Question Underwhelming RAM performance on Ryzen 3000

[aidan-astwood]

Installed 4 X 8 GB Flare X 3200MHZ CL 14 in my gigabyte X570 prowifi Ryzen 3800X build.
Runs stable with XMP enabled.
Checked mulitiple system analysis tools, all show expected frequency, timings, command rate, voltages etc.

However, performance is pretty bad. Benchmarking on UserBenchmark website says my RAM is performing significantly below expectations after multiple runs, no other significant programs running. Multicore bandwidth was 34.4 GB/S and singlecore was 30.7 GB/S. Latency was pretty typical for the kit and not a concern.
Now, looking at other benchmarks, this is a recurring theme, the same kit of RAM is performing below expectations on every 3600, 3600X, 3700X, 3800X build ive seen. Only one 3900X benchmark i saw looked normal.
This is contrasted to the 2700X which generally scores 40-45 GB/S multi and 35-40 GB/S single. And obviously the threadrippers with quad channel score quite a lot higher.
This also affects X470 boards running 3000 series CPUs too. And it affects 2 X 8 GB configs as well, bandwidth RAM speeds are much lower with ryzen 3000 series CPUs than 2000 series with the same RAM, whether 2 or 4X 8GB.

Looking at my passmark RAM benchmark however, the scores seem pretty normal.

Im thinking is this a bug in the UserBenchmark software, is it just not configured to benchmark RAM with Ryzen 3000 properly yet? Or is this an issue with immature BIOS or something else?

 

[rgd1101]

don't use userbenchmark.

 

[aidan-astwood]

don't use userbenchmark.

Is userbenchmark known for not being useful in ram benchmarking?

 

[alceryes]

No, Userbenchmark is okay and consistent.

The problem with most (all?) benchmarks is that they don't measure real-world performance well. Many of them also play with the graphs and numbers so that the scale is nowhere near representative of what you would 'feel' in real life.
E.g. your system is at the bottom of a chart with fifteen CPUs above it. You think you're having major performance issues until you realize that the whole chart represents about 3 seconds of real time.

Please post the link to your Userbenchmark result along with a CPUZ screen-grab of the memory and SPD tabs.

Here's one of my Userbenchmarks - https://www.userbenchmark.com/UserRun/17688023

I recently upgraded from an i3-8350k to an i9-9900k and one thing I noticed was that my 'memory kit' score went down about 15% with the upgrade. Even when overclocking the memory past what I had it at previously, I can't obtain the memory scores I was getting with the i3 (click through the history to see my attempts ). It looks like the extra on-chip cache causes more of a strain on the memory controller, slowing it down just a bit.

 

[rgd1101]

it good for comparing the same setup. not for comparing to everything(the expectations thing)

 

[aidan-astwood]

No, Userbenchmark is okay and consistent.

The problem with most (all?) benchmarks is that they don't measure real-world performance well. Many of them also play with the graphs and numbers so that the scale is nowhere near representative of what you would 'feel' in real life.
E.g. your system is at the bottom of a chart with fifteen CPUs above it. You think you're having major performance issues until you realize that the whole chart represents about 3 seconds of real time.

Please post the link to your Userbenchmark result along with a CPUZ screen-grab of the memory and SPD tabs.

Here's one of my Userbenchmarks - https://www.userbenchmark.com/UserRun/17688023

I recently upgraded from an i3-8350k to an i9-9900k and one thing I noticed was that my 'memory kit' score went down about 15% with the upgrade. Even when overclocking the memory past what I had it at previously, I can't obtain the memory scores I was getting with the i3 (click through the history to see my attempts ). It looks like the extra on-chip cache causes more of a strain on the memory controller, slowing it down just a bit.

https://www.userbenchmark.com/UserRun/18629943

https://valid.x86.fr/xphexn

Those are my scores running just 16 GB.

I also ran the passmark RAM benchmark, 6 of 7 tests were fine, actually good results, but memory threaded test was terrible relative to many others score with same RAM.

If you look at others scores on Userbenchmark with the same RAM, you can see everyone else with 3600, 3600X, 3700X and 3800X is getting bad scores compared to 2000 and 1000 series Ryzen CPUS. I had the same sticks with my Ryzen 2700X, it always performed above expectations with Userbenchmark. Interestingly this RAM is benchmarking well with the 3900X.

You may have a point about RAM performance varying with CPU architecture. 3800X and 3700X have 36MB of cache compared to 20 MB for the 2700x for example.
I dont think that is the issue in this case though. I think its the new chiplet design of Zen 2 CPUs, 3600, 3600X, 3700X and 3800X have a core die and a I/O die, the 3900X and upcoming 3950X have 2 core dies and a I/O die. Compared to the 1000 and 2000 series having a single monolithic die.

 

[dudeman509]

Performance may change as BIOS updates and Windows updates are released to better take advantage of the new architecture. Similar to the power plan issues with scheduling on Ryzen 1000, it sounds like the firmware/release issues of the 3000 series could have been ironed out a bit more before launch.

 

[Crosslhs82x2]

If it will help for a 2600x and flarex hers my run.
[URL="http://Www.userbenchmark.com/userrun/18324288[/URL]"]Www.userbenchmark.com/userrun/18324288[/URL]

 

[aidan-astwood]

If it will help for a 2600x and flarex hers my run.
[URL="http://Www.userbenchmark.com/userrun/18324288[/URL]"]Www.userbenchmark.com/userrun/18324288[/URL]

Thanks.
This shows my point. You've got them at 3400mhz which is why the result is higher than usual but even if you had it at XMP 3200 it would only be a little lower. With that peak of scores on the right like I used to get with my 2700X.
But now all the 6 and 8 core ryzen 3000 cpus get a score with that ram significantly lower with that.

 

[aidan-astwood]

Performance may change as BIOS updates and Windows updates are released to better take advantage of the new architecture. Similar to the power plan issues with scheduling on Ryzen 1000, it sounds like the firmware/release issues of the 3000 series could have been ironed out a bit more before launch.

Yeah, hope so.
See what eventuates in the next few weeks.
It's not a big deal. Little effect on real world performance but still a curiosity.

 

[Crosslhs82x2]

Have you tried 16gb using ryzen dram calculator to see what you can push them to?

 

[alceryes]

@aidan-astwood, you had mentioned that RAM voltage is 'as expected'. What voltage is it hitting?

Load up HWiNFO64 sensors and let it run in the background as you run the Userbenchmark (or some other benchmark that stresses RAM). After a few mins cancel the bench and then scroll down to motherboard section and report back what the DIMM 'maximum' is reporting.

 

[JayGau]

This thread is old now but here is the explanation:

In memory bandwidth, we see something odd, the write speed of AMD's 3700X, and that's because of the CDD to IOD connection, where the writes are 16B/cycle on the 3700X, but it's double that on the 3900X. AMD said this let them conserve power, which accounts for part of the lower TDP AMD aimed for.

Read more: https://www.tweaktown.com/reviews/9051/amd-ryzen-3900x-3700x-zen2-review/index3.html

It's why the ram always performs so bad on UserBenchmark with the 8 cores 3000x chips. With my 2700x my ram was always at "perform way above expectation" and now that I have a 3700x it always performs below expectation whatever I do (overclock, timing optimization, voltage, etc.).