Sooo, just how fast is 128 cores when rendering? EPYC 7742 vs Intel 8280
tj_1ca9500b
Posts: 2,057
https://www.phoronix.com/scan.php?page=article&item=blender-epyc-7742&num=1
The article compares 2x EPYC 7601 32 cores (64 total) vs 2xXeon 8280 28 corss (56 cores total) vs 2xEPYC Rome 7742 64 cores (128 cores total).
Notably, the dual 7742's suggested retail at Newegg is about the same price as a sinlge 8280 at CDW... ($14,903 for the 8280 at CDW vs $7,219.99 each for the 7742s at Newegg). CDW is probably waaay overpriced though, I've seen mentions of 8280's being priced considerably lower in various reviews. Intel REALLY needs to lower their prices now if they want to remain competitive though...
Most of us can't afford those prices in the first place, but it's fun to look at the benchmarks anyways, to see what's possible. It does make one wonder just how close to 2080 Ti benchmarks the 128 cores are these days though...
Here's a few 2080 Ti Blender benchmarks that I managed to track down. I KNOW I saw others somwerhere, I just have to find them...
http://boostclock.com/show/000242/gpu-rendering-nv-blender-tile-2019-01-01.html
If others know of other links with useful 2080 Ti or RTX Titan Blender benchmarks, for comparison purposes of course, feel free to share!
Of course, the other question would be, could Daz Studio even handle 256 threads?
Daz 3D is part of 
Connect
DAZ Productions, Inc.
7533 S Center View Ct #4664
West Jordan, UT 84084
Licensing Agreement | Terms of Service | Privacy Policy | EULA
© 2025 Daz Productions Inc. All Rights Reserved.
Comments
31,000+ in cinebench
cinebench is actually rubbish for benchmarking, just like 3dmark, compared to a 3d fractal generator http://www.mandelbulb.com/2014/mandelbulb-3d-mb3d-fractal-rendering-software/ for CPU only,
rendering a fractal will probably only work if you can run prime95 for atleast an hour without any issue;'s because of lack of volt's to core's and RAM etc https://www.mersenne.org/download/
..have seen a couple dual epyc workstations available, pretty expensive.
iray rendering should be extremely embarrassingly parallel, from what I understand, it scales perfectly with additional videocards.
my CPU is incredibly old (it's a 5930k on the x99 platform from 2014) and has only 6 physical cores (12 threads), my RTX2080ti is about 20x faster for iray rendering. So this would lead me to guess that 120 CPU cores would compare favourably with a single RTX2080ti GPU running on it's own.
Unfortunately, things are not that simple. There is a possibility that (even with 8 memory channels per processor, not kidding) there might be bandwidth issues. AFAIK iray does not support texture compression when using CPU emulation and GDDR6x is faster then even 16 channels of DDR4. AFAIK GDDR6x on RTX2080ti is 616 GB/s and 16 channels of DDR4 would be something like 240 GB/s. This is only important if memory bandwidth is an issue, which I have no idea about.
On the plus side, Intel and AMD have improved their floating point performance a lot since my CPU was produced, I believe they have both more then doubled the FP performance per CPU cycle. So it might be that only 60 CPU are needed to beat an RTX2080ti.
Would be interesting challenge, but I doubt it would be good value for money. Iray is produced by NVIDIA using CUDA and would be heavily optimised for a GPU.
..well for about 23,000$ you can get a dual 64 core Epyc workstation with 128 GB of memory (actually priced one out). Add another 7,000$ for dual Titan RTX GPUs with NVLink.
Apparently NVLink memory pooling doesn't work for iray in DAZ Studio. Some people were talking about it in a thread a while back, I think they said it is working in 3dMax or maybe it was something else.
Personally (if I won the lottery) I would just get 2 RTX Titans or even 4 of them. With $23k you could even consider Quadro RTX, the Quadro RTX8000 comes with 48GB of memory and it looks like they have come down in price according to https://techgage.com/news/nvidia-quadro-rtx-6000-rtx-8000-price-drops/ you can get 2 with a bridge (96GB total memory for apps that support GPU memory pooling) for $13k.
For an absolute top of the line CPU server chassis that's peanuts. Dual 8280 racks run over $50k. Note that 128Gb of RAM is on the low end for a dual CPU server these days. The 7742 has 8 memory channels per chip. With 2 sockets that means 16 channels which means a minimum of 16 and more likely 32 DIMM slots. 128/32 = 4Gb RAM sticks. That's pretty minimal for such a beast. I'd configure any that I build for 512Gb minimum.
..23,000$ for the workstation I priced out is about twice as much as I receive a year on Social Security. I wouldn't call that "peanuts".
For a drag-racer, it would be peanuts. They'll spend that on tires alone. For an over-the-road trucker, that would be a down payment.
My point is that it is all about context. As it always is. I am grateful for those who are the pioneers in this industry; the ones willing to spend a king's ransom to get their work done. Or even if it's just for bragging rights. Regardless of the motivation, it makes innovation possible. And THAT makes affordable hardware and software available for you and me a few years down the road.
My laptop's specs wouldn't have been possible at twice the price-point even 5 years ago. And for many folks, would not have been imaginable. But here we are, and this is a fabulous time to be alive.
I feel called out...
...I'm still working on 7 and 5 year old tech (LGA 1366 and 1150) CPUs with DDR3 memory and Maxwell GPUs that are 2 generations old. This is the best I can afford at the time.
Just came out of a meeting with a SuperMicro rep. They are offering a dual socket EATX server board with 2 7702 for $15k. In a decent chassis plus RAM and drives you're probably looking at something less than $25k.
That's amazing.
Gotta love the TCO numbers for EPYC Rome!
I'm not up on my Rome product stack yet, so I looked up the 7702. It's the slightly slower 64 core chip, with the 7742.being the top of the stack. It also has a lower TDP, which can be nice depending on your cooling and power requirements.
As others have pointed out, things are looking very interesting for 7nm Threadripper based on the EPYC Rome numbers.
In other news, dual 7742 EPYCs may have broken cinebench r20...
https://www.servethehome.com/crushing-cinebench-v5-amd-epyc-7742-world-record-edition/
I wish my Iray renders would complete in less than 8 seconds... Yeah it's an apples to potatoes comparison, but we all like faster render speeds. Of course, affordability is also a factor!
That benchmark is too used but it is almost directly relevant to things like Blender. The only serious question is, are the render engines designed to use all those threads. We've seen already from TR that there is a lot of SW out there that while multithreaded fail to use that massive number of threads.
...Carrara can use up to 100 CPU threads so one Epyc 7742 is more than enough..