Question GPU bench results lower after CPU overclock?

[grebgonebad]

Hey all,

I recently put together a rig for my nephew and have been having a play around with it to see how it runs, and as part of that I've been tinkering with the clocks a little to see what I can get out of the thing. Relevant specs are as follows:

AMD Ryzen 5 1600X
Corsair H100i Aio
16gb Corsair Vengeance RGB Pro (2 x 8) 3200mhz
EVGA SC GTX 970
Asus TUF B450M Plus Gaming
EVGA 850 GQ


I've been using Passmarks Performance Test 10 as my bench. I know, not exactly the best bench out there but it's one I'm familiar with, plus I'm only using it to get an idea of how much I've improved (or not) the different performance points of the system.

And so, this brings me onto my issue. At stock clocks the CPU (3.6ghz, 1.375V) and GPU (1317ghz GPU, 7010mhz Mem) scores were as follows:

CPU - ~13,000 (50-55C max temps)
GPU - ~8800 (65-70C max temps)

After OC'ing the GPU (1417mhz GPU, 7510mhz Mem) but NOT the CPU, the scores were as follows:

CPU - ~13,000 (as before)
GPU - ~10,000 (70-80C max temps)

Great news right? So then I turned my hand to the CPU and started fiddling with it's clocks, got it running stable (4.0ghz, 1.4V), ran the bench again and got the following scores:

CPU - ~14,000 (60-70C max temps)
GPU - ~8000 (65-70C max temps)

As I'm sure you can understand I was petty taken aback when I saw the staggeringly low GPU score after the CPU OC. The GPU clocks were still in their OC'd state, the only thing that had changed were the CPU clocks. I know from experience that sometimes power/thermal limits on the CPU can negatively affect GPU performance, however given the still more than reasonable CPU temps coupled with the fact that the score was improved modestly, I can't see this being the issue? However I will admit that this is my first time OC'ing a Ryzen based system so I won't be so bold as to assume I'm doing everything correct.

I will also note at this point that I reverted the CPU clocks back to default (as they were during my initial GPU overclock) and the GPU score went straight back up to where I expected it to be. So I'm certain it's something to do with my CPU overclock that's causing the issue.

So yeah, if anyone has any suggestions or could shed some light as to what's exactly going on here I'd very much appreciate it. Thanks in advance!

 

[rgd1101]

make and model of the psu?

 

[grebgonebad]

make and model of the psu?

EVGA 850 GQ

Pretty sure that's more than enough juice to handle whatever I throw at it. Ha ha.

 

[Phaaze88]

EVGA 850 GQ

Pretty sure that's more than enough juice to handle whatever I throw at it. Ha ha.

People frequently misunderstand the question about power supply. Internal component quality > wattage, not the other way around.
Either way, you're fine in this department: http://www.jonnyguru.com/blog/2015/11/09/evga-850-gq-850w-power-supply/6/

 

[grebgonebad]

People frequently misunderstand the question about power supply. Internal component quality > wattage, not the other way around.
Either way, you're fine in this department: http://www.jonnyguru.com/blog/2015/11/09/evga-850-gq-850w-power-supply/6/

Ha ha, yeah I understand. I never skimp out on the PSU though, had my fair share of headaches with sub standard, unbranded hunks of nightmare material in my time. I always stick to brands that I trust and have used for years. =)

Any idea's as to what could be causing the lower bench results then after the CPU overclock? It's the first time I've come across this particular issue and I'm honestly stumped as to what could be causing it?

 

[Phaaze88]

I don't really know, but knowing the relationship between the cpu and gpu... the OC may not be actually stable, but then again you've only run 3 samples?
Is it the same, back to back?

 

[grebgonebad]

I don't really know, but knowing the relationship between the cpu and gpu... the OC may not be actually stable, but then again you've only run 3 samples?
Is it the same, back to back?

Sorry, I probably should have clarified in my original post. The 3 samples that I provided were for context, I ran multiple tests across a range of clocks and voltages and the result is pretty predictable. As soon as I breath on the CPU clocks the GPU performance tanks (according to the bench at least).

I can't imagine it being a bottleneck issue across any aspect of the system, though as I stated originally this is my first time playing with Ryzen so I may admittedly be simply uneducated with the nuances of this specific architecture.

To be brutally honest I'm not that bothered about OC'ing this system anyway as it's an xmas present for my nephew who lives across the country from me, so I'll just be setting everything back to stock before I send it his way anyway. I'm only tinkering now out of morbid curiosity. Ha ha.

 

[Phaaze88]

Try working your way up to 4.0ghz - scores at 3.7, 3.8, 3.9 - and see how it behaves along the way.
If the cpu OC isn't stable, any hiccups are going to impact the gpu too.

 

[grebgonebad]

Try working your way up to 4.0ghz - scores at 3.7, 3.8, 3.9 - and see how it behaves along the way.
If the cpu OC isn't stable, any hiccups are going to impact the gpu too.

Yeah I was thinking about giving that a go tomorrow. Maybe a good nights sleep will help clear my head too. Ha ha.

I'll have another play around tomorrow and try increasing the clocks slowly and see how it goes. I'll post back here with an update once I've got some more solid data to go off.

 

[grebgonebad]

Update:

Spent an hour or so running a few more tests after work, not conclusive by any means, I definitely still want to run more. Also I've swapped out the 970 for a 1060 just out of curiosity to see if I get similar results with a different GPU. All other aspects of the system remain unchanged.

Average CPU temps are 36C and average GPU temps are 50C. It's pretty hot over here right now, that coupled with my small office means it gets pretty toasty in here. I powered up and let the air temp in the room level out before conducting the tests. Anyhoo, here's what I've got so far results wise. Results are arranged as follows:

Bench score>average temp reached>average wattage reached

Reboot

Clocks as follows:

Memory: 2133mhz
CPU: 3.6ghz 1.375V
GPU: Core 1885mhz Mem 4006mhz 1.062V Power Target 100% Temp Target 83C

1st run

Mem: 2173
CPU: 12448 54C 70W
GPU: 10133 62C

2nd run

Mem: 2180
CPU: 12481 52C 70W
GPU: 9172 61C

3rd run

Mem: 2182
CPU: 12501 53C 70W
GPU: 10089 61C

Reboot

Clocks as follows (Only GPU Temp & Power Targets):

Memory: 2133mhz
CPU: 3.6ghz 1.375V
GPU: Core 1885mhz Mem 4006mhz 1.062V Power Target 116% Temp Target 92C

1st run

Mem: 2168
CPU: 12601 53C 70W
GPU: 10055 61C

2nd run

Mem: 2164
CPU: 12638 62C 70W
GPU: 9164 62C

3rd run

Mem: 2200
CPU: 12615 53C 70W
GPU: 9174 62C

Reboot

Clocks as follows (Only CPU Clock):

Memory: 2133mhz
CPU: 3.8ghz 1.375V
GPU: Core 1885mhz Mem 4006mhz 1.062V Power Target 100% Temp Target 83C

1st run

Mem: 2156
CPU: 12780 73C 90W
GPU: 9276 61C

2nd run

Mem: 2156
CPU: 12768 73C 90W
GPU: 9771 61C

3rd run

Mem: 2183
CPU: 12754 73C 90W
GPU: 9213 60C


The mystery continues. Some anomalous results, not sure what to make of them. Though it is still an admittedly limited sample size so perhaps a more vigorous testing methodology is needed. What does surprise me is just how much the CPU temp and power draw went up after only bumping the multiplier up a couple notches. I haven't even touched the voltages yet. I wouldn't have thought that a simple .2ghz bump would result in a 20C rise in temps and a 20W jump in draw?

Anyway, I'll get around to running some more tests later, but for now that's all I've got to go on. Any thoughts?

 

[Phaaze88]

Looks better overall when you don't touch anything.
Power consumption is the only thing that comes to mind; neither device seems to care for the higher power consumption from pushing them.

 

[grebgonebad]

Yeah that was my initial thought. I'm still interested to find out why though. Things just aren't adding up. Plus I've got a bit of an addictive personality, now I've got a puzzle that I need to make sense of I'm kind of eager to try and solve it before I can be totally satisfied. Ha ha.

 

[Phaaze88]

The gpu one I get: You've raised the power limits, sure, but you've also increased it's operating temps and voltage draw. IF the card is hitting the voltage limit - or any limit for that matter - more often, then it's having to throttle back more frequently too, and that'll be comparable to stock settings, or even worse.

The cpu may be following something similar, but I'm not too sure.

 

[grebgonebad]

The gpu one I get: You've raised the power limits, sure, but you've also increased it's operating temps and voltage draw. IF the card is hitting the voltage limit - or any limit for that matter - more often, then it's having to throttle back more frequently too, and that'll be comparable to stock settings, or even worse.

The cpu may be following something similar, but I'm not too sure.

Interesting, but what about the CPU temps and power draw shooting up by so much after what is a pretty menial overclock? That definitely doesn't make sense to me?

 

[Phaaze88]

That one would be over my head. All I could tell you is that the voltage/frequency curve isn't linear, and overclocking anything discards power efficiency entirely. The only way to keep power in check is to use less voltage as far as is reasonably stable.

 

[grebgonebad]

Sorry for the hideously late reply, I've been busy preparing for the weekend. Ha ha.

So basically I swapped the EVGA 850 GQ (850W, obviously) for a Corsair VS550 (550W, again, obviously) and after running the same bench as before I'm pretty much getting the same results as before. So I'm certain that the issue doesn't lie with the PSU quality/capacity.

So once again I have to ask, why in the hell are the scores so different once I breath on the CPU clocks? I did try bumping the Mem clocks up to 2800mhz which made no difference to either score (predictable), but as soon as I touched the CPU multiplier and upped it (Only went as far as 3.7ghz this time around) the CPU Wattage and Temps rose dramatically.

Honestly, I spent years OC'ing, I only stopped messing with PC's around 2016, have things really changed that much in 4 years, or am I making a rookie mistake here?