Ryzen Vermeer (5000 series) CPU has now appeared in the CPU statistics page.

CPU model	                                                 Num. computers	cores/computer	GFLOPS/core	GFLOPs/computer
AMD Ryzen 5 5600X 6-Core Processor [Family 25 Model 33 Stepping 0]	11	   11.45	5.84		66.85

Interested to see if the numbers improve over the next few weeks.

Ryzen 9 5900X and Ryzen 9 5950X have also entered the list. Yet to see Ryzen 7 5800X (doesn't seem to be that popular)

The Ryzen 5 5600X has slid down the listing as the number of 5600X CPUs grow. This latest generation of AMD CPUs are not living up to the AMD hype at the moment. Currently performing same as, or worse than the previous generation.

This latest generation of AMD CPUs are not living up to the AMD hype at the moment. Currently performing same as, or worse than the previous generation.

I wouldn't pay the slightest bit of attention to the numbers on the CPU models list.
Due to the way GFLOPs are calculated, those numbers are pretty much meaningless.

And on Rosetta with it's fixed length runtimes the only way to compare performance is RAC- and once again due to the way it is determined, and not knowing just how many hours a day the system you are looking at is actually processing Rosetta Tasks, and how many cores & threads are in use, and if the system is busy doing other things at the same same, and it take 6-8 weeks for RAC to stabilise around it's nominal value, it makes it difficult (to almost impossible) to know if you are actually making a valid comparison.

Other projects that have Tasks that don't have fixed runtimes give a much better indication of a system's performance. The sooner it produces a Valid result, the more powerful it is.

---

Grant
Darwin NT

Due to the way GFLOPs are calculated, those numbers are pretty much meaningless.

Indeed, that's stange to me
I saw a lot of benchmark and 5xxx series are from 15% to 25% better than 3xxx series.
Obviously the code is not optimized (for example, GCC has introduced a Zen3 initial support, not complete, with good results), so numbers are "uncertain"

---

Other projects that have Tasks that don't have fixed runtimes give a much better indication of a system's performance. The sooner it produces a Valid result, the more powerful it is.

Thanks for the feedback Grant

Without finding a valid completed task, here or on another project.... a bit hard to find a way to compare performance on BOINC. I have seen all the reviews (including Phoronix, many thanks boboviz for the link). Now just need confirmation of BOINC performance.

If anyone with a Vermeer processor can chime in, that would be most appreciated.

Now just need confirmation of BOINC performance.

Talking about BOINC performance is meaningless.
BOINC is a manager- it manages computing resources for projects. It doesn't actually process any work. Each project produces it's own applications and they do the processing work.

You need to find a host with the CPU you are interested in, and look t it's returned Tasks & then compare them with other systems running the same application (some projects can have dozens of different applications, not to mention a single application can process many different types of data resulting in different computation times- you have to compare like with like).
Which really would be un-necessary IMHO as the benchmarks as posted by Phoronix and the non-synthetic ones by sites such as Anandtech are relevant results showing the actual performance of the CPUs being tested on actual programmes using real data.

---

Grant
Darwin NT

Which really would be un-necessary IMHO as the benchmarks as posted by Phoronix and the non-synthetic ones by sites such as Anandtech are relevant results showing the actual performance of the CPUs being tested on actual programmes using real data.

I think that a single person (Larabel) that created his own tools of benchmark, a site with a lot of different benchmarks (not only standard, but also with optimized app/kernel/compilations) and daily news about open software is simply GREAT!
And not "un-necessary".

---

Which really would be un-necessary IMHO as the benchmarks as posted by Phoronix and the non-synthetic ones by sites such as Anandtech are relevant results showing the actual performance of the CPUs being tested on actual programmes using real data.

I think that a single person (Larabel) that created his own tools of benchmark, a site with a lot of different benchmarks (not only standard, but also with optimized app/kernel/compilations) and daily news about open software is simply GREAT!
And not "un-necessary".

I suggest you re-read what i typed, and you quoted, but I'll spell it out anyway- it is unnecessary to go to the effort of tracking down comparable systems with comparable applications processing comparable data thanks to the work of Phoronix and Annandtech & other such websites.

---

Grant
Darwin NT

I suggest you re-read what i typed, and you quoted, but I'll spell it out anyway- it is unnecessary to go to the effort of tracking down comparable systems with comparable applications processing comparable data thanks to the work of Phoronix and Annandtech & other such websites.

ALL benchmarks say that 5xxx is better than 3xxx series.
If Rosetta@Home is the same (or worst) with 5xxx family, the problem is Rosetta code. Not cpu.
Let's not even talk about optimizations, it's wasted time

---

I suggest you re-read what i typed, and you quoted, but I'll spell it out anyway- it is unnecessary to go to the effort of tracking down comparable systems with comparable applications processing comparable data thanks to the work of Phoronix and Annandtech & other such websites.

ALL benchmarks say that 5xxx is better than 3xxx series.
If Rosetta@Home is the same (or worst) with 5xxx family, the problem is Rosetta code. Not cpu.
Let's not even talk about optimizations, it's wasted time

But surely the figures are not showing the work being done by rosetta@home but the totally artificial benchmark figures held against each host, in which case the low figure for the 5xxx series is likely a result of a few machines holding the default figures because the benchmarks have not been run.

This has been highlighted over the past couple of weeks as a known problem and would hit new hosts like the 5xxx to a greater extent than more established processors.

ALL benchmarks say that 5xxx is better than 3xxx series.

And once again, you completely ignore what it is that i typed.
*shrug*

---

Grant
Darwin NT

Ryzen 9 5900X and Ryzen 9 5950X have also entered the list. Yet to see Ryzen 7 5800X (doesn't seem to be that popular)

The Ryzen 5 5600X has slid down the listing as the number of 5600X CPUs grow. This latest generation of AMD CPUs are not living up to the AMD hype at the moment. Currently performing same as, or worse than the previous generation.

<cough>

Not set-up properly yet. And I may run out of Rosetta tasks before it is.
Currently very variable for reasons I don't yet comprehend, but I'll get there eventually

Edit: My i3-8350K is showing
Device peak FLOPS 5.52 GFLOPS

But the 5800X is showing
Device peak FLOPS 1.00 GFLOPS

I've run CPU benchmarks. Is there something else I should do or am I waiting for site feedback before that changes to whatever it really is? I'm out of my depth on this.
Or is it related to something else entirely?

FYI:
Measured floating point speed 5734.02 million ops/sec
Measured integer speed 23423.87 million ops/sec

Tweaking will definitely improve that once I get my head straight on how to optimise everything - currently running at only ~4482MHz

---

Device peak FLOPS 1.00 GFLOPS

I've run CPU benchmarks. Is there something else I should do

I don’t know of anything else you can do… Maybe it takes a while for the host details to propagate to task results? Give it a day and check again.

Device peak FLOPS 1.00 GFLOPS

I've run CPU benchmarks. Is there something else I should do

I don’t know of anything else you can do… Maybe it takes a while for the host details to propagate to task results? Give it a day and check again.

Thanks.
I'm patient, but it will be nice to see how it eventually compares

---

FYI:
Measured floating point speed 5734.02 million ops/sec
Measured integer speed 23423.87 million ops/sec

That shows that the benchmarks have been run.
For comparison my i7-9700K benchmarks,

Measured floating point speed 4928.32 million ops/sec
      Measured integer speed 18495.04 million ops/sec

From memory, it takes 10 or so Valid results for the APR (Average Processing Rate) to be determined, and i think the Device Peak FLOPS are based on that/make use of that value to get their value. While Rosetta does have it's own Credit mechanism, there are times where it does make use of Credit New, so until those numbers stabilise Credit will be much, much more random than usual.
And that tends to be the case whenever a new type of Task comes out- Credit all over the place till you (and a large number of other crunchers) have returned a dozen or so Valid results.


Edit- Your last couple of Valids are showing a Device peak FLOPS of 5.73.

---

Grant
Darwin NT

Measured floating point speed 5734.02 million ops/sec
Measured integer speed 23423.87 million ops/sec

That shows that the benchmarks have been run.
For comparison my i7-9700K benchmarks,

Measured floating point speed 4928.32 million ops/sec
      Measured integer speed 18495.04 million ops/sec

From memory, it takes 10 or so Valid results for the APR (Average Processing Rate) to be determined, and I think the Device Peak FLOPS are based on that/make use of that value to get their value. While Rosetta does have it's own Credit mechanism, there are times where it does make use of Credit New, so until those numbers stabilise Credit will be much, much more random than usual.
And that tends to be the case whenever a new type of Task comes out- Credit all over the place till you (and a large number of other crunchers) have returned a dozen or so Valid results.

Edit- Your last couple of Valids are showing a Device peak FLOPS of 5.73

So they do.
That 5.73 was at 4517GHz, which I've tuned to 5.75 at 4591GHz effective clock this evening with just a little way more to go.

My i3-8350K is 4 cores at 4738GHz giving 5.52, compared to 16 hyperthreaded cores at 4591GHz giving 5.75
The old FX-8370 I think was 8 cores at 3.95 GFLOPS at its peak, though dipped down to 2.78 towards the end of its life

Looking decent

---