The Ryzen Threadripper smashed the market with its combination of high cores and low pricing, and now the 2nd Generation is here including the 32 Core 2990WX. We test them out.
AMD Threadripper 2950X and 2990WX Review
Last edited:
AMD Threadripper 2950X and 2990WX Review
- Thread starter tinytomlogan
- Start date
The Ryzen Threadripper smashed the market with its combination of high cores and low pricing, and now the 2nd Generation is here including the 32 Core 2990WX. We test them out.
AMD Threadripper 2950X and 2990WX Review
Dark NighT
New member
What a beast of a cpu that 2990WX.
i actually have some issues with the video.
i did not read the article on the website (no time yet).
so excuse me when you adressed it in the written article.
one thing you forgot to mention in you TR2 video is that stuff like ZIP compression, AES encoding etc. (stuff that has a heavy memory throughput) is actually a lot slower on the 32 core than on the 16 core.
it is not just some "fancy" server stuff that sees a hit beause of the infinity fabric memory architecture on the 32 core CPU.
i watched the video and you made it sound as if only some very limited software will see a performance hit because of the infinity fabric memory architecture on the 32 core CPU.
handbrake is faster on a 16 core TR1 than on a 32 core TR2.
it´s also not a tiny tiny hit... sometimes, depending on the app, it is a pretty big hit.
i have seen the 1950x/2950x to be 10-20% faster than a 2990wx in some apps.
you really should expand your benchmark suite.
i don´t say that to badmouth TR2. i have two TR systems and will sure buy the 32 CPU model.
but people need to know where the 32 core shines and where not.
there are many cases where the 16 core is actually a faster CPU than the 32 core.
i do 3d rendering and programs like vray can use all 32 cores. also memory is not that big an issue.
so i can understand that people like me are as excited as TTL in the video.
but for a lot of people buying a 32 core instead of the 16 core TR2... could be a dissapointment.
there is always the warning "threadripper 2990x is not a gaming CPU".... yeah well it is not exactly the best threadripper for every workstation use either.
i did not read the article on the website (no time yet).
so excuse me when you adressed it in the written article.
one thing you forgot to mention in you TR2 video is that stuff like ZIP compression, AES encoding etc. (stuff that has a heavy memory throughput) is actually a lot slower on the 32 core than on the 16 core.
it is not just some "fancy" server stuff that sees a hit beause of the infinity fabric memory architecture on the 32 core CPU.
i watched the video and you made it sound as if only some very limited software will see a performance hit because of the infinity fabric memory architecture on the 32 core CPU.
handbrake is faster on a 16 core TR1 than on a 32 core TR2.
it´s also not a tiny tiny hit... sometimes, depending on the app, it is a pretty big hit.
i have seen the 1950x/2950x to be 10-20% faster than a 2990wx in some apps.
you really should expand your benchmark suite.
i don´t say that to badmouth TR2. i have two TR systems and will sure buy the 32 CPU model.
but people need to know where the 32 core shines and where not.
there are many cases where the 16 core is actually a faster CPU than the 32 core.
i do 3d rendering and programs like vray can use all 32 cores. also memory is not that big an issue.
so i can understand that people like me are as excited as TTL in the video.
but for a lot of people buying a 32 core instead of the 16 core TR2... could be a dissapointment.
there is always the warning "threadripper 2990x is not a gaming CPU".... yeah well it is not exactly the best threadripper for every workstation use either.
Last edited:
AlienALX
Well-known member
You will probably find that all of the issues (IE where 32 is slower than 16) can be easily fixed with updates and patches.
This always happens, and people don't usually revisit products to review them once they are all fixed.
not when it is an issue of the infinity fabric memory architecture.
the 2950x has only one link... the 2990x has six.
normaly i would (naively) think that when a 32 core uses only 4 cores and clocks at 3.8 GHz it is as fast as a 16 core at 3.8 GHz using 4 cores (same CPU generation of course). at least roughly the same performance.
that the 2990wx is sometimes a lot slower running at the same clocks as it´s 2950x brother is counterintuitiv. but it makes more sense when you look at how memory is handled on the 32 core chips.
ok 4 cores are maybe a bad example as they could all run on the same CCX with direct memory access. but you get what i mean.
it could be some overhead, maybe they could tweak memory handling of apps a bit.
but i doubt we will see a big improvement.
in fact for some apps i doubt we will see them get on par.... not to mention the 32 core pulling ahead.
there might be apps that can be optimized to make use of more cores, no question.
handbrake currently does not scale well past 6 cores.
but handbrakes performance normaly does not get worse with more cores (same clockspeed).
and parallelization has it´s limits. it does not work well for all cases.
it seems the infinit fabric implementation on the 32 core does have a great impact on performance with some memory sensitiv software.
i also wonder how this will work out when doing heavy multitasking and multiple memory intensive programs.
Last edited:
NeverBackDown
AMD Enthusiast
But can it run crysis?! 
mihapiha
New member
I have been waiting for this optimization for such a long time. I've been eying the TR4 motherboards but the lower CPU frequency was making me hesitant. With the new Precision Boost 2 manual overclocking hardly seems worth the effort.
I still have a 4930k and it is quite old now. Because I use 4k it luckily hasn't been effected that much. So I really love the results I see here. Just a little longer for the the Threadripper 1920X's replacement, the 2920x to come out.
I will move over to AMD for the first time since the release of the Core2Duo. These results are just spectacular.
I would have loved for TTL to do at least one test or two with SMT turned off, especially in gaming. Hardly any game will take advantage of this many threads....
I still have a 4930k and it is quite old now. Because I use 4k it luckily hasn't been effected that much. So I really love the results I see here. Just a little longer for the the Threadripper 1920X's replacement, the 2920x to come out.
I will move over to AMD for the first time since the release of the Core2Duo. These results are just spectacular.
I would have loved for TTL to do at least one test or two with SMT turned off, especially in gaming. Hardly any game will take advantage of this many threads....
NeverBackDown
AMD Enthusiast
Manual OC to sustain the higher clocks on all cores is really all it will do for yeah. Worth doing for 32 cores unless you are worried about efficiency.
mihapiha
New member
Well, my point was kind of different. Because I'm looking to get the 2920X (12-core) I was more focused on the results of the 2950X than the 32 core monster.
And it was quite spectacular to see the 2950X stock outperform a 8700k with an OC in a sintetic benchmark like 3DMark FireStrike Ultra. Which is basically the result I was curious about most, because I game at 4k.
Would it be on par with Intel at gaming with the Precision Boost 2? And apparently it is. Because it overclocks single cores higher than what you'd manually achieve for all cores, it performs spectacularly well considering. That's why it outperformed most manually overclocked results in the gaming benchmarks and in the single threaded CineBench run. Since 12 Cores is plenty for me, I am primarily considering getting the 12 core one and even turning SMT off. I just know that now with my 4930k there's quite a performance loss if HT is disabled, even though I have it overclocked to 4.4 GHz 24/7.
Basically AMD found a nice balance with this between both worlds. If you want a single threaded application speed, it does provide it, while not sacrificing that much multi-threaded either.
It's like nVidia Boost 3.0 (or whatever the call it now). Basically making manual overclocking obsolete for 24/7 usage.
Last edited:
a german websites reports that there are issues with TR 2990wx and nvidia cards.
they replaced a nvidia 1080 TI with a radeon vega 64 and the framerate increased a lot.
for one title it nearly doubled.
https://www.golem.de/news/32-kern-cp...08-136016.html
they replaced a nvidia 1080 TI with a radeon vega 64 and the framerate increased a lot.
for one title it nearly doubled.
https://www.golem.de/news/32-kern-cp...08-136016.html
Similar threads
