AMD threadripper 2990wx for phoenix FD

[fraggle]

First of all, sorry but I can't post on the phoenix forum as I used a trial license!
I'm about top purchased a new workstation and I'm wondering if the new AMD 2990wx make sens or not for phoenix.
Or should I buy a second one, with an old extreme CPU or OC cpu, single core as it seams that multiple core is not supported by Phoenix.

[georgi.zhekov]

Hello,

You can check our hardware recommendations for Phoenix FD here - https://docs.chaosgroup.com/display/...sterSimulating

Also it's important to note that Phoenix does support multicore machines.

[Paul Oblomov]

I have more digits for ya:

2699v4@2.4Ghz
Total elapsed
11m 15s

2990wx
Total elapsed
14m 31s

I think, i'm gonna wait for an intel 9900xe, tho maybe there will be the same mesh problems. Who knows...

[fraggle]

thanks, that's very useful.
That's why we all would like to get a Phoenix benchmark.
All the info grab on Chaos hardware advice will make the AMD lotr faster: faster BUS, Faster Ghz, more core ... but it is still slower!

[Paul Oblomov]

Maybe epyc will be better.

Chaos advice was pre numa like cpus came to mass market.

P.S. I think, we should try cutting half of the cores from the 2990 and retest. Looks like TR memory interface is a bottleneck, coz only half of the cores are directly connected to ram.

[fraggle]

P.S. I think, we should try cutting half of the cores from the 2990 and retest. Looks like TR memory interface is a bottleneck, coz only half of the cores are directly connected to ram.

Please let us know if it change the result.

[Paul Oblomov]

Oh no. I don't have it.
https://forums.chaosgroup.com/forum/...hardware-bench

[jacksc02]

Reducing the number of cores to 16 on the 2990wx does the burnout sim in 16mins 38secs
This sim uses 15gb of ram and from what I have seen so far is that only really large scenes (100gb+) seem to show a slow down in memory speed.

[fraggle]

Reducing the number of cores to 16 on the 2990wx does the burnout sim in 16mins 38secs
This sim uses 15gb of ram and from what I have seen so far is that only really large scenes (100gb+) seem to show a slow down in memory speed.

Sorry, but 16 min with 16 core vs how many with 32 core?

[jacksc02]

14mins 31secs with 32cores

[fraggle]

14mins 31secs with 32cores

Thanks,
So, I'm lost then. I was thinking that the AMD problem was NUMA and that's why it performed poorly compare to Older Intel CPU that are not NUMA. So when disabling half the core, I believe the it is no more a NUMA. I guess I don't understand what's NUMA.