AMD Threadripper 2990WX 32-Core and 2950X 16-Core Review

Post by compilerdev » Mon, Aug 13, 2018, 11:16 PM UTC

Hasknewbie wrote: post #41

MrUnderhill wrote: post #20

Earlier quoted context omitted.

For compiling, having many cores is fantastic. Granted, on a workstation, compilation normally just involves a few files (the ones that have changed since the previous build and their dependencies), but when you have to do a full rebuild, it is fantastic to be able to do `make -j16` and watch it chug through 16 files simultaneously. Interestingly, the benchmark in this review shows that the 16-core 2950X compiles Chr…

"this review shows that the 16-core 2950X compiles Chromium faster than the 32-core 2990WX, presumably this means something other than the thread count becomes a bottleneck after 16 threads" The article mentions that, due to the die packaging, only 16 of the cores have direct access to RAM. So for the 32-core version, half the cores are memory-starved and have to go through the 'connected' cores (also impacting these…

Post by Filligree » Mon, Aug 13, 2018, 11:53 PM UTC

vbezhenar wrote: post #167

Osiris wrote: post #93

Earlier quoted context omitted.

There are 2 dies in the 1950X, each one has 2 memory channels. Thus, it's possible to run a process on one (8-core) die that maxes out the memory bandwidth to it's two local DDR4 channels while the other die still has full bandwidth access to it's own DDR4 channels. Threadripper is able to switch between NUMA (non-uniform memory access) mode and "regular" mode. In NUMA, the OS knows that 2 channels are attached to 1…

Does it really work this way (with automatic memory and core pinning)? Both Windows and Linux can do that?

Post by glandium » Tue, Aug 14, 2018, 12:08 AM UTC

compilerdev wrote: post #170

glandium wrote: post #163

Earlier quoted context omitted.

There's a significant workload difference between compiling the Linux kernel (C code base) and Chromium (C++ code base). I'd very much like to see a Chromium build benchmark performed under Linux.

See my answer above, it's because the build was LTCG and in the best case at most 4 cores were used during optimization/code geneation/linking.

Post by hajile » Tue, Aug 14, 2018, 12:22 AM UTC

greglindahl wrote: post #140

somebodythere wrote: post #103

Earlier quoted context omitted.

If AMD's method scales so much easier, why didn't Intel just... do that? Honest question.

Intel has done that in the past, actually, their first "dual core" chip (2005) was actually two chips in a package. https://en.wikipedia.org/wiki/Pentium_D

Post by hyperpallium » Tue, Aug 14, 2018, 1:07 AM UTC

jandrese wrote: post #155

mjevans wrote: post #147

Earlier quoted context omitted.

That's still pretty much the case today. Though the underlying java GL library the desktop client uses has been updated a bit and the newer versions have migrated to higher versions of OpenGL; breaking a lot of mods pretty hard with the changes to the rendering engine. I hear the (Windows/console only?) version is faster for all the reasons you'd expect, but I'll probably end up never playing it.

Windows version has way better performance, but it doesn't support mods so nobody cares about it.