,000 camera.  The Pentax K-1 Mark II beats it but at higher ISOs.  I am able to control the lighting and exposure and am shooting at the lowest ISO (where many of the cameras converge in any case - about 11½ stops).
 
Lake Placid, 9 Sep 2019
And here is a look at the recently added a7R IV dynamic range from the Photons to Photos website.  I am just comparing it with the a7R III to make the plot simpler.
 
Again, the a7R III is the black line.  The a7RIV is the blue line.   They're pretty close which is interesting given the a7RIV's photosites are smaller - area-wise to the lens.  I've seen comments saying the a7RIV's photosites are deeper
so there is more capacity to collect photons (converted to electrons).
 
DXOMARK has ranked the a7R III favorably since the camera first came out - now virtually tied in the top three full frame cameras.
 
It is reported that Sony's post processing tool, Imaging Edge, when used on the a7R III sharpens the result when it creates the combined image.  But when I've just eyeballed it at 100%, I think I've seen detail in the combined image that is not simply the result of sharpening (it could be I'm looking at it wrong).  In any case, I will look at using other tools (e.g. SonyPixelShift2DNG, Raw Therapee) as noted in that article.
 
There do appear to be benefits using pixel shift of reduced moiré, reduced noise and improved color accuracy.
 
I realize this all has very little, if any, effect at "web page image sizes".  My hope is that its benefits will come into play at large format print sizes.
Lake Placid, 22 Aug 2019
That does appear to be the case.
References:
   - LENS DIFFRACTION & PHOTOGRAPHY
Lighting
Boynton Beach, 27 Nov 2019
Positioning
Boynton Beach, 28 Nov 2019
Polarizer
Armenia, 7 Apr 2019                 
I want to make polarized photos like the crystals on my gallery page.  It looks like that calls for 2 polarizing filters, a "retardation or wave plate" (mica, scotch tape or I'm going to try this retarder film) and a holder for the microscope slide. 
 
I'm planning to 3D-print a holder for them.  Click on the animation ->
Glass polarizers
Retarder film
with condenser
There are numerous descriptions of how crossed polarizers work online. This one by Olympus is particularly good. 
The graphic on the left (borrowed from this page at Georgia State University) seems to me a good illustration of 2 polarizers crossed at
90%.  The graphic on the right shows the effect of a birefringent sample.  It splits the polarized lightwaves that enter the sample into two new lightwaves which are able to pass through the second polarizer (called the Analyzer).  Retardation/wave plates enhance the color.

Here is a description of retardation or wave plates from this Facebook post:
Optical waveplates (also called wave plates or retarder plates) are transparent plates with a carefully chosen amount of birefringence. They are mostly used for manipulating the polarization state of light beams.
 
Two common types of waveplates are the half-wave plate, which shifts the polarization direction of linearly polarized light, and the quarter-wave plate, which converts linearly polarized light into circularly polarized light and vice versa. Addition of plates between the polarizers of a petrographic microscope makes it easier to identify optically minerals in thin sections of rocks. Or people use them to get pretty images.

One polarizer (called the Analyzer) and the Retardation Plate would be located in the tube between the tube lens and the objective.  The Slide and other Polarizer would be held in the 3D-printed holder.  I am able to insert multiple retardation plates in the holders, resulting in more varied colors reaching the camera.  Each of the retardation plates can be rotated to vary the colors.
 
Printer
Lake Placid, 13 Jul 2019
Here is the printer I'm using - a Canon Pro-4000.  It is capable of printing on 44" wide media.  I expect to mainly make 20" x 30" prints on canvas and fine art paper.  But have the ability to print much larger.
Paper and Canvas
Lake Placid, 21 Jul 2019
Here's what I have:
Type Brand Name Wt.
(g)		 Size Cost
$/sq.ft.		 Quote Roll / ICC / Profile
Canvas Breathing Color 17M Satin Canvas 350 24" x 75' $1.88 "industry-leading Dmax and color gamut. No Varnish Required. Use Photo Black." -Src
Lyve Matte Canvas 450 24" x 40' $1.25
Epson Glossy Exhibition Canvas Archival 420 24" x 40' $1.90
PremierArt Museum Bright Satin Canvas 350 24" x 40' $1.74
P&L Art A cheap polyester canvas 290 24" x 100' $0.50
Glossy Canon Glossy Photo 170 24" x 100' $0.42 2" / / Canon's builtin profile
Semi/ Smooth Glossy Canon Premium Semi-Glossy Paper 2 280 24" x 100' $0.75 3" / / Canon's builtin profile
Canson Infinity Baryta Prestige 340 24" x 50' $1.94 "richness and depth in tonal values and hues. Inks seem to blend in with the surface rather than sit on top, a very pleasing look. It is a “smooth gloss” paper, which means that it has the look and sharpness of traditional glossy paper while lacking the distracting sheen of a “hard” surface gloss." -Src 3" / / Canson's Infinity Baryta Prestige profile
Epson Legacy Platine 314 24" x 50' $1.90 "On the luster side, the Legacy Platine has a particularly pleasing feel and it gives the photos a bit of a distinctive richness that I believe will make this paper a favorite in the fine art photographic community." -Src 3" / / Canon's Pro Luster profile
Matte Epson Singleweight Matte Paper 120 17" x 131' $0.17 I bought this mainly to make an occassional test print to try to keep the printhead from clogging. 2" / / Canon's Pro Premium Matte profile
Hahnemuhle William Turner 310 24" x 39' $2.42 "Not only are the colors and tones excellent, but detail is great too. In addition to printing fine details well, the William Turner paper's texture also adds an additional appearance of detail and depth. For a finely-detailed image that is rich in texture, I'd opt for William Turner." -Src 3" / / Hahnemühle' Wm Turner profile
Metalic Red River #1771 Polar Gloss Metallic 255 (formerly 66lb. Polar Pearl Metallic) 255 24" x 100' $1.00 "closely matches the look of photo lab metallic prints" -Src
#1848 Polar Luster Metallic 255 255 24" x 50' $1.11 "related to Polar Gloss Metallic 255. The difference is that this paper has a luster texture similar to Arctic Polar Luster. Your images will take on a brilliant, almost luminescent quality. Everything from airplanes to flowers will look bold, more saturated, more intense." -Src
Satin Hahnemuhle Fine Art Baryta Satin 300 24" x 39' $1.91 "a satin-gloss surface on a bright white base. The barium sulfate in the coating works in conjunction with a microporous ink receiving layer to deliver vivid prints from subtle to intense color images: monochrome prints express deep, rich blacks as well as a wide gamut of grayscale values." -Src 3" / / HFA Baryta Satin
Red River #5129 68lb. UltraPro Satin 4.0 270 24" x 100' $0.47 "looks and feels just like traditional photo lab paper. The surface is textured to a medium depth. Bright white papers are particularly good at reproducing deep blacks and higher contrast." -Src 3" / / Canon's Photo Paper Plus Semi-gloss II
#1278 San Gabriel Baryta SG 2.0 315 24" x 50' $1.30 "tonality, depth, and feel of a traditional darkroom print. The paper has a lightly textured surface, similar to a fine satin." -Src
 Recommendations ...
Choosing paper and canvas:
Always go for a paper that has a TAPPI rating of around 90.
Paper Comments
RC Photo Papers
Metallic Papers
Fiber-Based Baryta Papers baryta papers have set the gold standard for fine photographic papers with incredibly rich blacks, wide gamuts, great contrast and sharpness, and smooth tonal transitions. These papers are perfect for black and white prints.
Cotton Fiber Papers
Alpha-Cellulose Papers
Hot and Cold Press Papers
Canvas Papers
Specialty Papers
What Process Why We Love It Great For Not As Great For
GLOSSY Inkjet Delivers colors that are saturated and bold and shows plenty of detail Competition prints, framed work and owners of high-resolution camera bodies Clients—cast-coated glossy papers can look and feel more like plain paper
SATIN and LUSTER Its textured surface reflects less light and can smooth skin tones Portraits Images where very fine details are critical
MATTE Matte’s smooth and non-reflective finish and variety of weights make it a beautiful choice for fine-art prints Landscapes, black-and-white prints, lower-contrast photos and portfolio images Prints that will be handled often. Pigment inks can sit on the surface of matte papers, making them prone to scratching
COTTON and RAG Cotton papers usually deliver a warm white and a premium feel Portraits, fine art, nature scenes High-volume prints, as it can be pricey
BARYTA Delivers colors that are saturated and bold and shows plenty of detail Competition prints, framed work and owners of high-resolution camera bodies Prints that will be handled often as the surface is typically delicate
CANVAS The classic fine-art media, canvas can be stretched and hung for beautiful wall art with its no-glare surface Wall art, gallery wraps Prints that will be handled often as ink can be scratched off the surface (coating it helps)
METAL PRINTS Dye-sub transfers or direct-to-metal printing using UV printers The metal resists damage from humidity, fading and is scratch-proof Images with silver, high-contrast, high-resolution and metallic subjects like cars, skyscrapers and robots Portraits, since skin tones don’t always translate well on brushed aluminum
SILVER HALIDE Digital photographic This versatile material has a wide color gamut and continuous tone Mid-gloss photographic papers (between a matte and full glossy) are particularly suited for portraits Long-term archiving as silver halide prints don’t resist fading as well as some inkjet paper or ink combinations
Use Paper Notes Have
Monochrome Canson Infinity Baryta Prestige 340gsm The baryta nomenclature tags it as having the look and feel of silver-halide prints. Inks seem to blend in with the surface rather than sit on top, a very pleasing look. 1
Rich Color with a Satin Surface Hahnemühle FineArt Baryta Satin a satin-gloss surface on a bright white base. The barium sulfate in the coating works in conjunction with a microporous ink receiving layer to deliver vivid prints from subtle to intense color images: monochrome prints express deep, rich blacks as well as a wide gamut of grayscale values. 1
Art Prints and More Moab Entrada Rag Textured 0
Color Richness Epson Legacy Platine smooth yet not so shiny that it reflects back like “harder” glossy papers are apt to do. . On the other hand, it has the print “snap” and contrast edge enhancement we’ve come to associate with a glossy surface. 1
The “Profiled” Print and the Case for Printer-Branded Papers Canon and Epson Canon Glossy 1
Multimedia Red River Paper 88lb. Polar Matte Magna 0
Cotton/poly is the most common. The reason I've heard from manufacturers is that it is the best compromise between durability, lack of shrinkage and flexibility. 100% cotton is usually more flexible and easier to stretch but at the expense of possibly having to restretch it again later.
 
Besides composition, the other variables are thickness, surface texture and whether it contains OBAs.
 
From what I gather from the various printing forums, the most commonly used canvas is Breathing Color's Lyve (no OBAs). It is a matte canvas that offers a good combination of price, a relatively smooth texture, and a high quality image. It is also one of the few canvases that can be used with either PK or MK ink and still look good (I use it with PK).
 
The smoothest surface is provided by Canson's Museum ProCanvas, which is 100% cotton and OBA free. It comes in either matte or satin and is very soft and easy to stretch.
 
The canvas with the whitest background (without OBAs) is Premier Art's Museum Bright Satin canvas.
 
Supposedly the whiter the coating the better the colors, but I haven't seen much difference compared to either Lyve or Museum ProCanvas. It's also slightly more expensive than the others.
For your best images, consider satin luster photo paper if your content has a high level of contrast or is highly saturated. The textured surface yields brilliant color saturation and deep black density.
Coating canvas:
A fast PC (PC2)
Armenia, 5 Apr 2019
This will run Zerene doing the image stacking (using Zerene's "capability to ingest and stack new images on the fly as they are acquired by a tethered camera").  PC2 also runs Photoshop and the Zerene editor for post processing. 
Lake Placid, 3 Nov 2019
Components I used:
What Which Details Price
Motherboard MSI MPG Z390 Gaming PRO Carbon AC Supports i9-9900K LGA1151, 4 x DDR4 memory slots, supports up to 128GB, 2 x M.2 32Gb/s; 6 x SATA 6Gb/s, 3 x PCI-E 3.0 x 16 slots, DisplayPort / HDMI, 1 x Intel I219-V Gigabit LAN, USB 3.1 Gen 2 (3 x Type-A, 1 x Type-C), 2 x USB 2.0, Intel Wireless-AC 9560 (WiFi 2.4GHz/5GHz, BT 5), 1 x DisplayPort 1.2 4096X2304@60Hz, ATX $210
CPU i9-9900K 8 cores, up to 5.0 GHz Turbo unlocked, Intel UHD Graphics 630, 3 x 4096x2304@60Hz $485
Cooler Cryorig R1 Universal CR-R1B Dual Tower CPU Heatsink with XF140 and Xt140 fans. It just barely fit. $132
RAM CORSAIR VENGEANCE LED 64GB (4x16GB) DDR4 3200MHz I think I should have gotten 128GB. It was a lot more expensive. $818
GPU PNY NVIDIA Quadro P1000 (VCQP1000-PB) NVIDIA's Pascal, 2.713” x 5.70” Single Slot, 1.9 FP32 TFLOPS, 640 CUDA Cores, CUDA API, 47 W $329
SSD 2 x Samsung SSD 960 EVO Series - 1TB PCIe NVMe
(MZ-V6E1T0BW)		 2 x $418
Disk Drives HITACHI 0F14683 Ultrastar A7K4000 4TB 7200 RPM, 64MB cache, SATA, 6.0Gb/s, 3.5" $84
Power Supply Corsair TXM Series, TX750M, 750 Watt, 80+ Gold Certified $113
Case CORSAIR OBSIDIAN 750D Full-Tower Case - Airflow Edition (CC-9011078-WW) Nine PCI-E expansion slots, up to ten 3.5" or 2.5" drives and three 5.25" drive bays. $158
O/S Microsoft Windows 10 Pro $188
Monitor BenQ SW271 27 Inch 4K HDR Professional IPS Monitor 10-bit with 14-bit 3D LUT, hardware calibration, 99% Adobe RGB coverage $1,099
Network Attached Storage (NAS)
Lake Placid, 15 Sep 2019
I bought a Synology 8 Bay NAS (DS1819+) and two 10 TB WD Red Pro drives.  That should be enough to store 29,000 61MP TIFF images (assumming I configure the drives for mirroring, ie. RAID 1) or 7,200 240MP TIFF images (likewise, configured RAID 1).  I plan to configure the RAID that way.
 
Eight bays plus an option of two 5-bay
add-ons for a maximum capacity of 216 TB (8x12 TB + 2x5x12 TB).  Four Gigabit Ethernet ports.  Good reviews.  At a reasonable price.

Lake Placid, 29 Oct 2019
Here are the cost and storage capacity (16-bit TIFF images) for Seagare Ironwolf disks in Oct, 2019.  I initially bought two 10 TB disks which I have configured as RAID 1 (mirroring).
TB Oct 2019 Cost # 61 MP TIFF files # 240 MP TIFF files
1 $60 2,897 723
2 $80 5,793 1,446
3 $97 8,690 2,169
4 $100 11,586 2,893
6 $165 17,380 4,339
8 $218 23,173 5,785
10 $272 28,966 7,232
12 $354 34,759 8,678
14 $430 40,553 10,124
16 $530 46,346 11,570
Here is a plot of cost per 1000 16-bit TIFF images vs disk size.
Surge Protection
Lake Placid, 1 May 2019
I read that Ethernet Switches are vulnerable to power glitches.  I see many refurbished switches for sale. 
 
At first, I thought I should get a battery backup, ie. an Uninteruptable Power Supply (UPS).  So, I estimated the power requirement of a UPS as follows:
  1. 46W max for the switch.
  2. 5.7W x 48 = 274W for the Raspberry Pi's.  (using the max. of 5.7W, 5.1W and 0.980 x 5.1V = 5.0W (400% busy, no attached devices)).
  3. 40W for the NAS (source).  Although that doesn't include the optional DX517 expansion unit.
The total of 360W+ led me to this 450W CyberPower CP800AVR (which was highly rated in this review).
 
Great, but then I saw that that review also said,
"Unfortunately, most affordable UPS units don’t offer much protection compared with a dedicated surge protector. In previous tests, electrical engineer Lee Johnson took apart our UPS samples to examine their guts. Based on his assessment, we found that our picks should protect your equipment about as well, if not for as long, as basic surge protectors we’ve tested before."
 
So, here is their most recent review of surge protectors.  It has a pretty glowing review of the Furman Powerstation 8.  A little pricey (for me) but hopefully worth it.
 
I've ordered four.  One for the ethernet switch and NAS, one for the printer and two for other electronics.
Furman Powerstation 8
Post processing, Stacking, etc.
Lake Placid, 22 Aug 2019
 My Target system

Armenia, 25 Apr 2019
Lake Placid, 14 Aug 2019
Optically, this is the sort of system (left) I'd like to initally emulate using my components (right).  That is, add to my current system the linear polarizers, retardation
Olympus CX31-P
My target
(click on image to enlarge)
material, a light source and whatever bits of 3D-printed holders I need to glue it all together (the grey and yellow parts in the animation).

I feel it takes a long time to process a large stack - converting the Raw Pixel Shift images from the camera to TIFFs for Zerene being much of it.  It requires several manual steps - easy drag-and-drop operations, but I have to be there to do it.  I'd like to speed up the process by souping up my computer system and removing the manual steps.
 
Computationally, this is the system I'd like to build.  My goal is to automate the end-to-end process.  And convert and stack the images as they come off the camera.  To accomplish that, I am putting together a Multi-processor (MP) from single board computers (Raspberry Pi's)
and planning to build a fast PC (PC2) to do the stacking and post processing.
 
I'm planning to locate the MP, NAS and Ethernet switch in a separate room to avoid their cooling fans from affecting the image capture.  Everything is connected together by Gigabit Ethernet.  There is also a Wifi connection from the Controller on PC1 to a web server-based monitor so I can check on status using my laptop, tablet or phone.
 
References:
- Old Microscopes
Leitz Dialux, Early 1970s many parts and accessories are available on the market
American Optical Series 110, Mid-1980s [a so-so review]
Zeiss Standard, Late 1970s higher level are the Universal or Photomic models
Nikon Labophot, Mid-1980s an "entry level" benchtop scope, less expensive than the Optiphot line (which are rare in POL), unsurpassed optics, large controls
Wild Heerbrugg M21, Early 1970s considered by some experts to be the best microscope ever made
Olympus BH-2 BHTP, Late 1980s "the microscope I prefer over all others"
- New Microscopes
- Research Microscopes
Fourier Ptychography:
Mesolens:
Mesolab website -- 4X mag, NA 0.47, field size 6mm, WD 3mm, resolution 0.7 µm lateral, 7 µm axial
Polychromatic polarization microscope:
- DIY Microscopes
        Bertrand Filters
            New Bertrand Lens on Ebay
        Condensers
            Substage Condensers
        Lighting
            Tungsten halogen vs white LED
            Optiphot modular light source
            FAQ:How (and why) to use electronic flash at the microscope?
            Microscope Conversion to LED-Light
            HOW BRIGHT IS YOUR LIGHT PART III: LED CONVERSION
        Rheinberg Filters
            Making Rheinberg illumination discs
            How to Make Rheinberg Filters
            Rheinberg Filters for Photomicrography
   - Technique
            A Practical Guide to Creating Superresolution Photos with Photoshop
- New Microscopy
Raspberry Pi MP
Armenia, 15 Apr 2019                 Implementation
I'd like to speed up the processing of the images (conversion of the 4 ARW images from the camera to the ARQ pixel-shifted image to a 16-bit TIFF image which is then merged into the current stack) by creating a pipeline for that processing.  A pipeline that could keep up with the image capture (4 ARW images every 10 or 15 seconds).  A cost effective way to do that may be to develop a small Multiprocessor (MP).  Well, the fun way.
 
I'd like to build this Multiprocessor out of the latest version of the Raspberry Pi (Raspberry Pi 3 Model B+).  The Pi 3 B+
Photo by Joshua Kiepert
is clocked at 1.4 GHz, has 1 TB of RAM and supports Gigabit Ethernet at 300 Mbps.
 
I will model mine after Joshua Kiepert's RPiCluster.  He has documented his system well, including Eagle files for his power distribution cards.
 
At the end of that article, it says his 32-node system (using an older version of Raspberry Pi) does 10.13 GFlops, where "the first Cray-2 supercomputer in 1985 did 1.9 GFlops".  I look forward to benchmarking mine.  Let's see, Summit did 200,795 TFlops in the latest Nov 2018 Top500 List.  How close will I be?

Status: My new MP monitor based on my old Raspi Monitor is coming along.
 
Lake Placid, 24 Oct 2019
Since 2008, I have been spending my winters on my boat in the Bahamas.  Unfortunately, Hurricane Dorian has totalled my boat.  This winter, I plan to spend Jan-Apr in Peru.  I'm bringing my Raspberry Pi MP with me to work on.  I've made up a travel kit consisting of: I chose each component to be as small and capable as possible.  They mostly fit in a small plastic box as shown here (along with the power cords for the USB power supply and Ethernet switch).  Seven 1' USB cables, six 1' Ethernet cables and the keyboard will travel in a plastic baggie.
 
I need to cut holes in the box for access to some of the connections and to secure the components.  I may 3D print a more custom storage box.
 
It takes just a couple minutes to convert from travel mode to operational - just a matter of connecting things together.
 
This is probably half the size of the equivalent kit I brought to Colombia for a month this past April. 
Set for travel
Setup for use
GPU
Lake Placid, 3 Nov 2019
Somebody asked why I'm using Raspberry Pis and not GPUs (Graphics Processing Units).  After all, contemporary supercomputers use GPUs.  My feeling was that Pis would be easier to use, be more widely wrung out of the kinks and better supported in general.  That there would be more open source available thus enabling the large developer community the access they need to sort out bugs and procedures.  I figured they would simply be more fun.  But I took a look at some GPUs, summarized here.  The difference in LINPACK FP32 floating point performance is significant.
Series Model FP32
(TFLOPS)		 Power
(W)		 Price W/
TFLOP		 $/TFLOP
Quadro Pxxx P1000 1.894 47 $316 25 $166
P2000 3.000 75 $453 25 $151
P4000 5.300 105 $750 20 $144
P5000 8.873 180 $1,625 20 $183
P6000 10.882 250 $3,539 21 $325
GeForce 10xx, 16xx, 20xx GTX 1650 2.661 75 $150 28 $56
GTX 1060 3.470 120 $183 35 $53
GTX 1660 4.308 120 $230 28 $53
GTX 1660 Ti 4.608 120 $265 26 $58
RTX 1070 5.783 150 $519 26 $90
RTX 2070 7.465 175 $539 23 $72
GTX 1080 8.228 180 $450 22 $55
GTX 2080 10.600 225 $799 21 $75
RTX 2080 Ti 13.450 250 $1,463 19 $109
Titan RTX 16.300 280 $2,499 17 $153
Quadro RTX x000 RTX 4000 7.119 160 $865 22 $122
RTX 5000 11.151 230 $2,272 21 $204
RTX 6000 16.3 295 $3,489 18 $214
RTX 8000 16.3 295 $5,500 18 $337
Rasp-
berry Pi*		 3b+ 0.000245 3.5 $35 14,300 $143,000
4 0.000927 4.4 $35 4,750 $37,800
Intel* i9 9900K 0.476 180 $472 378 $992
* - These Power Consumption and Prices include the CPU, so not a fair comparison with the GPU-only values above.  Thus, they are greyed out.

Here is a plot of the PNY NVIDIA Quadro cards (squares) and NVIDIA GeForce cards (circles).  You can see the effect of progressing technology in the Price-performance of the Quadro GPUs vs the GeForce GPUs - the Quadro GPUs were introduced in late-2016 to early-2017 while the Series 16 and 20 GeForce GPUs were introduced in late-2018 to mid-2019.  The large green points at [13.45, $1,463, 2080 Ti] and [16.3, $2,499, Titan] look like sweet spots.
 
By happenstance, I already have a starter GPU - a PNY NVIDIA Quadro P1000 (the lowest performer, lowest cost GPU shown here) on my new PC.
 
I have some spare slots in my PC case and as a next step, may get one of the 2080 Ti or Titan RTX later.  Although, prices will continue to decrease and performance increase, so no rush - especially if the P1000 meets my immediate needs.
 
Then, if I want to get serious (ie. expand beyond what will fit in my current case), I should be able to rebuild my system without throwing away too much.  It looks like GPU Servers (for Deep Learning, etc) are configured with multi-core i9s (which I have).  GPU cases (with room for 8 and more GPUs, say) will take ATX motherboards (which I have) and are fairly cheap ($100-150).
 
Notes:
  • My immediate goal (converting ARQs or possibly ARWs to TIFFs) may not be represented by this benchmark (LINPACK FP32).  There are bound to be some surprises.  For example, my application may be I/O intensive (getting to the image files on the NAS).  Will that be a bottleneck in the GPU configuration compared with the MP?  TBD.
  • There is a GPU on the Raspberry Pi.  The measurements above probably didn't use it.  And may not have used the hardware FP feature on the Pi's CPU.
  • I do have other things that I'd like to use the MP or GPU on.  GPUs would clearly be better for deep learning, say.  On the other hand, a Pi-based MP would likely be more useful for playing with IoT (Internet of Things).  Of course, this is all very grandiose at this point.
  • To allow me to play with programming a GPU while travelling,
    • I looked at eGPUs and being able to program them from my Macbook.  They seem bulky and heavy. 
    • I see now that the thing to do is get a "Gamer's Laptop".  The Razer Blade Stealth 13 GTX FHD Model with an NVIDIA GTX 1650 looks nice.  It is the same size and weight as my 13" Macbook Pro.  The FP32 isn't too shabby.  I think that is what I'll do - carry it as my laptop on my next trip (4 months in Peru) along with the Pi MP travel kit described above.   Four months is a long time.
Model GPU FP32
(TFLOPS)		 Size (in) Weight
(lbs)		 Price $/TFLOP Notes
Acer Predator Triton 700 GTX 1060 6GB
GTX 1080 8GB		 3.935
8.228		 15.5 x 10.5 x 0.74 5.4 $1,242
$1,499		 $315
$182		 "A Powerful laptop for deep learning"
Asus ROG Zephyrus GX501 GTX 1080 Max-Q 8GB 8.228 14.9 x 10.3 x 0.7 5.0 $1,549 $188
Dell Inspiron i5577 GTX 1050 4GB 2.3 15.7 x 10.4 x 0.99 5.7 $1048 $455 "The cheapest laptop for Deep learning"
No longer listed on Dell's site
Eluktronics Pro-X P650HS-G GTX 1070 8GB 5.783 15.1 x 10.6 x 1.13 6 $2,000 $346
GIGABYTE Aero 15X GTX 1070 8GB 5.783 14 x 9.8 x 0.74 4.5 $2,000 $346
Lambda TensorBook (specs) RTX 2070 Max-Q 8GB
RTX 2080 Max-Q 8GB		 7.465
13.45		 15.0 x 9.80 x 0.73 4.2 $2,600
$3,100		 $350
$230		 Some pretty negative reviews
Razer Blade 15 GTX 1060 Max-Q 6GB
GTX 1660 Ti 6GB
RTX 2080 Max-Q 8GB		 3.470
4.608
10.600		 14.0 x 9.25 x 0.78 5.2 $1,560
$1,600
$3,000		 $450
$347
$283		 "Ultra Fast. Ultra Small. Ultra Powerful"
Razer Blade Stealth 13
GTX FHD Model 		GTX 1650 4GB 3.0 12.0 x 8.27 x 0.60 3.1 $1,800 $600
Ethernet Switch
Armenia, 12 Apr 2019
I'd like to be able to support up to 48 Raspberry Pi's in the MP described above.  Add a port for the Master Node, two for the NAS, one for the PC1+PC2 (that I'll bring in from a separate switch
where they're located at my desk).  Making 52 ports all together.  This 52-port switch looks good.  Gigabit speed at a reasonable price.

Lake Placid, 21 Oct 2019
I've already bought one of the above 52-port switches and am having a bit of buyer's remorse.  I'm guessing Power Over Ethernet (PoE) will become more common and affordable in these large Ethernet switches in the future.  I see it increasingly available.  And will be built into a future version of Raspberry Pi at some point (soon?).  PoE is presently available on the Raspberry Pi as an optional $30 "hat".  And this $15 splitter looks interesting.  But I guess I'll wait until the PoE is integrated into the Pi.

-- FIN