AMD 2nd Gen Ryzen Threadripper 2990WX review
video description
Date: 2022-03-15
Related videos
Comments and reviews: 10
Edsel
I i think I know why task in some -multi threaded- application runs lower in threadripper. Some apps has certain number of worker tasks. They maybe multithreaded but it doesnt mean all workers or all threads are used to call those worker threads. The reason why 3d renderers are much better with thread ripper is cause it works per bucket, the square region that gets render this region has a good amount of pixels, so the more buckets the more pixels the more thread gets rendered. For 2D stuff like video encoding and decoding is a bit questionable. but you can use FFMPeg to benchmark instead of handbrake and adobe premiere try FFMPeg its tested.
reply
I i think I know why task in some -multi threaded- application runs lower in threadripper. Some apps has certain number of worker tasks. They maybe multithreaded but it doesnt mean all workers or all threads are used to call those worker threads. The reason why 3d renderers are much better with thread ripper is cause it works per bucket, the square region that gets render this region has a good amount of pixels, so the more buckets the more pixels the more thread gets rendered. For 2D stuff like video encoding and decoding is a bit questionable. but you can use FFMPeg to benchmark instead of handbrake and adobe premiere try FFMPeg its tested.
reply
Sam
I haven't seen you mentioning. This in any of your videos so far and it was a big deal and selling point for me with the 1950x: what you can do at the same time with very little penalty. Let's call it multitasking. The fact that you could play a game, live stream, record it, plus have video editing open in the background encoding and you didn't need to close chrome with 30 tabs open? All at the same time without skipping a beat? I mean you start talking about taking on insanely CPU intensive tasks - all at the same time without issue or slow down.
reply
I haven't seen you mentioning. This in any of your videos so far and it was a big deal and selling point for me with the 1950x: what you can do at the same time with very little penalty. Let's call it multitasking. The fact that you could play a game, live stream, record it, plus have video editing open in the background encoding and you didn't need to close chrome with 30 tabs open? All at the same time without skipping a beat? I mean you start talking about taking on insanely CPU intensive tasks - all at the same time without issue or slow down.
reply
65
Second Generation Threadripper will be a Virtual Machine monster. Instead of having eight different computers with eight threads and 16GB of RAM consuming ungodly amounts of electricity, I could have all that running on one PC and consume no more than 850 watts on a platinum rated power supply.
Side note: My current 8-core FX-8350 consumes 233 Watts idle and 365 Watts under full load. Overclocked to 4. 8Ghz, it consumes 350 Watts idle and 500 Watts under full load. Imagine having eight of these. 1800-2900+ Watts easily
reply
Second Generation Threadripper will be a Virtual Machine monster. Instead of having eight different computers with eight threads and 16GB of RAM consuming ungodly amounts of electricity, I could have all that running on one PC and consume no more than 850 watts on a platinum rated power supply.
Side note: My current 8-core FX-8350 consumes 233 Watts idle and 365 Watts under full load. Overclocked to 4. 8Ghz, it consumes 350 Watts idle and 500 Watts under full load. Imagine having eight of these. 1800-2900+ Watts easily
reply
Kyamil
Anandtech showed that it's not just core-core-core latency but also that infinity fabric is consuming a lot of power and leaves the CPU with a big chunk of power hunger meaning that IF has a diminishing return effect - it helps scaling, but also holds it back after a certain power threshold. That, coupled with the core latency and software not utilising all them threads leaves the part feeling a bit out of date but in the reverse meaning, haha.
reply
Anandtech showed that it's not just core-core-core latency but also that infinity fabric is consuming a lot of power and leaves the CPU with a big chunk of power hunger meaning that IF has a diminishing return effect - it helps scaling, but also holds it back after a certain power threshold. That, coupled with the core latency and software not utilising all them threads leaves the part feeling a bit out of date but in the reverse meaning, haha.
reply
Witold
hi, can we have some machine learning (ML) or data processing (DP) benchmarks? For ML you could try xgboost training speed or lightGBM training, and for DP some data manipulation times in SPARK or DASK. And there is also linux - in one benchmark I saw 7zip compression under linux to be 2x fast comparing to win10.
reply
hi, can we have some machine learning (ML) or data processing (DP) benchmarks? For ML you could try xgboost training speed or lightGBM training, and for DP some data manipulation times in SPARK or DASK. And there is also linux - in one benchmark I saw 7zip compression under linux to be 2x fast comparing to win10.
reply
Unreal
at around minute 20, do not doubt I actually think I will be rendering 2 jobs at once. It'll come to a point where saving the EXR images and the single core processes that precedes the cores kicking in will take longer than the actual render so it makes sense that we would double up on these workloads.
reply
at around minute 20, do not doubt I actually think I will be rendering 2 jobs at once. It'll come to a point where saving the EXR images and the single core processes that precedes the cores kicking in will take longer than the actual render so it makes sense that we would double up on these workloads.
reply
movax20h
Please do not report benchmarks results in terms of time to completion. Instead do benchmarks per minute. I. e. do not say 20 seconds to do a benchmark, instead do 3. 0 benchmarks per second. There is a huge difference. It also makes easier to interpret graphs and read all of them in the same way.
reply
Please do not report benchmarks results in terms of time to completion. Instead do benchmarks per minute. I. e. do not say 20 seconds to do a benchmark, instead do 3. 0 benchmarks per second. There is a huge difference. It also makes easier to interpret graphs and read all of them in the same way.
reply
IceColdKilax
This Video is Hands Down Proof of what everyone knows. It takes 32 AMD ThreadRipper 2 Cores to Barely, beat 18 7980XE Cores. These guys proved it. And JAYZ2Cents Proved 7980XE will hit 5Ghz. Overclocked
I-d bet money that an Overclocked Intel 7980XE will beat an Overclocked AMD 2990WX.
reply
This Video is Hands Down Proof of what everyone knows. It takes 32 AMD ThreadRipper 2 Cores to Barely, beat 18 7980XE Cores. These guys proved it. And JAYZ2Cents Proved 7980XE will hit 5Ghz. Overclocked
I-d bet money that an Overclocked Intel 7980XE will beat an Overclocked AMD 2990WX.
reply
Alf
Why are you using a so old version(0. 9. 9) of handbrake? Handbrake is now at 1. 1. 1. Maybe that will fix the problem.
And if its for comparison with the other CPUs sake, the newer tech has a disadvantage. Newest hardware equals newest software.
Again with the 7-zip from 2010. -PCWorld
reply
Why are you using a so old version(0. 9. 9) of handbrake? Handbrake is now at 1. 1. 1. Maybe that will fix the problem.
And if its for comparison with the other CPUs sake, the newer tech has a disadvantage. Newest hardware equals newest software.
Again with the 7-zip from 2010. -PCWorld
reply
Five
I'm a little surprised no one mentioned that it might be the Windows scheduler since we had a very similar issue when the Ryzen 1000 series launched. Linux' performance in some of the tests the 2990WX struggled in scaled properly. Never underestimate MS' incompetence.
reply
I'm a little surprised no one mentioned that it might be the Windows scheduler since we had a very similar issue when the Ryzen 1000 series launched. Linux' performance in some of the tests the 2990WX struggled in scaled properly. Never underestimate MS' incompetence.
reply
Add a review, comment
Other channel videos
