- Messages
- 20,926
- Name
- Steve
- Edit My Images
- Yes
If you were building a dual CPU and dual GPU system which would you go with and why?
Dual Xeon or dual AMD EPYC
- Thread starter SFTPhotography
- Start date
As an ex server man, I would always go with Intel because I have found the Intel chipsets more stable.
Like @Brazo, I am curious about what you would use it for? I have played with using servers as 'desktops' in the past and found the inability of the O/S and applications being able to simultaneously exploit the hardware fully was an issue.
A server platform running multiple instances of O/S images with Vmware or Hyper-V is more able to make better use of the CPU multi-socket environment than a single 'client' O/S.
Like @Brazo, I am curious about what you would use it for? I have played with using servers as 'desktops' in the past and found the inability of the O/S and applications being able to simultaneously exploit the hardware fully was an issue.
A server platform running multiple instances of O/S images with Vmware or Hyper-V is more able to make better use of the CPU multi-socket environment than a single 'client' O/S.
OP
- Messages
- 20,926
- Name
- Steve
- Edit My Images
- Yes
It would be used for image, video editing and general use (gaming etc). The real reason for wanting a dual socket system is the RAM upgrades. Scaleable processors can take on much more ram than a typical I9 or Ryzen (usually capped at 128gb). As tech moves on and apps, OS become more resource hungry being able to support higher RAM becomes valuable. They also use EC memory so far more stable too.
Last edited:
Brazo
Suspended / Banned
- Messages
- 3,687
- Name
- Mark
- Edit My Images
- Yes
Steve don’t worry about ram. 32Gb is more than enough for most applications. By the time you get to 128GB the socket and cpus will be so long in the tooth that’s the ram will be pointless.
Get an AMD 3700-3950x series (Although is wait now for series 4) and you’ll be golden and can upgrade as and when needed. Trying to future proof with dual cpus just won't work and dual gpu’s are dead in the water now, even Nvidia has quietly killed it off in the last week.
Get an AMD 3700-3950x series (Although is wait now for series 4) and you’ll be golden and can upgrade as and when needed. Trying to future proof with dual cpus just won't work and dual gpu’s are dead in the water now, even Nvidia has quietly killed it off in the last week.
I run dual Xeons, dual Quadro gpus and 128gb at the moment in my workstation .....
Before you go splashing cash - research what software you’ll be using and whether it can take advantage of your hardware. From my experience - dual gpus are not necessary as the drivers and software never work properly it’s a nightmare, memory, I haven’t done video editing, but for photos the most I’ve seen Lightroom use is 60gb - I can run a test in dxo & capture one if you’d like, I process d810 images to give you an idea of the size of images I work with.
My machine is a Fujitsu r series - beautifully built but it’s a little too bespoke for my liking when it comes to things like drivers and general support. I’ve got a SAS controller in mine that I don’t think has ever worked properly as the driver support stopped years ago - that happens with these things
My memory is all ecc registered but for workstation applications I’m not convinced it’s of any benefit ....
More modern software may leverage your gpu more over cpu nowadays - so that’s something to consider.
If starting again - I’d look at the software packages I’d plan to use - Let’s take Lightroom for example - LR does multi thread but it’s much much better to have faster cores than more slower ones. Ram - 64gb - my LR doesn’t really use my gpus - so it’s a waste having them.
Most important I believe is actually your data workflow - m2 drives etc to get the information to and from the memory as fast as possible.
I upgraded my processors last night to 2 8 core Xeons @2.9-3.8ghz. It’s dropped my lr export average time from 12 seconds and image to 4.5 .... huge difference
Before you go splashing cash - research what software you’ll be using and whether it can take advantage of your hardware. From my experience - dual gpus are not necessary as the drivers and software never work properly it’s a nightmare, memory, I haven’t done video editing, but for photos the most I’ve seen Lightroom use is 60gb - I can run a test in dxo & capture one if you’d like, I process d810 images to give you an idea of the size of images I work with.
My machine is a Fujitsu r series - beautifully built but it’s a little too bespoke for my liking when it comes to things like drivers and general support. I’ve got a SAS controller in mine that I don’t think has ever worked properly as the driver support stopped years ago - that happens with these things
My memory is all ecc registered but for workstation applications I’m not convinced it’s of any benefit ....
More modern software may leverage your gpu more over cpu nowadays - so that’s something to consider.
If starting again - I’d look at the software packages I’d plan to use - Let’s take Lightroom for example - LR does multi thread but it’s much much better to have faster cores than more slower ones. Ram - 64gb - my LR doesn’t really use my gpus - so it’s a waste having them.
Most important I believe is actually your data workflow - m2 drives etc to get the information to and from the memory as fast as possible.
I upgraded my processors last night to 2 8 core Xeons @2.9-3.8ghz. It’s dropped my lr export average time from 12 seconds and image to 4.5 .... huge difference
OP
- Messages
- 20,926
- Name
- Steve
- Edit My Images
- Yes
D810 is what I process and capture one is something I use and will continue to use. I was planning to upgrade to medium format digital in the next few years - hence the requirement of high spec machine.
I'd be curious to know how much RAM C1 and DXO use - my understanding is C1 is particularly power hungry.
This is interesting - https://techgage.com/article/a-look-at-phase-one-capture-one-performance/
I'd be curious to know how much RAM C1 and DXO use - my understanding is C1 is particularly power hungry.
This is interesting - https://techgage.com/article/a-look-at-phase-one-capture-one-performance/
Last edited:
I'll run some tests for you with D810 images and report back to this thread
- Messages
- 18,695
- Name
- LongLensPhotography
- Edit My Images
- No
You will be more than OK for general image editing with a normal Ryzen 9 or Intel i9, AMD being faster and much cheaper. Video editing is where you will be absolutely hammered.
I found this quite helpful to tailor my spec list https://www.pugetsystems.com/recomm...-Adobe-Photoshop-139/Hardware-Recommendations
Adobe Lightroom Classic
Our workstations for Adobe Lightroom Classic are tested and optimized to give you the best performance and reliability.
www.pugetsystems.com
Most MF bodies are only as "bad" for computing power as a humble Canon 5Ds or Sony A7RIV.... Only one or two go above 100MP.
Now come video editing it is shock and horror https://www.pugetsystems.com/recomm...-DaVinci-Resolve-187/Hardware-Recommendations
You might as well get 3 cards in there if you are serious about it! P.S. If you plan on using linux, be warned that nvidia drivers are a total proprietary POS and PITA. I'm going all AMD I think.
I might try waiting till black friday sales and new models which are due out any day now but it's getting super painful.
Right - I'm back.....
I was planning on running some tests this weekend following the processor upgrade (changed from xeon e5-2630 up to e5-2690) machine runs 128gb ddr 3 ecc ram and 2 x quadro k4000 gpus.
Anyway, I have tested the following 3 bits of software:
LR
DXO photolab 3
Capture One 20
Findings:
Capture 1 test
Images: 728
Format: RAW
Imported with preset SPR-09
Import observations:
Coming from card (sandisk extreme pro USB3.0)
CPU: approx 16%
Memory: 9gb
GPU: 6% (on a single card, 2nd card not showing activity)
Import from card: 4mins 3.5s
Previews: 9mins 18 seconds for 728 previews
Sharpening and nr for all images
export as jpeg @ 100% with no sharpening for screen
Export v slow: both gpu's working between 0-40%
CPU approx. 9%
mem: 11gb
approx time: 1hr
OPEN GL disabled : export time approx 1hr 20 mins -
cpu @ max 65%
DXO photolab 3:
Copy files with windows from cf card: 1min 30
DXO prime NR on all images + unsharp mask and local adjustments
Export set to process 32 images at a time
RAM: 35gb
CPU 100% @ full turbo speed after 2 mins
32 images in 13mins 16 seconds : avg. 25 seconds / image (down from 45 with 2x6 core @ 2.6)
DXO open gl was disabled as DXO recognised gpu's and suggested cpu was faster than gpu
Lightroom:
cpu 100%
ram circa. 35gb
av. 4.5 seconds / image
General observations:
In general - huge ram is not going to be of benefit with any of these packages.
GPU grunt will get you the best performance and fast individual clock speed will help for LR and DXO.
Capture One requires a very fast/powerful gpu to get the maximum performance from the software. I didn't have time to disable one of the gpu's but i could see this yielding better results. My set up is not optimized for Capture One.
DXO performs in a similar way to LR, but not as ram hungry as LR. Processor heavy and faster individual cores are preferred over more, slower cores.
LR performs well, but sucks ram up in a big way, especially when doing editing etc. HDR image merging really consumes rams, so if that's a use case then go heavy on memory.
LR exports faster than DXO, however DXO 'prime nr' is notoriously slow, regular nr yielded similar results to LR, with DXO being slightly quicker.
FYI - fan speeds were maxed out doing these tests, especially on the DXO test, I would recommend looking at cooling if heavy cpu loading will be regular.
Not the most scientific of tests, but hope it has been helpful.
Chris
I was planning on running some tests this weekend following the processor upgrade (changed from xeon e5-2630 up to e5-2690) machine runs 128gb ddr 3 ecc ram and 2 x quadro k4000 gpus.
Anyway, I have tested the following 3 bits of software:
LR
DXO photolab 3
Capture One 20
Findings:
Capture 1 test
Images: 728
Format: RAW
Imported with preset SPR-09
Import observations:
Coming from card (sandisk extreme pro USB3.0)
CPU: approx 16%
Memory: 9gb
GPU: 6% (on a single card, 2nd card not showing activity)
Import from card: 4mins 3.5s
Previews: 9mins 18 seconds for 728 previews
Sharpening and nr for all images
export as jpeg @ 100% with no sharpening for screen
Export v slow: both gpu's working between 0-40%
CPU approx. 9%
mem: 11gb
approx time: 1hr
OPEN GL disabled : export time approx 1hr 20 mins -
cpu @ max 65%
DXO photolab 3:
Copy files with windows from cf card: 1min 30
DXO prime NR on all images + unsharp mask and local adjustments
Export set to process 32 images at a time
RAM: 35gb
CPU 100% @ full turbo speed after 2 mins
32 images in 13mins 16 seconds : avg. 25 seconds / image (down from 45 with 2x6 core @ 2.6)
DXO open gl was disabled as DXO recognised gpu's and suggested cpu was faster than gpu
Lightroom:
cpu 100%
ram circa. 35gb
av. 4.5 seconds / image
General observations:
In general - huge ram is not going to be of benefit with any of these packages.
GPU grunt will get you the best performance and fast individual clock speed will help for LR and DXO.
Capture One requires a very fast/powerful gpu to get the maximum performance from the software. I didn't have time to disable one of the gpu's but i could see this yielding better results. My set up is not optimized for Capture One.
DXO performs in a similar way to LR, but not as ram hungry as LR. Processor heavy and faster individual cores are preferred over more, slower cores.
LR performs well, but sucks ram up in a big way, especially when doing editing etc. HDR image merging really consumes rams, so if that's a use case then go heavy on memory.
LR exports faster than DXO, however DXO 'prime nr' is notoriously slow, regular nr yielded similar results to LR, with DXO being slightly quicker.
FYI - fan speeds were maxed out doing these tests, especially on the DXO test, I would recommend looking at cooling if heavy cpu loading will be regular.
Not the most scientific of tests, but hope it has been helpful.
Chris
I find this site useful for comparing raw computing power between processor types and vendors. You can use it to see what impact dual CPUs has vs a single CPU system.
If you compare the CPU Mark rating for a single Intel E5-2690 V4 (rated at 17678) vs dual (rated at 27563) you will find the increase in processing power adding a second CPU is around 56%
By comparison a single i9 9940X is rated at 28242, with lower power consumption and a faster individual core speed.....
PassMark Software - CPU Benchmarks
PassMark Software - CPU Benchmarks - Over 1 million CPUs and 1,000 models benchmarked and compared in graph form, updated daily!
www.cpubenchmark.net
If you compare the CPU Mark rating for a single Intel E5-2690 V4 (rated at 17678) vs dual (rated at 27563) you will find the increase in processing power adding a second CPU is around 56%
By comparison a single i9 9940X is rated at 28242, with lower power consumption and a faster individual core speed.....
- Messages
- 18,695
- Name
- LongLensPhotography
- Edit My Images
- No
I find this site useful for comparing raw computing power between processor types and vendors. You can use it to see what impact dual CPUs has vs a single CPU system.
PassMark Software - CPU Benchmarks
PassMark Software - CPU Benchmarks - Over 1 million CPUs and 1,000 models benchmarked and compared in graph form, updated daily!www.cpubenchmark.net
If you compare the CPU Mark rating for a single Intel E5-2690 V4 (rated at 17678) vs dual (rated at 27563) you will find the increase in processing power adding a second CPU is around 56%
By comparison a single i9 9940X is rated at 28242, with lower power consumption and a faster individual core speed.....
AMD 3950X in turn destroys that and the mega expensive Threadripper is way ahead at the top. It is impressive to see all the entries at the top occupied by AMD. Can't wait for 4th gen!