Help with Ryzen 5600x and XMP

[Thaddeus_Pi]

I recently built a new computer:

 

CPU: Ryzen 5 5600x using the Hyper 212 Black Edition Cooler

Mobo: Gigabyte Aorus Elite X570 (with latest BIOS)

RAM: G.SKILL Trident Z RGB Series 32GB (2x16GB) DDR4 3600MHz CL17 Dual Channel Memory Kit 1.35V (F4-3600C17D-32GTZR)

GPU: GTX 1070

 

I ran benchmarks on stock (cinebench r23, 3DMark, Unigine Valley, Crystal disk mark). I monitored with HWInfo64 - I was getting an all core boost at 4.65GHz. All the thermals looked good (highest was 70C using 3DMark). No issues.

 

I turned XMP on, and it worked fine, but when I went to bench, my scores were lower, and HWInfo64shows the all-core boost was reduced to 4.3. I couldn't even hit 4.65 on a single core. The RAM was operating at 3600MHz though.

 

Why is my CPU operating at a lower clock with XMP on and and how do I fix it? Should I manually tune the RAM to a lower clock (say 3200MHz)?

 

Please feel free to state the obvious - I'm clearly new to this.

 

 

[Nathanpete]

There should be zero correlation between the two. The two do no affect each other other than the CPU and the memory controller share the same CPU package. 

 

Shut down the PC, wait a while, power it back up, and it if it is doing it again, then yeah, i guess you can try to manually tune it down to 3433 or 3200 MHz. 

[Thaddeus_Pi]

How would one manually tune it?

 

I thought it might be a power issue - the increased RAM speeds taking away from the power budget for the CPU.

[riklaunim]

3 minutes ago, Nathanpete said:

There should be zero correlation between the two.

Assuming poor cooling higher temp on the memory controller would reduce the core clock speeds.

 

9 minutes ago, Thaddeus_Pi said:

I turned XMP on, and it worked fine, but when I went to bench, my scores were lower, and HWInfo64shows the all-core boost was reduced to 4.3. I couldn't even hit 4.65 on a single core.

What's the CPU temperature?

[Thaddeus_Pi]

8 minutes ago, riklaunim said:

Assuming poor cooling higher temp on the memory controller would reduce the core clock speeds.

 

What's the CPU temperature?

 

Idle: <30C

Gaming: around 50C

Benching: 60C (70C in 3dMark)

[Thaddeus_Pi]

A bit of an update here - I turned on XMP and looked closely at HWInfo.

 

What change between XMP and non-XMP is in the "Frequency Limit - Global". With XMP on it goes down during Cinebench testing.

 

I can still register 4.6Ghz all core boosting, but for whatever reason, during Cinebench testing, my all core frequency drops with XMP on.

 

I ran 3DMark (Time Spy) and my scores improved for both CPU and Graphics - going from 7497/5765 (CPU/GPU) to 8301/5800 with XMP on.

 

Thermals are great.

 

Can anyone shed some light on the differences between Cinebench and 3DMark? Is it a power budget issue in cinebench?

[Nathanpete]

10 hours ago, Thaddeus_Pi said:

A bit of an update here - I turned on XMP and looked closely at HWInfo.

 

What change between XMP and non-XMP is in the "Frequency Limit - Global". With XMP on it goes down during Cinebench testing.

 

I can still register 4.6Ghz all core boosting, but for whatever reason, during Cinebench testing, my all core frequency drops with XMP on.

 

I ran 3DMark (Time Spy) and my scores improved for both CPU and Graphics - going from 7497/5765 (CPU/GPU) to 8301/5800 with XMP on.

 

Thermals are great.

 

Can anyone shed some light on the differences between Cinebench and 3DMark? Is it a power budget issue in cinebench?

CInebench uses AVX instructions. AVX instructions are known as a "power virus" as they require a ton of wattage to run at full speed. So much so that motherboards have a special offset for CPU clock speed when running AVX instructions. It is called the AVX offset and is usually 100-200 MHz (or 0.1-0.2 GHz) below the maximum turbo speed. 

 

3DMark is primarily a GPU test but the CPU test uses more common instruction sets that are easier to run and aren't considered a power virus. They still pin the CPU to 100%, but they are running at a higher clock speed than AVX code. 

[Thaddeus_Pi]

10 hours ago, Nathanpete said:

CInebench uses AVX instructions. AVX instructions are known as a "power virus" as they require a ton of wattage to run at full speed. So much so that motherboards have a special offset for CPU clock speed when running AVX instructions. It is called the AVX offset and is usually 100-200 MHz (or 0.1-0.2 GHz) below the maximum turbo speed. 

 

3DMark is primarily a GPU test but the CPU test uses more common instruction sets that are easier to run and aren't considered a power virus. They still pin the CPU to 100%, but they are running at a higher clock speed than AVX code. 

 

Well this makes a lot of sense. I'm seeing exactly that drop. The only difference is I see the drop again when XMP is enabled. Does XMP bring the CPU closer to the total allowed wattage?

[Nathanpete]

24 minutes ago, Thaddeus_Pi said:

 

Well this makes a lot of sense. I'm seeing exactly that drop. The only difference is I see the drop again when XMP is enabled. Does XMP bring the CPU closer to the total allowed wattage?

The memory controller and SoC take very very little wattage, and your BIOS allows you to lift the power limit somewhere deep in the BIOS. 

[Thaddeus_Pi]

23 hours ago, Nathanpete said:

The memory controller and SoC take very very little wattage, and your BIOS allows you to lift the power limit somewhere deep in the BIOS. 

 

Well then I'm still confused why the Cinebench r23 score is lower with XMP on.

 

I found this thread, with other uses discussing lower cinebench scores with XMP on. I don't think any of them tried with XMP off.

 

That said, 3d Mark Time Spy score improved, as I said above, so perhaps this is a bit of a nothing issue. I'd still like to understand what is going on.

 

[Nathanpete]

1 hour ago, Thaddeus_Pi said:

 

Well then I'm still confused why the Cinebench r23 score is lower with XMP on.

 

I found this thread, with other uses discussing lower cinebench scores with XMP on. I don't think any of them tried with XMP off.

 

That said, 3d Mark Time Spy score improved, as I said above, so perhaps this is a bit of a nothing issue. I'd still like to understand what is going on.

 

I can't help you better understand your situation. Cinebench is a good benchmarking score for CPU OC, but its use of AVX instructions makes it unrealistic for your normal user. 3Dmark Time Spy is actually useful using real world examples, and if you score improved there I consider that a win. 

[freeagent]

42 minutes ago, Nathanpete said:

I can't help you better understand your situation. Cinebench is a good benchmarking score for CPU OC, but its use of AVX instructions makes it unrealistic for your normal user. 3Dmark Time Spy is actually useful using real world examples, and if you score improved there I consider that a win. 

To be fair, cinebench is not that bad to run. Avx is not an unrealistic load, to me it sounds like your cooling is not up to snuff..

 

Cinebench doesn’t really care a ton about memory.

[Nathanpete]

10 hours ago, freeagent said:

To be fair, cinebench is not that bad to run. Avx is not an unrealistic load, to me it sounds like your cooling is not up to snuff..

 

Cinebench doesn’t really care a ton about memory.

I mean but then why is his scores decreasing when modifying his memory speed? Sure he is using a Hyper 212 Evo and a 5800X, but still. 

 

AVX is not a popular instruction set, especially for games, in fact some older CPUs don't support it and games are specifically patched to remove AVX, AVX2, or AVX-512 instructions so that games don't crash when encountering this code. In fact one of the first few daily CP 2077 patches was to specifically remove AVX code that appeared around the 1 minute prologue cutscene because players on pretty old AMD CPUs (how they ran on there IDK) would be stuck here because the game would crash in the same spot everytime. 

[freeagent]

XMP may make some of his subtimings pretty slack.. his 212 may not be keeping his 5600X cool, who knows..

[Thaddeus_Pi]

10 hours ago, Nathanpete said:

I mean but then why is his scores decreasing when modifying his memory speed? Sure he is using a Hyper 212 Evo and a 5800X, but still. 

 

AVX is not a popular instruction set, especially for games, in fact some older CPUs don't support it and games are specifically patched to remove AVX, AVX2, or AVX-512 instructions so that games don't crash when encountering this code. In fact one of the first few daily CP 2077 patches was to specifically remove AVX code that appeared around the 1 minute prologue cutscene because players on pretty old AMD CPUs (how they ran on there IDK) would be stuck here because the game would crash in the same spot everytime. 

 

10 hours ago, freeagent said:

XMP may make some of his subtimings pretty slack.. his 212 may not be keeping his 5600X cool, who knows..

 

My thermals are great - I don't break 60C in Cinebench. My scores went up in 3dMark Time Spy. My core clocks were maxed using 3dMark as well.

 

The issue is the drop in core clock in Cinebench when XMP is enabled (compared to when it is disabled).

 

[freeagent]

When I let it control itself I have asus performance enhancer enabled so it keeps it clocks, maybe your gigabyte bios has something similar..

 

I just run one speed one voltage works great..