AMD Opteron 32-Core - is the Adobe Suite using 32 Cores?

[Fabian Benninghoven]

Hi folks,

like many others i am waiting for the best MacPro alternative option. Maybe Z800/Z820, or a DIY system with an I7 2x6-Core.
But now i checked AMD and the prices are low atm.

CPU: 2x AMD Opteron G34 16 Core x 2,60 Ghz (6282) = 1850 Euro
Ram: 64 GB (16x4) A-Data Reg. ECC 10667R = 594 Euro
Mobo: Supermicro H8DGi-O Dual CPU Sockel G34 = 490 Euro
Grafix: Asus GTX 680 = 500 Euro
SSD: 128 GB Crucial = 100 Euro
HDD: 4x 2TB Raid-5 = 400 Euro
Tower: = 250 Euro
Power: 1100 W = 180 Euro

-> 4364 Euro

Does the Adobe Suite support 32 cores? Maybe AMD is an alternative to Intel. What you guys think about?

[Subhadip Sen]

Before you dive in, consider that AMD's CPU architecture (Bulldozer) is vastly different from Intel's (Sandy Bridge). A single Opteron 62xx features 16 'cores', but more accurately 8 modules, while a single Xeon E5 26xx features 8 cores and 16 threads. Really though, the concept of 'cores' is less relevant today so we should consider that both CPUs are basically 16 threads or 8 modules. Intel's approach is to have 8 powerful cores with a hyper-thread, while AMD's approach is to have 8 tightly packed modules with each module having 2 dedicated integer units but sharing a lot of other things. In short, Intel = 8 big threads + 8 tiny threads, AMD = 16 medium threads.

Considering a dual-CPU configuration, Xeon 16 core / 32 thread vs Opteron 16 module / 32 thread, if your application is going to use only 8-16 threads, then Intel will be much faster. However, if your application can saturate 32 threads, the Opterons will come into their own - in heavily multithreaded stuff like X264 it is not far away from Intel's Xeon E5s for nearly half the price. Once you consider the pricing, it is a no-brainer - Opterons are far ahead on price/performance. This is why Cray and many supercomputer vendors (where applications are optimized) have embraced Opteron in favour of Xeon. These Opterons were so successful in the high-performance computing market that it was out of stock in general retail for nearly 3 months after release. I did consider an Opteron G34 solution last year, but decided against it as much of my applications were failing to utilize 16 threads reliably. In many ways, the Bulldozer architecture is way ahead of its time - it is designed for a future where every application is multi-thread aware, schedulers are aware of the modular architecture, floating point operations are transferred to the GPU via OpenCL / C++ AMP / CUDA while CPUs are dedicated to integer operations, applications use advanced instructions like FMA4 or XOP, etc. AMD may have bet right - this future is approaching, but it is ultimately up to the software developers.

Back to your crucial question - does Adobe Suite support 32 cores? CS5.5, not really. However, CS6 is more promising. There's a test buried somewhere in the Premiere 6 thread where Luigi shows CS6 getting 100% usage across 24 threads for a transcode from R3D! So, yes, it is time to start looking at Opteron (once again) as a serious alternative. But we need a little more information as to how well CS6 does utilize 32 threads in different scenario.

If you need the most powerful system that will be running several GPUs, Xeon E5 remains the best solution, but it is difficult to complain considering what AMD are offering for the money - you could only build an 8 core Xeon E5 solution for the same price! Of course, the G34 motherboards can be upgraded with next-gen Piledriver CPUs / Opteron 63xx which will bring significant improvements in 4 months' time.

[Petri Teittinen]

Love your hardware posts, Subhadip. Kudos!

[Elsie N]

Love your hardware posts, Subhadip. Kudos!

+1. Geek prose that anyone can understand.

[Fabian Benninghoven]

Thanks... and now i am at the same point like always: Don't know which decision to make.... :001_huh:
We should buy all three systems: Z800 with 2x6Core, Selfmade I7 Extreme with 2x 6-Core and an Opteron 32 Core system.

Please donate :devil:

[Corey Culp]

Before you dive in, consider that AMD's CPU architecture (Bulldozer) is vastly different from Intel's (Sandy Bridge). A single Opteron 62xx features 16 'cores', but more accurately 8 modules, while a single Xeon E5 26xx features 8 cores and 16 threads. Really though, the concept of 'cores' is less relevant today so we should consider that both CPUs are basically 16 threads or 8 modules. Intel's approach is to have 8 powerful cores with a hyper-thread, while AMD's approach is to have 8 tightly packed modules with each module having 2 dedicated integer units but sharing a lot of other things. In short, Intel = 8 big threads + 8 tiny threads, AMD = 16 medium threads.

Considering a dual-CPU configuration, Xeon 16 core / 32 thread vs Opteron 16 module / 32 thread, if your application is going to use only 8-16 threads, then Intel will be much faster. However, if your application can saturate 32 threads, the Opterons will come into their own - in heavily multithreaded stuff like X264 it is not far away from Intel's Xeon E5s for nearly half the price. Once you consider the pricing, it is a no-brainer - Opterons are far ahead on price/performance. This is why Cray and many supercomputer vendors (where applications are optimized) have embraced Opteron in favour of Xeon. These Opterons were so successful in the high-performance computing market that it was out of stock in general retail for nearly 3 months after release. I did consider an Opteron G34 solution last year, but decided against it as much of my applications were failing to utilize 16 threads reliably. In many ways, the Bulldozer architecture is way ahead of its time - it is designed for a future where every application is multi-thread aware, schedulers are aware of the modular architecture, floating point operations are transferred to the GPU via OpenCL / C++ AMP / CUDA while CPUs are dedicated to integer operations, applications use advanced instructions like FMA4 or XOP, etc. AMD may have bet right - this future is approaching, but it is ultimately up to the software developers.

Back to your crucial question - does Adobe Suite support 32 cores? CS5.5, not really. However, CS6 is more promising. There's a test buried somewhere in the Premiere 6 thread where Luigi shows CS6 getting 100% usage across 24 threads for a transcode from R3D! So, yes, it is time to start looking at Opteron (once again) as a serious alternative. But we need a little more information as to how well CS6 does utilize 32 threads in different scenario.

If you need the most powerful system that will be running several GPUs, Xeon E5 remains the best solution, but it is difficult to complain considering what AMD are offering for the money - you could only build an 8 core Xeon E5 solution for the same price! Of course, the G34 motherboards can be upgraded with next-gen Piledriver CPUs / Opteron 63xx which will bring significant improvements in 4 months' time.

Between you and Jeff, Subhadip, I can't be more grateful for what you guys bring to the table. Always insightful and informative, but never critical or spiteful of others looking for answers and some education.

[Stefan Antonescu]

Love your hardware posts, Subhadip. Kudos!
+ 1

I feel very fortunate having Subhadip and Jeff on Reduser. :)

[Joseph S.]

CS6 supports all 32 threads (16 cores with hyper-threading), and it uses each and every thread at near 100% during renders.

[Gunleik Groven]

+ 1

I feel very fortunate having Subhadip and Jeff on Reduser. :)

+11

[luigivaltulini]

CS6 supports all 32 threads (16 cores with hyper-threading), and it uses each and every thread at near 100% during renders.

I'm trying with Joe, some test renders with its sandy bridge 32 core on windows 7 64bit platform.

for example, rendering red epic 4:23 clip 5k down 1080p h264. with REDCINE-X 1h 12 min , with Adobe premeire CS6 about 5 minutes. impressive.
There lighthouse 'knowledge in the coming days.


thanks
Luigi

[Joseph Coleman]

I'm trying with Joe, some test renders with its sandy bridge 32 core on windows 7 64bit platform.

for example, rendering red epic 4:23 clip 5k down 1080p h264. with REDCINE-X 1h 12 min , with Adobe premeire CS6 about 5 minutes. impressive.
There lighthouse 'knowledge in the coming days.

thanks
Luigi

5 minutes? Is this with or without rocket?

[Joseph S.]

5 minutes? Is this with or without rocket?

without a rocket.

Joe

[Jeff Kilgroe]

I'm starting to be able to test and confirm here. Finally have one of my dual E5-2687W boxes up and running. I'm being limited by disk bandwidth at the moment until all my other parts get here, but it's looking very promising. More tests / numbers soon.

The problem with REDCINE-X is it's hardly multi-threaded. Old Redcine (pre-Rocket days) was actually multithreaded alright, but didn't scale well beyond 4 CPUs, so it was always dragging its feet. The current RC-X Pro is optimized for the Rocket, but mostly ignores all your system's processors.... Too bad. Anyway, the numbers are starting to show that with properly multithreaded software, the Rocket is becoming less relevant.

[luigivaltulini]

I'm starting to be able to test and confirm here. Finally have one of my dual E5-2687W boxes up and running. I'm being limited by disk bandwidth at the moment until all my other parts get here, but it's looking very promising. More tests / numbers soon.

The problem with REDCINE-X is it's hardly multi-threaded. Old Redcine (pre-Rocket days) was actually multithreaded alright, but didn't scale well beyond 4 CPUs, so it was always dragging its feet. The current RC-X Pro is optimized for the Rocket, but mostly ignores all your system's processors.... Too bad. Anyway, the numbers are starting to show that with properly multithreaded software, the Rocket is becoming less relevant.


Jeff you have new Hp z820?

[Jeff Kilgroe]

Jeff you have new Hp z820?

I will have one of those as soon as HP ships it to me! It was supposed to ship on the 19th, then on the 26th, then yesterday... Now they're telling me the 8th. WTF???

I received some of the RAM I ordered from Micron the other day. So I have a system with SuperMicro X9DAi motherboard with dual E5-2687W CPUs and 64GB RAM up and running, but I'm lacking some other parts and need to swap some GPUs around for testing. I have a Quadro 6000 card installed right now, but I'm going to do some testing with an EVGA GTX580 Classified Ultra when I get a chance to pull it from my i7-3960X system.. and probably a GTX680 when the 4GB models arrive.

[luigivaltulini]

I will have one of those as soon as HP ships it to me! It was supposed to ship on the 19th, then on the 26th, then yesterday... Now they're telling me the 8th. WTF???

I received some of the RAM I ordered from Micron the other day. So I have a system with SuperMicro X9DAi motherboard with dual E5-2687W CPUs and 64GB RAM up and running, but I'm lacking some other parts and need to swap some GPUs around for testing. I have a Quadro 6000 card installed right now, but I'm going to do some testing with an EVGA GTX580 Classified Ultra when I get a chance to pull it from my i7-3960X system.. and probably a GTX680 when the 4GB models arrive.

Very good, let me know when you get hp z820, I really like.
Here in Switzerland are not yet available.
I'm curious to know your opinion about it :)

definitely try cs6 remain surprised how fast ;)

Luigi

[Joseph S.]

I look forward to Jeff's tests. I wish I was more knowledgeable of what tests to run. All I know is 5K to default Vimeo 1080 in CS6.

Jeff, Luigi, and Subhadip, big thumbs up. Learned a lot from you guys :)

[Subhadip Sen]

I look forward to Jeff's tests. I wish I was more knowledgeable of what tests to run. All I know is 5K to default Vimeo 1080 in CS6.

Jeff, Luigi, and Subhadip, big thumbs up. Learned a lot from you guys :)

Here's some more to try:

- General playback tests. See if it plays back full res, or it continues to 'hit the wall'. To be fair, 1/2 res is already overkill for general editing...
- Multiple streams... drop to 1/2 res and try 3-4 streams, maybe Multicam.
- Try some of the in-built video effects, maybe the new color corrector, some blurs, see how it plays back. (Of course, I am assuming you have a GTX 580) Then try to render this.
- Render to 4K H.264. Under MainConcept, choose Level 5.1 and 4096, go for a bitrate of 120 Mbps or so.
- Render to 5K DPX. (If you have ProEXR, then to EXR PIZ. Also, if you have Cineform Studio Premium then to 5K Cineform AVI. OK, I am going overboard now, but maybe 5K UT Lossless Codec as well...)

Obviously, keep an eye on the Task Manager to see how much of the CPU is being utilized for each of these tests.

I know that's a lot to ask for, so thanks a lot in advance. The results of these tests will be very helpful!

[Joseph S.]

Here's some more to try:

- General playback tests. See if it plays back full res, or it continues to 'hit the wall'. To be fair, 1/2 res is already overkill for general editing...
- Multiple streams... drop to 1/2 res and try 3-4 streams, maybe Multicam.
- Try some of the in-built video effects, maybe the new color corrector, some blurs, see how it plays back. (Of course, I am assuming you have a GTX 580) Then try to render this.
- Render to 4K H.264. Under MainConcept, choose Level 5.1 and 4096, go for a bitrate of 120 Mbps or so.
- Render to 5K DPX. (If you have ProEXR, then to EXR PIZ. Also, if you have Cineform Studio Premium then to 5K Cineform AVI. OK, I am going overboard now, but maybe 5K UT Lossless Codec as well...)

Obviously, keep an eye on the Task Manager to see how much of the CPU is being utilized for each of these tests.

I know that's a lot to ask for, so thanks a lot in advance. The results of these tests will be very helpful!

-- edit: CS6 is playing back 5K at full res. No caching, no nothing. Just straight instant playback in full res.

Unfortunately no high-end cards yet. Just a pair of Quadro 1800's for now. Waiting on the 690's hopefully in a week.

1/2 res playback in CS6 is flawless. Not even a tiny glitch. Video effects are immediate on 1/2 res as well. Three way color correction plays back real-time as I make the changes (assuming that's just the norm). With a bunch of grading, a 4:23 file is encoded in 5 minutes to H.264 1080 CBR at 20Mbps.

I did a 5K encode to DPX 4096x2048 max bit depth 4:23 file in about 30 minutes. But it took 1 hour with color correction (assuming the 1800 cards are the bottleneck).

Every render maxes out the threads the entire way through rendering.

This stuff threw me off, since I'm not a pro:

- Render to 4K H.264. Under MainConcept, choose Level 5.1 and 4096, go for a bitrate of 120 Mbps or so.
- Render to 5K DPX. (If you have ProEXR, then to EXR PIZ. Also, if you have Cineform Studio Premium then to 5K Cineform AVI.

[Joseph S.]

Another thing I found. During a QuickTime encode of 5K to uncompressed YUV 10 bit 4:2:2 (1920x1080), (4:23 took 7-8 minutes), strange that the memory maxes out at 83GB (out of 128GB). There's apparently a limit of around 80GB where it doesn't use anymore RAM, even though it was only 75% complete.

Also, during the render, there are 16 processes running called "ImporterREDServer.exe *32". (Qty 16 matches cores and not threads). 32bit and not 64?

[Olivier Madar]

+ 1

I feel very fortunate having Subhadip and Jeff on Reduser. :)

+1 Thank you !


Olivier

[Subhadip Sen]

Brilliant, thanks a lot again! Let me get this out of the way first of all - Rocket has been absolutely obsoleted.

That is pretty incredible performance, it is fair to say that Premiere Pro CS6 is now a real-time colour corrector. Quadro 1800 is definitely a bottleneck. You can check for GPU usage. If you have "Max Render Quality" checked, a GTX 580 can accelerate the render several times.

About 4K H.264, it is in the same area, choose H.264. Under Basic Video Settings select Profile as High and Level as 5.1. Then go up to the resolution and make it 4096x2048. Go for VBR 1 pass and target bitrate 120 Mbps.

For uncompressed rendering it is normal for the RAM to be cached out. You can look under Preferences > Memory, by default Adobe allocates 75% to CS applications and 25% to other applications, you can increase that to use more than 80GB. Really, you can make it 120GB, Windows does not require more than 4GB (or 1 GB, for that matter...).

The R3D SDK is 32-bit AFAIK, so all implementations will only run 32-bit.

[Joseph S.]

About 4K H.264, it is in the same area, choose H.264. Under Basic Video Settings select Profile as High and Level as 5.1. Then go up to the resolution and make it 4096x2048. Go for VBR 1 pass and target bitrate 120 Mbps.

.

4 minute 5K file with that exact setting is about 25 minutes.

BTW, saw the option for RAM, and it was already set to 120GB. Something else must be limiting it to 80GB. However, in the export above, it sat around 45GB.

[Subhadip Sen]

4 minute 5K file with that exact setting is about 25 minutes.

BTW, saw the option for RAM, and it was already set to 120GB. Something else must be limiting it to 80GB. However, in the export above, it sat around 45GB.

Did you have "Maximum Render Quality" checked? The GPU will be bottlenecking then. Even so, those are pretty impressive results all around.

Getting back to the original topic, it does seem like an Opteron G34 system could be an excellent option for Premiere CS6, aided with lots of RAM and a powerful GPU.

[Fabian Benninghoven]

I'm trying with Joe, some test renders with its sandy bridge 32 core on windows 7 64bit platform.

for example, rendering red epic 4:23 clip 5k down 1080p h264. with REDCINE-X 1h 12 min , with Adobe premeire CS6 about 5 minutes. impressive.
There lighthouse 'knowledge in the coming days.


thanks
Luigi

-> He really got a 2x 16 Core Intel? wtf? I thought there was only 8-core processors atm.


So what you guys think about an Opteron system? 32 cores and 64/96GB Ram would be nice. The only payable option is an z800 or a DIY system with I7 like Joe's.
Nobody tested an AMD yet i think... but if 32 cores are in use in CS6, maybe it's worth a thought.

[Subhadip Sen]

-> He really got a 2x 16 Core Intel? wtf? I thought there was only 8-core processors atm.

I think he meant 32 thread, i.e 2 x 8 cores.

There's a Xeon E7 out there, you can slap in 4 of those for a 40 core / 80 thread system. And then interconnect multiple systems by InfiniBand. The top supercomputers get to 100,000+ cores this way. The possibilities are endless...

I wonder if CS6 could scale to hundreds of threads? That way you could possibly transcode dozens of times faster than Rocket. 2 hour HDRx 3D clip? Done and dusted in 2 minutes.


So what you guys think about an Opteron system? 32 cores and 64/96GB Ram would be nice. The only payable option is an z800 or a DIY system with I7 like Joe's.
Nobody tested an AMD yet i think... but if 32 cores are in use in CS6, maybe it's worth a thought.

Most certainly worth a shot. Obviously it is not going to be as fast as a top end 16C Xeon E5 system, but for the price it is unbeatable provided your application can use all 32 threads. Now that we have one that can, it's worth it.

Edit - After checking out some benchmarks and doing some calculations, it seems like the Opteron 6282 SE will perform somewhere between a Xeon E5 2660 and 2670 once all 32 threads are used. Considering the 2 x Opteron 6828 is priced $2000 and the Xeons are $2700 and $3100 respectively, it seems like a no brainer. Plus, the motherboards cheaper too. I may be wrong... but even in a worst case scenario the AMD system seems like a great option.

[Fabian Benninghoven]

Thanks... maybe i will try this.

[Subhadip Sen]

Another interesting point is that 6282 SE is actually the worst value in the AMD line-up!

6272 is only $550 per CPU, which is insane value. Same 16 threads, base clock is 2.1 GHz, turbo is up to 3.0 GHz. Also consumes less than 115W.

Think about it - AMD is offering 32 cores for the same price as Intel's 6 cores.

[Fabian Benninghoven]

@ Sen:
You're right, 2x 6272 and the base clock maybe set up to 2,8 Ghz: only 2x 490 Euro - best price!


I called Adobe and asked if CS6 is using all threads from 32 cores. The answer from the support: you don't need more than 8 cores atm.
But the guy from the support will check this "Opteron issue" and will give me another call back later. I'll keep you posted.

[Petri Teittinen]

Think about it - AMD is offering 32 cores for the same price as Intel's 6 cores.
This certainly made me re-think my next PC hardware purchases; thanks!

BTW, is the jury still out on the new gaming-oriented Nvidia 680/690 cards and their performance in video applications?

[Fabian Benninghoven]

2nd call from Adobe: it's important to have more CPU power than CPU's in the system. The support max. recommended a 8-core system, with the highest Ghz available.
4GB Ram per Core -> 24GB.

Hmm...

[Olivier Madar]

2nd call from Adobe: it's important to have more CPU power than CPU's in the system. The support max. recommended a 8-core system, with the highest Ghz available.
4GB Ram per Core -> 24GB.

Hmm...

Very interesting thread !

I will follow the Clevo route with P170EM


Nvidia Geforce GTX675M *2Go
Intel® i7-3920QM 4x2 cores - 2.9GHz à 3.6GHz - 8 Mo cache - 55 Watts (Ivy Bridge)
DDR3 Sodimm 32Go 1333Mhz quadri-channel
ssd for system
2x750 esata raid 0 for media


Plus 3 x usb 3,1x esata


I would have wait for thunderbolt, but i am not sure the Rocket will be indispensable

Any suggestions from our Gurus ?

[Brian Merlen]

wow i seen a 64 core model on titanus! it can have 512gbs of ram too! talk about crazy specs

[Joel Arvidsson]

Did anyone try the amd with premiere yet?

[Jeff Kilgroe]

The Xeon E5's smoke these AMD systems in per core and per clock comparisons. The advantage of the AMD is price, in particular they can offer more cores for less money, but the cores aren't as good. And very few applications can scale well beyond 2, maybe 4 CPUs. Premiere does a pretty good job getting to 8 with MPE, but after 8 processors, the diminishing returns really become obvious. Anything more than about 12 threads seems to almost start slowing things down.

The monster AMD systems are good for render farms, massive database servers...

All the good dual Xeon E5 and larger motherboards / workstations will accept 512GB RAM too....

[Joel Arvidsson]

Jeff, you made me take another look at the xeon again. Was suppose to just order a 3930 system but from what I have read here it seams that alot of 3930 systems seams to struggle abit on the 4k but works ok in quad hd. I like to be able to run 4k at half res solid and at the same time output with a black magic card to hd/2k monitor.SLI

What do you think about this system below? It may not scream like some monster builds on reduser but for 4k I think this could be pretty solid. I have not found any with this processor here so I don't know what perfomance I should expect.
The system will be used for premiere and resolve. Fore storage I plan to run a windows 8 server connected with 4gigbit lan cables but thats another story.

-------------------------------------------------------------------------------------------------------------------------
ASUS Z9PE-D8 WS, 2xSocket-2011
SSI-EEB, C602, 8xDDR3, 7xG3-PCIe-x16, SLI/CFX, 2xIntel GbLAN, 2xFW, iKVM

(two of those below)
Intel® Xeon® Processor E5-1650
Socket-LGA2011, 6-Core, 3.2Ghz, 12MB, 130W, TRAY/without FAN

(two of those below)
Kingston DDR3 1333MHz 16GB ECC KIT
2x8GB 1333MHz DDR3 ECC CL9 DIMM

Fractal Design Define XL Black Pearl
Fan:1x 140mm Front, 1x 180mm Top, 1x 140mm Bakside, noise reduced, m/e-ATX, USB3

Intel® SSD 520 Series 240GB 2.5"
SATA 6 Gb/s (SATA3.0), 550MB/520MB/s read/write, bulk (OEM)

ZOTAC GeForce GTX 680 4GB PhysX CUDA
PCI-Express 3.0, GDDR5, DVI-D+DVI-I, native-HDMI, DisplayPort,

Noctua NH-U12DX CPU Cooler
Socket 771, Intel Xeon 5xxx, 7xxx,1300 RPM, 92,3 m³/h, 19,8 dBA, 120mm FAN

Cooler Master Silent Pro M2 1500W PSU
ATX 12V V2.3, 80 Plus Silver, Modular, 6x 6+2pin PCIe, 6x 8pin PCIe, 12x SATA
---------------------------------------------------------------------------------------------------------------

Also do you need sli to run on premiere and resolve? If you dont need sli I found a better workstation moderboard.
ASUS Z9PE-D16, 2xSocket-2011
E-ATX,C602-A,16xDDR3,4xPCIe(3.0)x16,3xIntel i350 GbLAN,1x Mgmt LAN, VGA, iKVM

Another confusion is that on resolves recomandation there is a
Intel Xeon E5-2620 6C/12T
2.00 GHz, TLC: 15 MB, Turbo: Yes, 7.2 GT/s, TX/RX300 S7, RX350 S7
It cost almost the same as the
Intel® Xeon® Processor E5-1650
Socket-LGA2011, 6-Core, 3.2Ghz, 12MB, 130W, TRAY/without FAN
Wich one is the better for premiere and resolve?


Hope there some helpful people on this thread. I really like to order this within a week or two.

Cheers!