A Ramble About Computers...

[simsurace]

Having a lot of memory definitely makes a big difference when stitching.  With 8 GB on the Macbook pro things get really slow if there are a lot of frames.  I mean two cups of coffee slow.  With 16gb on my desktop 16 frames are stitched pretty quick.  32 GB should handle anything I will ever do.

I don't know what you stitch with, but the latest PTGui version added GPU support, now stitching is much faster. I didn't time it, but I don't think you could even make half a coffee when you have less than 30-40 images to stitch. I'm using a late 2014 MBP with SSD and 8Gb of RAM.

[Ron Scubadiver]

Have been stitching with ACR.  Time to look at ptgui

[Bjørn Rørslett]

PTGui can indeed be surprisingly fast when you stitch a large number of frames. Having massive amounts of RAM and SSD disk(s) helps too.

[Michael Erlewine]

a custom built desk top can deliver huge performance advantages at affordable prices, particularly if you are into heavy image stitching or stacking. 

I stack and I stack 36MP layers, sometimes 150 at a shot. I use Zerene Stacker, but there are others that work too.

As for computers, I stopped buying premade computers systems years ago and began having them built for the specs we need to process stacks, panos, video, etc. I am on my third custom-built computer and each has been increasingly expensive, because time is worth life, and I am getting older.

It is one thing to have a fast computer, but you need software that can access that speed as well, especially with graphics. IMO, forget about building your own. Even forget about having the local computer guru build one for you. Spend some money and have it tuned to what you actually need.  I had one built by Boxx. It worked, but the staff there tried too much to upsell me with B.S..  Even tried to tell them to desist, but they were just rude.

The best company I know of is Puget Systems, out of Auburn Washington.

https://www.pugetsystems.com/

This company and the folks who work there are beyond what we can expect. They even call and re-tune your system, as well as provide a complete notebook with backup disks, photos of your particular system, and every last part on it. I cannot recommend them enough. That’s the good news.

The bad news is no different than any of these companies. These fast computers cost money and a lot of it. They have to be budgeted just as we budget money for cameras and lenses. At this point I have a very fast system, but it is still too slow sometimes. I have several Macs, but I only use PCs for photography because I find they are faster, less expensive, and more easily interfaced to peripherals.

Right now I am using a water-cooled system based on:

Asus X99 Deluxe Motherboard
Intel Core i7 5960X 3.0 GHs eight-core 20MN 140W (overclocked) CPU
128 GB DDR-4 2133 REG ECC RAM
Dual PNY Quandro K5200 PCI-E 8GB GPU video cards
And a variety of internal SSD and SATA drives, connections, etc.

The bad news is that it cost more than $10K....

This machine does well with still and video photography, but is IMO still held up by Adobe’s software, as far as I can tell.  Here are some photos.

[Ron Scubadiver]

I checked on the Ptgui site.  GPU computing does not support INTEL IGP's which is what I have on my notebook.  My desktop is pretty fast with ACR except on the largest ones I have done where the thing runs out of memory.

As for home brew computers, I have done many.  Puget Systems is a good source of information regarding the parts they use.  Lovely system, Michael, especially the dual Quadro gpu's and tons of memory.

[simsurace]

Yes, I forgot to say that I use a dedicated graphics card.

[Frank Fremerey]

Ptgui?
Try Kolor.fr
Evalutate.
Smokes them all.

[Akira]

Ptgui?
Try Kolor.fr
Evalutate.
Smokes them all.

LOL!  This is amazin, Frank!

[Erik Lund]

PtGui is about precision, calculated re-positioning of pixels, works for my stitched architecture superbly.

[Frank Fremerey]

M. Erlewine. That is a serious sytem. Based on a Lian Li case and a great power supply.
That looks pretty much like my 2009 system. Apart from the fact that I use only one
Graphics Board and a slow one. I do not play much and I am not sure
If the computing power of the GPUs can be fully utilized for number
crunching. Maybe it is possible today.

[Ron Scubadiver]

Just for fun I looked up a few of the components in Michael's workstation;

CPU  $1,050
GPU, around $2,000 each
Memory is around $1100, but I  could not find quickly find a quote for a 16x8 ecc kit.

Other components are not super expensive, but a lot of the cost is in having someone getting all of it to work together.

[Andy]

Memory is around $1100, but I  could not find quickly find a quote for a 16x8 ecc kit.

Ron,
the CPU doesn't support ECC memory (only Xeon's do). No need to pay a premium for ECC memory. Not sure, why there is 128 GB memory in the system, as the CPU supports 64 GB max.
http://ark.intel.com/products/82930/Intel-Core-i7-5960X-Processor-Extreme-Edition-20M-Cache-up-to-3_50-GHz

If you want to speed up your system, avoid connecting your fast drives on the SATA/SAS ports located on the motherboard (starting with 3 SSDs the DMI is the bottleneck), use a cheap LSI host bus adapter on PCI Express for your SATA SSDs or connect your SSD directly to the PCI Express bus. Don't spend money on RAID controllers, not needed.

In case you want to build a fast system without too much financial investment:

The dual Quadro Setup is about as fast as a single Titan Black (approx 1000$; FYI: the Titan has no ECC graphics memory). If you don't need high double precision performance, use the GTX 980 Ti instead. This card is faster than the Titan in graphics and costs about 600-700$

If you want lots of cores, look up Intel's Xeon E5-2670 (v1) on eBay. They are well below 100$ / piece and you get 8c/16t per socket. Imho, currently the best bang for the buck if you intend to run parallel aware applications. Original SRP was 1500$/pc then. (look for C2 stepping - SR0KX)
http://ark.intel.com/products/64595/Intel-Xeon-Processor-E5-2670-20M-Cache-2_60-GHz-8_00-GTs-Intel-QPI

If you have really computationally intensive workloads where CPUs get hot, use watercoolers for approx 60$ each. Performance increase is usually in the 30% range. But only if the load is for longer time. Short spikes don't make much difference.

Combine it with a dual socket workstation MB - ca. 300$. The architectural limit of the Xeons is 768 GB main memory - in case memory prices go down and your needs up.

Use cheaper (but more) 1600 MHz ECC memory. Usually, the perf difference to 2133 MHz memory is in normal apps in the single digit range. 64 GB is most often enough (unless you have a stitching app, which can really use more memory). Be conscious about your memory needs. Too much memory in the system often slows down the startup times of apps. (The OS need to clean the memory for security reasons before the app can use the allocated memory. As example: Cleaning 256 GB takes about 60 seconds, unless a special API is used. Most apps don't use it. With the optimized API it takes about 2 seconds)

Building a balanced system is always fun - enjoy

Andy

[Ron Scubadiver]

Andy, I appreciate your analysis, but you are talking about things that are way beyond my needs.  Someone who needs cutting edge performance will read your post and be grateful.  My main gripe is 5 years of technology has not produced much of an improvement in photo editing performance owing to Intel's quest for energy efficiency, larger MP files and Adobe releasing consecutively slower versions of ACR and Lightroom.  I may have to let the desktop upgrade go for a month or two because I have some other more pressing challenges .

[Frank Fremerey]

That is a cool Idea: Intel's Xeon E5-2670.

On German Ebay you get two of these for ~260 Euros incl. deductible VAT. Intel's original asking price was 1550 US$ per piece.

16 physical Processors & 32 Threads for less than 600 Euros including new motherboard? WOW.

Thank you.

Question: Which Xeon-Motherboard (two way) do you recommend?
Is this Xeon fit for Two Way?
I remember in olden days Tyan and Supermicro were the weapons of choice.

[Erik Lund]

Andy that's serious good advise here! Thanks