Okay finally time to upgrade my workstation. I was waiting and waiting for 7nm AMD and here it is. Now the problem which one to go with.
For me it’s not only rendering but lot of modeling and multitasking (rendering, zbrush sculpting at the same time etc.) Also cloth and particle simulation. SO I need one powerful workstation that can do it all 
Is it worth getting the 32 core beast 3970x or a bit overkill and go with 24 cores as it’s 100mhz (3.8Ghz) base clock faster which is important in 3dsmax as it’s one core app.
I’m not sure about Ryzen 3950x as I read that it will require water cooling and because if Threadripper size I assume air is still good and quieter option.
Whats your take an all this?
I vote for 24cores. Already bought 256gigs
any reason why 24 and not 32?
Cheaper. Faster one threaded
It`s not going to be faster one threaded. Both can go up to 4.5Ghz. In single threaded app, cpu will be looking for best core to run the thread on. And it`s not going to be at 3,7 VS 3,8, but much higher, possibly near to 4.5ghz…
Then, cheaper
i am very interested to see how the 8 channel chips compare.. nothing is certain at the moment, but there is definitely a chipset for threadripper coming with 8 memory channels (wrx80) and presumably chips specifically for that chipset. i assume (and i could be wrong) that the 48 and 64 core threadrippers (due in jan next year or so) will be 8 channel, since 4 channels is already rather feeble for a 32 core chip.. makes sense that the newer higher core models will have to address this.
zen2 seems to work very well given the limited bandwidth (they have tons of cache) however there are limits, and more is better.. so maybe the 8 channels will perform noticeably better than simply having more cores. of course, clock speeds might be much lower…
anyone from Chaos care to give an opinion on zen2 vray performance? is it limited by memory bandwith significantly on the higher core models? or does it generally play well with the large caches available?
Ill wait for 3990 coming in January
+1 for waiting for 64 cores.
Yes, at least in Vray.
I would not invest into 16 core nowadays
I am interested in your choice,are there any better options for you to build a powerful workstation?
It will be really appreciate if you can share your experience.
32 threads just dont cut it for me. I am not surprised intel beats amd on that front plus they have a lot better memory handling. Iam reliant on cpu because our projects use a lot of memory. Therefore i go for highest thread count machines with good price per core ratio. TRs are great, but the lower end ones are easily replaced by intels. If you have good enough budget, go for 32 core at least 
i am very pleasantly surprised by the massive improvement in vray over the 2990.. makes me hesitate about waiting for the 8ch ones.
for me, i suspect the 48 core might be the sweet spot.. double the memory bandwith than the 32 core, 50% more cores (more bandwith per core) and hopefully more reasonable price/performance and higher clockspeeds than the 64 core.
trouble is, by the time *those* come out, we are already going to have more concrete information about zen3, which is also due out next year and promises even bigger IPC and clockspeed improvements than zen2…
i might just buy some AMD stock.
@super_gnu yeah, i wish they released whole lineup same time, would have been easier to make a decision. I guess they just postponed it so they can hold on to the 8core chiplets for as long as possible - to sell it in EPYC market. Selling those in HEDT, less profit
I was also expecting 48core to be sweet spot, but after yesterday i think 32 core is the sweetspot for me. It`s a solid upgrade over my 2990WX. Specially single thread performance was a huge surprise.
48/64 core will bring significantly lower clock speeds , 64 probably as low as 3Ghz per core and boost to 3.4Ghz. That would be my guess, based on the 280W TDP they want to keep. Such a frequencies will be sub-optimal for desktop/single thread, daily work. And from $ per thread efficiency it looks questionable too, they going to ask 4k for it maybe?
They were positive on the future compatibility of TR40/80/W80 platform,so probably next year zen3 will work on current boards, thats good news.
From that benchmark seems that 3970 is almost twice as fast as the 2990(same number of cores), an ~85% jump in IPC seems unrealistic and not in line with every other benchmark(less than 20% on average), experience tells me that when something is too good to be true, it is usually false(or at least represents a very specific scenario). Also core for core Intel had a slight advantage over AMD but looks weird to see the 10980 being significantly faster than the 2990 since is a 18core VS a 32core. I’m not sure what is going on but a common problem with all recent rendering benchmarks is that both CPU and render engines have become extremely fast while scenes remained fairly simple.
The new TR will be great for rendering(not that much though) considering the price, for editing I’m not so sure considering the previous generation performance.
There are plenty of data around to show how much better TR3 is over 2nd gen, specially over 2990WX, which is a seriosly gimped architecture(specialy on windows),that is choking on complex scenes with limited bandwidth. I`m using it daily over a year, on both Linux and Windows so this is first hand experience.
“looks weird to see the 10980 being significantly faster than the 2990 since is a 18core VS a 32core.” - nothing weird here. It`s not only about core count…
If you want to check some non mainstream, non youtubers , but solid reviews check Phoronoix https://www.phoronix.com/scan.php?pa…0x-3970x&num=7 , the geometric mean of 111 benchmarks comes to 57% speedup. Sound good enough to me!
Those benchmarks shows exactly what I’m saying, every modern renderer should maximize core usage so results should be mostly consistent across different engine(of course some renderer may be more fine tuned than other so slightly difference should be expected). I mean.. just looking at that results, there are benchmarks where the 10980 is faster than the 3970, other where is faster than the 2990, other where is slower than both, pretty much unconclusive to me;)
I’m sure that 3970 will be a great choice for render tasks and will be significantly faster than the 2990(and of course Intel offer), but twice as fast in most scene? likely not, I’ll believe that when I’ll see a complex render made in 1h instead of 2h using bucket sampler.
BTW, I never look at youtube;) usually look at Phoronix and Servethehome reviews, that being said I take all of that with a large grain of salt since those reviews are made by people with zero knowledge in 3D and benchmarks maybe not representative of real world usage. I always prefer firts hand experience(hopefully unbiased) from people who actually use the hardware in a 3D pipeline, especially if they are doing scenes similar to mine.
Im not sure we are looking at the same benchmark page there? To me, that says (as concluded by the review too) that the 3970 rocks. Averaging 57% faster than its predecessor for the same core count. not really inconclusive at all. A few outliers where (probably) either the compiler was extremely intel preferential, or the renderer made heavy use of avx512 which only the intel chip has.
i agree that real-world testing is far preferential to benchmarking, and i certainly wouldnt be buying one without seeing feedback from real-world users. however, given the architecture is a totally known quantity now, due to the previous ryzen 3xxx, the clock speeds are now known, its not exactly a stretch to imagine how good it will be real-world. The only unknown really is memory bandwidth and latency issues, which is what caused all the 2nd gen 32 core problems. Results in all reviews so far seem to suggest that this issue has been put to bed.
vray was always a bit sensitive to NUMA issues, which means the 81% speedup may not be too far from the truth.