PC Processors

[hamishka]

Hey guys. Recently I built a system with the Ryzen 7 5800x and a RTX 3070TI. I play CS:GO mostly and the big problem is that i only get around 160 fps on the benchmark map, in general low fps.

Everything is on stock in BIOS, only XMP active and still no more fps. I made some benchmarks and the results are realy bad. Especially the single core performance in cpu z benchmark is max 142 (LOL) and in yt videos they get around +500 on stock.

Is the CPU broke or something ??

System Info: Ryzen 7 5800x - B550 Gaming Plus - RTX 3070TI - 2x8 GB CORSAIR VEGEANCE

If you need some other screenshots pls tell. I really need your help

[Gwillakers]

Nah,  CPU is not broke of something...

Um...    This is a good one...  Lets get some more info...

(suspicion it may have something to do with Memory subsystem)

1.  Take a look at task manager when single core CPU-z is running....  Is the execution bouncing among Cores? or is it more a steady same core execution?

      You can see the same thing in Ryzen Master if you expand that die caret...

2.  Take a look at event viewer ==> custom views ==> administrative events  (are there many WHEA errors?  

3.  Run AIDA ... and take a look at read and write bandwidth for memory

The reason I ask you to do the above is that all 4GHz executions are not the same as other 4GHz executions.

 The CPU waits and waits for memory   During that waiting, the core is active... it just is not getting any work done.

 You see memory waits are supposed to be of a short duration, thus the Core does not give up control.

 Unlike a core that issues a I/O to the disk subsystem.  There, that wait is expected to be of a long duration.

The CPU gives up control and the task is suspended until the Disk interrupts the CPU after completion of the I/O.

 In the meantime, the Core is free to work on something else.

Oh by the way...  There use to be software that allowed you to turn off caches so you could test the Dimms, and not just a repeated hammering of the L3 cache.   If you ever turn off cache, be prepared for your very fast 4GH+ computer to run no faster than a 386 from the mid 90's.    You can literally watch the screen getting built. It is that slow.

[hamishka]

Hey thank you for the long answer.

Everything is okay in task manager while cpu z is running. Only one core goes up.

Event manager also looks good.

The third point is a little bit confusing for me. I noticed that when i turn xmp on in bios my ram mhz stays at 1000 mhz (2000) in cpuz but i have a 3600 mhz. Also after rebooting into the bios, it says 2133mhz on top but xmp is active.

The aida test results:

Read

Write:

This one looks strange. Latency:

[ryzen_type_r]

Your mem bandwidth read and write scores are a little higher than mine (I'm running 3200 mem), so your memory is running at the right speed (a bit higher than 3400 actually since your mem clock says 1766) despite what your mobo reports but there is definitely something wrong with those latency numbers, they should not be almost twice as slow as a Celeron D from 2004!  LOL

What is your Fclk (infinity fabric clock) running at?  HWinfo will tell you.  AMD says for least latency it should run 1:1 with the mem controller clock, so in your case should be 1766, but maybe your mobo is decoupling them and running the Fclk much lower or something.

Make sure your BIOS is updated.  I suspect the BIOS is causing the issues.

[Gwillakers]

Your Read bandwidth looks good,   I had 2x but am running quad channel.

Your Write speed dropped off, seems about half of your read speed. My write was actually faster than my read.

And like Ryzen_type_r stated your Latency is way off.    Mine is not good at 70ns but yours is really bad.

I am pretty darn sure now that the memory subsystem is the root cause of this problem.

What I was trying to say in part 3 of my post before, was that Frequency or 4.3GHx does not determine how much work is done.

If the processor is waiting for the memory subsystem, (and yours is with those latency numbers) then it is not getting as much work done as processors who are not waiting as much on their memory.

Something is messed up somewhere.  But it is with your memory system, not with your processor.

Sometimes a different bios will  at bootup train the memory differently.  So a different bios might give you better results.

But before loading a new bios... (or an older bios)...

1. Make sure you are running the sticks in the right slots. (although this will not account for this whole problem)

2. Take XMP off.    Load Optimized defaults.   Reboot.

     This will run the memory at a slower speed,   however with the selection of new timing parameters you may solve your latency issue.   Run AIDA again, with the other benchmarks.    Even with the slower memory settings, you should not be experiencing that poor of a performance.

[ryzen_type_r]

@Gwillakers  My Aida numbers were showing the same thing for memory write speeds.  Did a little digging and apparently this is normal for single-CCD Ryzens.  Part of an architectural change starting with Zen2.  AMD says this has little effect on real world performance.

https://community.amd.com/t5/processors/low-memory-write-speed-zen2-3700x/td-p/51401

But something's definitely up with hamishka latency numbers.

[ryzen_type_r]

Yes, there's something wrong.  Your MT performance is about right, but my 5800X is getting 650 on single threaded in that same benchmark, at stock settings.

Is this a clean install of Windows?  If you've installed any motherboard or other kind of hardware tuning/tweaking utilities (aside from Ryzen Master) try disabling or uninstalling them, sometimes they can interfere in strange ways.

Make sure your BIOS is up to date.

Also try posting up a 7-Zip x64 19 benchmark, with dictionary size at default 32MB and threads set to 1.  I'll compare it to mine just to rule out the possibility of something wonky with your CPU-Z (I highly doubt it though).

P.S.  in HWinfo, don't look at the Core clocks, look at the Core Effective Clocks instead - these will show you which cores are actually doing real work.  When running the benchmark, you should see one of the cores boost all the way up.  The CPUZ benchmark only runs for a short time, use the 7-zip bench b/c it just keeps running until you tell it to stop.  Easier to keep an eye on things when it's constantly running.  If you see the single core clock is abnormally low, that would explain the low bench scores (but wont tell you why though).

[hamishka]

Here is the benchmark. Its a clean install of windows yes