Also keep it cool, and dont start a flame war thanks!.
1 Like
No, Ryzen has no cold bug.
2 Likes
That’s not entirely true, you can’t run single die chips full pot, and there’s a boot bug below a certain threshold. Otherwise, though, it’s much more tolerant than recent intel chips.
Phoronix does a ton of benchmarks. AMD knew the 2990WX did well in Linux, and they needed those benchmarks to show the poor Windows performance is due to Windows and not the chip. If you look at the buzz on the internet, it worked.
1 Like
I agree. Phoronix are like machines when it comes to benchmarking. Cause they made it a machine.
Their the news its very active on the site as well.
Does anyone have any clue how 2990WX vs 2950X would perform in virtualization? I got really worried by poor AES256 performance with 2990WX as I use storage encryption :C
In general poor memory performance sounds quite bad. Especially nowadays when many workloads are memory bound. From what I saw 2990WX has effectively lower per-core memory bandwidth than even mainstream Ryzen…
Im a no body but. 2cent. Storage is way slower than ram and CPU’s have been compressing and encrypting drive data for while now faster than the raw speed to record the data to drives.
It will be fine.
Tech tubers have kinda of shit the bed. To put it bluntly. Windows cant handle the CPU and they have a small suite of tools and games. Kids love games.
Have a read of the AnandTech review. It even has the full Epyc 8 channel ram 128 PCI chip in the graphs but I believe the Epyc chip by default does not clock as high or use as much power.
https://www.anandtech.com/show/13124/the-amd-threadripper-2990wx-and-2950x-review
Lastly If you dont have a HEDT need and application you use. The 2950 is nice if you dont have a 1950 already.
If you needed 32c64t you would have a task in mind and could test it. On top of that windows is a terrible HEDT platform. If your a tech tuber your HEDT is adobe which make it even worse. Adobe made iGPU’s seem useful. You think there targeting 64 thread monsters when promoting iGPU’s ?
The new CPU has flaws and windows is the worst of them. AMD is breaking the rules. Its also breaking OS’s.
1 Like
^ as per above: My dual core i5-5287U in my macbook pro keeps up pretty well with the SSD in it for full disk encryption.
Would love to see a video showing how many VMs/containers you can run concurrently. Preferably also showing off k8s? Pretty Please?
Well, I mean it IS because of Window’s scheduler not being strong with NUMA type CPUs.
Really? I’d say no, given the minimal difference in 99% of cases between dual channel and quad on the intel HEDT platform for the majority of tasks.
I’d say tasks that scale to 64 threads are more likely processing bound. TR2 has some big local caches on die to help with the memory IO. Sure there are things that can trip it up, but you’re getting 32 cores for very very cheap (like, 15-20% the cost of 28 core intel at lower clock). Trade offs were made.
At the end of the day, as @Marten said, this CPU is showing up flaws in Windows, and the tech youtubers aren’t really throwing appropriate workloads at it. Or an appropriate OS.
No one gives a fuck (or rather no one should) how fast it runs games in artificially limited 1080p or whatever. That isn’t what it’s for. If you’re a gamer you buy 2700X or 8700K. The “Bad” gaming performance on this chip is still very much “good enough” as a secondary usage. Which is all gaming on this would be. Or you’re nuts buying it.
The point is heavy multitasking / VMs / rendering / other massively CPU heavy workloads.
That said; if you’re looking to buy 32 core TR, and THINK you may be memory bound, you should also consider 32 core (or maybe even 24 core) EPYC. Which isn’t constrained in that way.
If you’re memory or IO bound, AMD have a solution for that. If you’re not, and there’s plenty of stuff for desktop that is not, then TR2 will fill a niche.
Speaking as an engineer, a developer, and as a consumer I have mixed feelings about 2990wx.
As a consumer, it’s double the price of 2950x, but it doesn’t make compute (CPU+memory bound) workloads double the speed / half the latency in most cases, it costs an extra 1000 and would only be useful in a small number of cases.
As an engineer, it’s double the price and even though you’re paying a premium to AMD considering it doesn’t cost twice as much to manufacture, if your 2950x system costs 5000 and your 2990wx system 6000, you’re paying only 20% more for 30-50% more performance in CPU + memory bound workloads – great! .
As a developer, just give me whatever kind of cores and I’ll make use of them and don’t bother me or distract me with this real-world hardware stuff again until there’s something at least half an order of magnitude faster.
That said I’ve been in the market for a new workstation for a while… and we have a 2990wx coming out at the same time as Nvidia 2080ti for some light gaming on the side and a new version of enthoo evolv x, that should save my desk from secondary system clutter. RAM is about the only thing messing things up for me slightly, but I can live with that. Upgrading the gas boiler at my house to a tankless one can wait until spring.
3 Likes
Compute performance never scales linearly with dollars… you always pay more as you climb the product stack.
I think irrespective of the cost vs. 2950x, you need to compare to the alternative CPU you would have to purchase for this many cores.
the intel desktop variant doesn’t exist, but their 28 core xeon server CPU on their extreme core count server socket is over $10,000 USD in quantities of 1000. and has lower clocks.
sure, you could maybe go for 2x 16 core xeons in a server board, but then you’re dealing with the issues associated with mulitple sockets, still way over budget, etc.
the short version is that you’re essentially paying a ton more money to do free QA for AMD
also:
…hwut
Holy shit that is quick!
Still it’s not better in “absolute” way over 2950x. Memory bandwidth issues are concerning. This is imho problem with 2970wx and 2990wx. Not price, but that you need to be 100% sure that your workload won’t be affected by memory bottlenecks.
I’ve been testing a 2990wx for like two days. Really impressed with it. At 3200 the memory bottleneck is less, but having trouble replicating phoronix results because it shows up quite good in benches there. I’m on fedora 28 but some benches like jtr won’t work on fedora …complains about missing libssl even though I have it. Probably a bad path.
I gotta open up the power and thermal limits to 450w to even get close to their results.
3 Likes
I think arguably more representative test would be with ECC 2666. Especially for loads like virtualization and other pro-grade workloads. Many people looking for those beasts will pair it with ECC. I’m still puzzled between 2950x and 2990wx for KVM host (using ECC).
I have 128gb of the stuff right now . Fastest unreg ecc you can get. Tho I can manually oc the Samsung 2400 that uses the faster ICs.
Phoronix used 3200 tho so…
1 Like
Is that the Crucial 2666 ECC that you have?