Ryzen 9 3900X vs. Ryzen 9 3950X vs. Core i9 9900KS In Nearly 150 Benchmarks

[vegabook]

I would like AMD to delay Zen3 for their own good. Just reap every dollar they can from Zen2, so they have the cash to fight intel/nvidia when the time comes.

No. Please No. This is how Intel thinks and is why we were stuck with 4 cores forever. They need to think more like Andy Grove: constant Paranoia.

[Rymgas]

Since Ryzen has separated float and ALU pipelines would be amazing to see one more test. Run all test at same time and see how both cpu would handle that. Because I would expect ryzen to be significantly faster here.

[creative]

No. Please No. This is how Intel thinks and is why we were stuck with 4 cores forever. They need to think more like Andy Grove: constant Paranoia.

Bleh.... Come up with something new. This is like celebrity worship except it's brand worship. Oh and intel has been make more than 4 core cpu's for years, AMD actually lied "with the FX garbage" that they actually had more than they had, their magical non existent cores/wannabe hyperthreads. Both companies take their sweet time getting demands answered.

95% of people and companies are lazy liars.

[duby229]

Bleh.... Come up with something new. This is like celebrity worship except it's brand worship. Oh and intel has been make more than 4 core cpu's for years, its just that AMD lied that they actually had more than they had, their magical non existent cores/wannabe hyperthreads. Both companies take their sweet time getting demands answered.

95% of people and companies are lazy liars.

you do kinda have a good point in regards to AMD's CMT architectures. there's no clear distinguishing line for them because they are different, but if you consider a core to be everything it takes to be a complete processor then what AMD called a "module" was actually the core and what they called the "core" was actually just one of two integer pipelines and didn't even consider the floating point pipeline or the cache pipeline.

now with that said, those integer pipelines only had two integer units each, if AMD had put three or even four integer units each l, then it would annihilated everything else ever conceived. not even Zen would stand a chance against a scaled up CMT architecture on a modern fab node.

[vegabook]

Bleh.... Come up with something new. This is like celebrity worship except it's brand worship. Oh and intel has been make more than 4 core cpu's for years, AMD actually lied "with the FX garbage" that they actually had more than they had, their magical non existent cores/wannabe hyperthreads. Both companies take their sweet time getting demands answered.

95% of people and companies are lazy liars.

But I don't actually celebrate AMD uncritically. And while I know it's unfashionable to pitch free-market cliches, there is just no doubt that by trying very hard, they've become a competitive force which is disrupting CPUs for the first time in years. And that's only because they're *not* thinking like a marketing MBA, but like engineers. They're product driven. The OP was talking about market arbitrage - playing game theory. I propose that AMD should simply try as hard as it can, all the time. to make better products. Just like Intel should have. Mine is a critique of the marketing people who have come to dominate engineering firms (see Boeing).

As for being critical of AMD, there is quite a lot going for Intel's AVX512 instruction set which they are promising will extend to provide a seamless coding abstraction into GPU-space. This is massive and something AMD should have been doing ages ago given they've owned ATI for 14 years. I'm actually very bullish on Intel's architecture/software roadmap. Coding for AVX512, is very attractive from the latency perspective. You can actually do some pretty mean computation without having to worry about having a big enough batch to overcome the latency penalty of GPU. If Intel manages to extend that to bigger batch being moved to GPU, this is a fantastic advantage. GPU-space has been far too disjointed for a long time and I cannot understand how AMD screwed this up having had a decade with x86 and top-flight GPU in house. ROCm certainly doesn't cut it.

[dad_ph]

And ... still my core i7 3960X @ 4 GHz (not AMD 3960X - only the 3960 index is the same ) looks not so bad in 2020: I run almost all of these 150 benchmarks (ubuntu 20.04) and 9900KS has only ~2X(200%) geometric mean. This is 5 GHz and 8-core 2019 i9 against 4 GHz 6-core 8(!) year old HEDT CPU: not very impressive win. R9 3950X is almost 3X geomean which is much better, but has 2.5X more cores than i7 3960X. I'd expect 5-10X average improvement in 8 years, not 2-3X. And 2-4X improvement per core, not 1.5-1.8X.

[dad_ph]

Interesting that for code compilation there doesn't appear to be any reason to go for the more expensive 3950X vs 3900X

Thoughts?

Compilation performance very heavily depends on programming language and compiler used(amount/layout of code being compiled also matters). Classic C/C++ and gcc scale quite well, but there are other PL/compilers which gain almost nothing from additional cores For example scala/dotty (PTS includes one benchmark for it) are single-threaded for almost all compiling phases except some trivial steps(code generation). So almost no profit from 16 core CPU compared to 4-8 core...

[nonrumor]

Although the benchmarks are very good maybe you should have included the standard i9-9900K for the test? Since the differences between the KS and the standard K are so marginal when it comes to performance but not to price

[DanielG]

Interesting that for code compilation there doesn't appear to be any reason to go for the more expensive 3950X vs 3900X: the best improvement is for the Linux kernel compilation and it's just 12%. The rest is in the 5-7% area. I have two theories as to why this might be the case.

The first is that the limited memory bandwidth starts to bite. That would also explain why the Linux kernel sees the best scalability (which is generally seems to be the case): lots of smallish C-files that don't take a lot of RAM to compile (unlike, say, large C++ files). But then things don't appear to be scaling well when we go from 3950X to the 24-core Threadripper (which has twice the memory channels). For example, for GCC compilation, going from 12 cores to 16 we see 7% speedup while going from 16 cores to 24 (plus the two extra memory channels), we see 9% speedup (based on these results: https://openbenchmarking.org/result/...AS-COMPILING50). One thing that we are not taking into account in this second comparison is the RAM speed, though.

The second theory is that for compiling real projects, single-core performance also matters a lot because of the serial linking steps. Again, the Linux kernel is the outlier here since there is only one linking step (I believe this benchmark does not build the modules, but even if it did, those are also quite parallelizable). Compare this to GCC which goes through quite a few linking "bottlenecks". And in this regard (single core turbo), the three processors are essentially the same.

Thoughts?

It might not even be memory bandwidth but simply the amount of memory - who would build a rig with 12 or 16 cores and put only 16GB in it?!
On the 3950X that's only 500MB per thread (=> per compiler process)!
The results with 64 or at least 32 GB of RAM would be very interesting.

[Pariston]

@Michael;
Are you going to do a Cascade Lake-X vs 3950X comparison?