Many More AMD Ryzen 5000 Series "Zen 3" Linux Benchmarks

[phoronix]

Phoronix: Many More AMD Ryzen 5000 Series "Zen 3" Linux Benchmarks

Over the past week we have published our Linux performance reviews of the Ryzen 5 5600X, Ryzen 7 5800X, and Ryzen 9 5900X + 5950X "Zen 3" processors. For as much data that has been available in those reviews, here is even more data accumulating thanks to the open-source Phoronix Test Suite and OpenBenchmarking.org. Thousands of data points are building up for these very exciting AMD Zen 3 desktop processors...

http://www.phoronix.com/scan.php?page=news_item&px=AMD-Ryzen-5000-Zen-3-Linux-Perf

[edwaleni]

And if someone (anyone) gets their Mac Pro on the 17th, can you please give Michael shell access so we can get reality on the M1 compared to Ryzen? Nuvia is saying they are even better than the M1. I would like to see a good round of PTS results so everyone can get context.

[Michael]

And if someone (anyone) gets their Mac Pro on the 17th, can you please give Michael shell access so we can get reality on the M1 compared to Ryzen? Nuvia is saying they are even better than the M1. I would like to see a good round of PTS results so everyone can get context.

I'll have extensive M1 benchmarks around the ~18th.

[f0rmat]

I'll have extensive M1 benchmarks around the ~18th.

OUTSTANDING! I am really looking forward to seeing them.

[vein]

So sick and tired of hearing about that shitty in read lock-in m1 processor. Can you just shut up about it.

[creative]

Oh man, the availability is tearing me asunder. I really wish I could order the 5900x. I have never wanted a specific processor so intensely... well I think this is the first time I have had new cpu fever this bad.

The pain!

[Jabberwocky]

Oh man, the availability is tearing me asunder. I really wish I could order the 5900x. I have never wanted a specific processor so intensely... well I think this is the first time I have had new cpu fever this bad.

The pain!

I'm just happy we aren't not stuck on 4core 8thread CPUs anymore!

New BIOS updates for x470 are aparently coming early next year. Waiting patiently for people to test iommu groupings before I upgrade.

[BillBroadley]

And if someone (anyone) gets their Mac Pro on the 17th, can you please give Michael shell access so we can get reality on the M1 compared to Ryzen? Nuvia is saying they are even better than the M1. I would like to see a good round of PTS results so everyone can get context.

One of the unique things about the M1 is the on package memory chips. I believe it's 4 x 32 bit channels, not the usual 2 x 64 bit. This should show up on parallel (from multiple cores/threads) random access. If you need a benchmark for that I've written one.

[polarathene]

I really wish I could order the 5900x. I have never wanted a specific processor so intensely...

I'm still on the fence between 5800X and 5900X. I'd rather not spend the extra for 5900X and it seems that unless I'm doing something that's going to use all cores/threads, the advantage is minimal vs all cores in the same CCX, which should be less scheduler concerns?

The 5900X isn't double 5600X in perf, and 5800X often when showing gains over a 5600X can be more than 25%, being closer to the 5900X results, other than that the 3 CPUs seem to scale fairly linearly.

Not sure why some benchmarks like FFmpeg are showing 5600X having a notable lead over the bigger models though? Is this a temps thing? Or CCX related for 5900X + 5950X? Some benches favor the 5600X and 5800X by about 10%, presumably single thread might be hopping between CCX on the multi-CCX variants?

Especially odd when the 5900X shows an improvement over 5600X but a 5950X performs worse than the 5900X, sometimes with a 5800X also performing better than the 5950X, eg libgav1 and LeelaChessZero. Curious what causes 5950X in many benches to be about the same as 5900X.

Something seems incredibly bizarre with LULESH results, everything scores around 10-11, except the 5600X doing a ridiculous 993 in comparison? I raised an issue to inquire about it here.

"OneDNN 1.5 Matrix Multiply Batch Shapes Transformer - Data Type f32", this is about 3x perf difference between the single CCX (slower) and multi-CCX. Amount of cores/threads seem to make little difference to scaling that performance, I assume this one is an example of where the doubled L3 cache is making a difference?

---

So what are the main differences to argue between 5800X and 5900X? (ignoring multi-core/thread perf and price)

Lower overhead/latency of a single CCX, which is capable of bit better single thread perf(at least for 5800X vs 5900X), but PBO may also be more limited due to thermals (Full 8-core CCX vs 2-cores disabled per CCX and further thermal load balancing between CCX). The doubled L3 cache for multi-CCX is probably useful too?

[creative]

I'm still on the fence between 5800X and 5900X. I'd rather not spend the extra for 5900X and it seems that unless I'm doing something that's going to use all cores/threads, the advantage is minimal vs all cores in the same CCX, which should be less scheduler concerns?

The 5900X isn't double 5600X in perf, and 5800X often when showing gains over a 5600X can be more than 25%, being closer to the 5900X results, other than that the 3 CPUs seem to scale fairly linearly.

Not sure why some benchmarks like FFmpeg are showing 5600X having a notable lead over the bigger models though? Is this a temps thing? Or CCX related for 5900X + 5950X? Some benches favor the 5600X and 5800X by about 10%, presumably single thread might be hopping between CCX on the multi-CCX variants?

Especially odd when the 5900X shows an improvement over 5600X but a 5950X performs worse than the 5900X, sometimes with a 5800X also performing better than the 5950X, eg libgav1 and LeelaChessZero. Curious what causes 5950X in many benches to be about the same as 5900X.

Something seems incredibly bizarre with LULESH results, everything scores around 10-11, except the 5600X doing a ridiculous 993 in comparison? I raised an issue to inquire about it here.

"OneDNN 1.5 Matrix Multiply Batch Shapes Transformer - Data Type f32", this is about 3x perf difference between the single CCX (slower) and multi-CCX. Amount of cores/threads seem to make little difference to scaling that performance, I assume this one is an example of where the doubled L3 cache is making a difference?

---

So what are the main differences to argue between 5800X and 5900X? (ignoring multi-core/thread perf and price)

Lower overhead/latency of a single CCX, which is capable of bit better single thread perf(at least for 5800X vs 5900X), but PBO may also be more limited due to thermals (Full 8-core CCX vs 2-cores disabled per CCX and further thermal load balancing between CCX). The doubled L3 cache for multi-CCX is probably useful too?

Sorry but I can't ignore the multi-core and thread performance on this one.

The main different between the two in my opinion is the ability to basically have 12 threads of rendering power while having 12 threads of gaming power all at the same time. I am learning blender, do video editing/rendering and music production with Mixbus. I want to get into virtual machines as well. Personally the 5900x fits the bill for its price point.

Also it is a chip I would very much like to keep in service for the next 8-10 years, provided my x570-p lives that long with its Delta superflo fan—which from what I understand is much like a fan in laptops and those can last for quite some time. I am hoping the L10 rating for the bearing holds up which equate to 7 years of 24/7 use supposedly.

Those are my reasons. RIght now I have a 3600 and it is awesome but I would very much enjoy twice the thread power and that additional IPC single thread uplift. Not a bad upgrade from what I have in my opinion, and it runs much cooler than the 3900x/3900xt.

I can't isolate and speak for the advantages of the higher L3 cache but I am sure its part of the total performance of the processor package.