Question Ryzen 7 1700X Overclock: Decreased Performance!?

davew1860

Prominent
Feb 6, 2019
145
29
620
I increased my base speed from the stock 3.4GHz to 3.8GHz. I also disabled core performance boost. Nothing else was changed. I left the voltages at auto.

The OC appears to be stable. I have run Prime95 several times, all without crashes, including a few sessions of "Small FFTs" for 10 minutes each time and a couple sessions of "Blend" for one hour each time.

My CPU temperature stays below 80 C during these stress tests.

Next I ran almost identical test runs in Battlefield V, one before the OC and one after the OC. Before the OC, my average FPS was 126. After the OC, my average FPS was only 122!? Hence, I actually lost 4 average FPS!?

Please advise.

-------------------------------------------------------------
CPU: AMD Ryzen 7 1700X
Motherboard: ASUS ROG Strix B450-F Gaming
Video card: SAPPHIRE Radeon RX Vega 64 8GB (with 19.4.1 driver version)
RAM: G.SKILL TridentZ RGB Series 16GB
SSD's: SanDisk Ultra 3D SSD 512GB / Samsung SSD 860 EVO 250GB
HDD's: Western Digital Black 7200 RPM 1TB / Seagate BarraCuda 7200 RPM 250GB
PSU: CORSAIR RM850x White 850W
 
What speed are your memory modules, and which slots are they installed in? Could be relevant. Probably isn't, but it's possible, so would be good to know.

Is your BIOS up to date with the MOST recent stable BIOS version?

Just because your core temps remain below 80 degrees doesn't mean your VRMs are not throttling, and that can cause a definite loss of performance. I'd download and run HWinfo (NOT HWmonitor or ANY other utility) and look at the VRM thermals if they are listed. Choose the "Sensors only" option when you start up HWinfo. The Summary option is mostly useless.

Often I'd tend to say this might be a power delivery problem, and the overclock is taking the PSU past the point where it can adequately deliver the necessary power to either the graphics card or the CPU, but that RM unit is not only somewhat higher capacity than is strictly necessary, it's also a very good model, so unless it's a fairly old unit that's unlikely.
 
  • Like
Reactions: davew1860

davew1860

Prominent
Feb 6, 2019
145
29
620
What speed are your memory modules, and which slots are they installed in? Could be relevant. Probably isn't, but it's possible, so would be good to know.

Is your BIOS up to date with the MOST recent stable BIOS version?

Just because your core temps remain below 80 degrees doesn't mean your VRMs are not throttling, and that can cause a definite loss of performance. I'd download and run HWinfo (NOT HWmonitor or ANY other utility) and look at the VRM thermals if they are listed. Choose the "Sensors only" option when you start up HWinfo. The Summary option is mostly useless.

Often I'd tend to say this might be a power delivery problem, and the overclock is taking the PSU past the point where it can adequately deliver the necessary power to either the graphics card or the CPU, but that RM unit is not only somewhat higher capacity than is strictly necessary, it's also a very good model, so unless it's a fairly old unit that's unlikely.

Memory modules are at 3200MHz and occupy slots #2 and #4.

I have the second most recent BIOS, which is version 2202 from April of this year. But I see there is now a new version from May 15th?

I do have HWiNFO64 but cannot find any readings for "VRM". Could it be named something else?

Thanks!
 

davew1860

Prominent
Feb 6, 2019
145
29
620
Here's some more data, and it just confuses me more:

I ran the OC test run in Battlefield V again. This time my average FPS rose to 123, which is still 3 average FPS less than WITHOUT the OC. So my results are reliable here but disappointing.

In the Heaven Benchmark, on the contrary, my average FPS rises by 2% WITH the OC. At least it is a performance gain, but it isn't much.
 

davew1860

Prominent
Feb 6, 2019
145
29
620
Have you monitored your per core frequencies, during testing, to see that the core clocks are actually staying at full speed where they are set?

I just ran the Heaven Benchmark and monitored.

All core clocks stayed between 3,791.6MHz and 3,793.0MHz.

The max CPU/thread usage varied from a minimum of 1.3% to a maximum of 95.0%. The average was 46.2%.

The total CPU usage varied from a minimum of 0.2% to a maximum of 29.5%. The average was 6.9%.

Does all of that seem normal?
 

Kurtisdede

Commendable
Jun 10, 2017
72
11
1,565
That is not a good way to benchmark anything. Games can vary in avg framerate depending on where you're looking at and stuff like that. You got less frames on your overclocked BF1 bench because you were doing something intensive.
You should test your cpu with something like Cinebench R15.
First test at stock, and then overclock, and re-test.
 
That is not a good way to benchmark anything. Games can vary in avg framerate depending on where you're looking at and stuff like that. You got less frames on your overclocked BF1 bench because you were doing something intensive.
You should test your cpu with something like Cinebench R15.
First test at stock, and then overclock, and re-test.

Not sure how you figure Heaven tests differently per run or varies in framerate due to "where you're looking at" since it's the same benchmark every time.

I will concede that if your system does not have automatic defragmentation, automatic cleanup and other automatic processes turned off then it's possible a windows process could be running to account for a difference in performance but in reality I think the difference you are seeing is so minimal as to not matter. ~4% is nominal, and while I agree that the overclocked value should be higher than the stock value, it could be anything really.

I would download Realbench, run the Realbench benchmark, and compare the two runs. Make sure you turn off automatic defragmentation and cleanup first, and any other automatic processes that are not necessary to be running, before doing the tests so that nothing skews the results.
 
  • Like
Reactions: davew1860

davew1860

Prominent
Feb 6, 2019
145
29
620
Not sure how you figure Heaven tests differently per run or varies in framerate due to "where you're looking at" since it's the same benchmark every time.

I will concede that if your system does not have automatic defragmentation, automatic cleanup and other automatic processes turned off then it's possible a windows process could be running to account for a difference in performance but in reality I think the difference you are seeing is so minimal as to not matter. ~4% is nominal, and while I agree that the overclocked value should be higher than the stock value, it could be anything really.

I would download Realbench, run the Realbench benchmark, and compare the two runs. Make sure you turn off automatic defragmentation and cleanup first, and any other automatic processes that are not necessary to be running, before doing the tests so that nothing skews the results.

I am going to run RealBench really soon. It is actually my next step. Thank you for that suggestion!

Update: I did change my core voltage from auto to 1.3375V. That should at least save me some electricity. I experienced crashes in Prime95 when the voltage was set below 1.3375V.

Here's my new order of testing:

1) Prime95 "Small FFTs" test for 15 minutes, with AVX disabled.

I actually just completed that test for thermal purposes, and my results were pretty good. My max CPU temp at 3.8GHz and 1.3375V was only 79.8 C. I consider that a passable performance.

My next steps are . . .

2) RealBench "Stress Test" for 8 hours, for system stability purposes.

3) Prime95 "Blend" test for 8 hours, for system stability purposes.

4) Heaven Benchmark, for my video card overclock.

5) Battlefield V test runs, for real-world purposes.

I will keep you posted about my results.

What do you think of my plan?
 

davew1860

Prominent
Feb 6, 2019
145
29
620
That is not a good way to benchmark anything. Games can vary in avg framerate depending on where you're looking at and stuff like that. You got less frames on your overclocked BF1 bench because you were doing something intensive.
You should test your cpu with something like Cinebench R15.
First test at stock, and then overclock, and re-test.

I agree that using BF V for benchmarking is not as reliable as, say, Cinebench, but it still is valid for measuring average FPS. If you play long enough and try to replicate the game's conditions as closely as possible across multiple test runs, then you can still get very useful real-world performance data.
 
If you haven't looked at this already, my recommendations for testing and validating an overclock are included here and are mostly in line with your premise that was outlined in your earlier post anyhow. There are some additional items of note included however including custom settings for Prime memory testing which could definitely be of importance on a Ryzen system since it can tend to be sketchy when it comes to memory stability.

Of MOST importance is to validate a solid CPU overclock FIRST, BEFORE you make any changes to the memory configuration from the default memory profile at 2133mhz, and THEN once you are satisfied that the CPU is as stable as it's going to get, THEN set your memory to the XMP or manual configuration you plan to run it with and test again for stability.

 
  • Like
Reactions: davew1860

davew1860

Prominent
Feb 6, 2019
145
29
620
If you haven't looked at this already, my recommendations for testing and validating an overclock are included here and are mostly in line with your premise that was outlined in your earlier post anyhow. There are some additional items of note included however including custom settings for Prime memory testing which could definitely be of importance on a Ryzen system since it can tend to be sketchy when it comes to memory stability.

Of MOST importance is to validate a solid CPU overclock FIRST, BEFORE you make any changes to the memory configuration from the default memory profile at 2133mhz, and THEN once you are satisfied that the CPU is as stable as it's going to get, THEN set your memory to the XMP or manual configuration you plan to run it with and test again for stability.


Oh, you were the one who wrote that super guide! Thank you for that service!

I'm guessing the reason for testing first at the base memory speed of 2133MHz is to eliminate the memory as being the instability culprit while testing the stability of the CPU?
 

TJ Hooker

Titan
Ambassador
I agree that using BF V for benchmarking is not as reliable as, say, Cinebench, but it still is valid for measuring average FPS. If you play long enough and try to replicate the game's conditions as closely as possible across multiple test runs, then you can still get very useful real-world performance data.
The difference you're seeing is only around 3%. That's probably within the margin of error of dedicated benchmarks that are scripted, let alone you just trying to do things as similarly as possibly from one run to the next. Meaning your before and after results can be treated as being more or less the same. And the reason they're the same is likely because your CPU can already boost up to 3.8 GHz for lightly threaded loads at stock settings, and/or you aren't CPU limited in BFV.

I'm not really seeing any issue here.
 
  • Like
Reactions: davew1860

davew1860

Prominent
Feb 6, 2019
145
29
620
The difference you're seeing is only around 3%. That's probably within the margin of error of dedicated benchmarks that are scripted, let alone you just trying to do things as similarly as possibly from one run to the next. Meaning your before and after results can be treated as being more or less the same. And the reason they're the same is likely because your CPU can already boost up to 3.8 GHz for lightly threaded loads at stock settings, and/or you aren't CPU limited in BFV.

I'm not really seeing any issue here.

I was actually wondering about how the core performance boost affects the game performance. Thanks!

So would you only expect to see significant average FPS gains from overclocks above that turbo boost speed of 3.8GHz?
 

TJ Hooker

Titan
Ambassador
Not unless you were CPU bottlenecked, i.e hitting 100% usage on one or more cores. Even then, most 1st gen Ryzen chips can only hit around 4.1 GHz at most, so best case scenario you'd only see an ~8% improvement compared to stock boost clocks (maybe a little more if the game was heavily threaded enough such that the CPU wasn't hitting it's 3.8 GHz boost).
 
  • Like
Reactions: davew1860

davew1860

Prominent
Feb 6, 2019
145
29
620
Not unless you were CPU bottlenecked, i.e hitting 100% usage on one or more cores. Even then, most 1st gen Ryzen chips can only hit around 4.1 GHz at most, so best case scenario you'd only see an ~8% improvement compared to stock boost clocks (maybe a little more if the game was heavily threaded enough such that the CPU wasn't hitting it's 3.8 GHz boost).

Ah, I see.

Well, I’m certainly not being held back by my CPU. It never even comes close to 100% utilization during gaming.

Personally, although I would consider 8% significant, I’m not sure 4.1GHz would be worth the extra voltage.

I will continue my testing, though, then see where I will go from there.

Thanks!
 

davew1860

Prominent
Feb 6, 2019
145
29
620
Well like I said, you'd only see fps scale linearly with CPU frequency if you were 100% CPU limited (and assume that CPU performance scales 100% with CPU frequency), which is not a realistic scenario.

Gotcha.

Even without significant performance gains, I find it fun to (intelligently) experiment with hardware.

Hopefully I will be able to share something of value after my stress testing and benchmarks are done.
 
That probably also GREATLY varies from game to game. In some games where you are not using much more than 40-50% peak CPU usage it would not be helpful however in any game where you DO see very high CPU usage, say 80-95% (Very unusual to ever see 100% CPU usage in an actual game, unless you are SERIOUSLY CPU limited) or in games where higher single core performance, per core, is beneficial, then you might see some small to moderate gains from a 4.1Ghz OC which is as TJH mentioned about an 8% increase.

We tend to take ANY increase that is free so long as stability is still maintained. Why? Because we can. We are enthusiasts, and that is what we do. LOL.

Consider, in my case on my current system for example. 6700k@4.7Ghz. Maximum stock boost speed is 4.2Ghz, so that's an 11.9% gain. Add to that the memory overclock to 3200mhz from 3000mhz, plus a drop in latency by tightening the timings down to 14-15-15-28 from 15-16-16-35, plus some adjustments to a few of the secondary timings which overall (A lot of work for little gain, but still, it's FREE gain, aside from my time) results in about a 1.5% improvement in performance. So in total, you have a 13.4% gain in performance, for free.


Funny thing is, people say well that's not very much, and yet, people will pay HUNDREDS and HUNDREDS of dollars to upgrade to a new system that offers maybe an 8-10% gain in IPC, if you're lucky, from one gen to the next. SO, I think in reality we take our digs where we can get them. Especially if it doesn't mean you have to buy a bunch of cooling hardware JUST to get that gain, because you already have that hardware OR needed it anyhow. Just my 2cents.
 
  • Like
Reactions: davew1860