Very low benchmark performance - Suggestions?

B

byusinger84

Gawd
Joined
Feb 1, 2008
Messages
807
**** UPDATE 5-20-18 ****

Turns out my issue was with my memory kit. Once I went from 64 GB of RAM to 32 GB of ram, I could run the RAM at the XMP settings and all of my slowness issues disappeared. Everything is running as it should. Guess I probably should have tried that earlier...ha.

******************************


While benchmarks aren't everything, they clearly can help know if your system is running correctly or not.

This past weekend I worked on a project some of my friends and I are doing benchmarking the various systems that we have at home and compiling the results into a Google Doc. The results for the most part are what you would expect, however, my main gaming rig has some major issues, at least based on the benchmark results.

My system has the following specs:

Ryzen 1700X running at 3.9 at 1.45v
ASRock X370 Fatal1ty X370 Professional Gaming
Corsair DDR4 4x16 GB running at 2666
EVGA 1080 Ti FTW3
Corsair HX1050W PSU

The system I'm comparing to, is very similar:

Core i7 7700K at stock
ASRock Fatal1ty Z270 Gaming K6
Corsair DDR4 2x8 GB running at 2400
EVGA 1080 Ti FTW3
Corsair 750W PSU

My friend's system very handily beats mine at 1080p (which isn't a surprise given how Ryzen performs with 1080p) but both of us game at 1440p. Based on reviews I see online, both systems should perform very close for the most part at 1440p but our results show at least a 15-30% delta in most cases.

Any thoughts as to why?

If you need more info, let me know. Thanks for your help in advance!
 
Last edited:
What games?

Assuming everything is working properly, older games would likely favor the i7 to a larger degree than more modern games. This is both because they tend to be more lightly threaded, and because they simply predate the Zen architecture and may not play as nicely with it.

It's also possible your CPU is thermally throttling itself given your overclock. You should benchmark your overclocked results against your own system when it isn't overclocked, and make sure you're getting the benefit you should.

I don't have a Ryzen system, but from what I understand they appreciate speedy memory, as opposed to the i7 which mostly doesn't care. You have 4 sticks, which is a bit harder, and you're only running it at 2666. You might try playing around here, make sure you're using the XMP timings for your memory and such.

Lastly, the Ryzen isn't a poor performer at 1080p compared to the i7; in actuality, at 1080p you are CPU limited for many games, which vanishes at higher resolutions when you become GPU limited. However, you are basically running the fastest GPU out there, and you may still be CPU limited at 1440p for many titles, allowing the i7 to stretch its legs a bit more. If you bump up to 4K, I would assume the difference between the platforms would vanish. The important part is that there isn't anything magical that makes the i7 better at 1080p than the Ryzen; it's just a faster CPU clock-for-clock at lower thread counts, so whenever a game utilizes 4 or fewer CPU cores and you are not GPU limited, then the i7 will win. As games expand to use 6+ CPU cores, the i7 will start to lose ground as the higher core count on Zen will give it an advantage to offset the higher clockspeed and per-clock performance of the i7.
 
The previous comment touched on it but when I get very low scores on a graphics bench it's because my GPU core or memory clock is too high not the CPU. Neither of those CPUs would bottle neck any graphics bench I'm aware of. This is why we work clocks up slowly and bench as we go with each step up. Always work core clock up 1st when you hit it's max typically your first clue is the score is lower in spite of the higher clock. OC memory 2nd by backing off your max core clock, again benching as you go when score falls you back mem oc off a bit. Then do some on line gaming because it is typically more demanding and will tax you OC to the max. and in many cases necessitate lowering clocks again. Pick one graphics test like Superposition 8K optimized or 3DMark DX12. It should go without saying but you should always do the same bench at all stock clocks CPU & GPU as a base line and make sure it is performing as well as other systems with similar specs. This will reveal issues that need to be resolved before overclocking anything. PCMark 8 or 10 is a good tool for that or User Bench. Both of those can be quite long so be prepared for that. The two systems in the OP should have similar performance with the Intel having a slight edge in Graphics test especially as sinisterDei described at 1080p.
 
My first thought - holy crap that's a lot of voltage for 3.9. Have you checked if you are throttling? I would guess your VRM's are roasting.
 
Thank you for your replies. I will try to explain more so that we're on the same page:

LigTasm - Yeah, I know it's a little high, but the system is stable, doesn't appear to throttle at all as CPU temps stay under 60c on load (the CPU is on a water loop by itself). Going to 1.4v it's mostly stable but under certain games it would freeze, hence the higher voltage. That being said, I've read a lot about Ryzen overclocking ever since the CPU came out and I have tried every combination I can think of to get the CPU to run with less voltage. AMD states that 1.45v is the maximum safe voltage for overclocking. I know 1.35 is the recommended 24/7 setting, but my CPU simply will not clock at 3.8 GHz at 1.35v and be stable. If you know something I don't know, I'd love to hear it. Also, from what I understand the VRMs on my motherboard are somewhat overkill and according to what I've read are fully capable of supporting what I am doing. Again, let me know if you think I'm wrong here.

owcraftsman Thank you for your input. I don't think I was clear enough, but I actually did run everything at stock first. That's when I first realized there was a problem (see below). My GPU is not overclocked, however my CPU is, as that appeared to be the bottleneck. OCing the CPU helped get my scores closer to where they should be based on similar systems but from what I can tell still fall very much short.

sinisterDei I am not benchmarking with any games, per say. Most tests were done with things like Cinebench, Novabench, 3D Mark (Time Spy and Fire Strike), Unigine Heaven and Superposition.

Cinebench was mostly for the CPU, but the OpenGL test was extremely low. My GTX 770 in an Intel system beat that test which is what first indicated to me that there was a problem. Looking online at other users with Ryzen 1700X systems at similar clocks and with GTX 1080 Ti's, it was clear to me that my system was indeed under performing in a major way as I was comparing essentially apples to apples. Note that this was all done at stock speeds. I had not at this point overclocked the CPU in any way. I had a fresh install of Windows, the AMD chipset drivers, and I was utilizing the AMD Ryzen power plan. I also tried on High Performance.

I then moved on to other benchmarks thinking it must be a fluke, however test after test indicated to me that I was severely under performing my Intel counterpart in my original post and when comparing to other Ryzen systems with nearly identical specs and clocks. I tested at 1080p and 1440p to be sure I covered all my bases and I can only conclude that something somewhere is messed up.

Below is an example of what I am seeing. As you can see the clocks are essentially identical:

unknown.png


This isn't the only example, but one that was right in line with everything else I've been seeing.
 
If your apples-to-apples performance at stock compared to equivalent systems is poor, then that is definitely troubling.

I would be looking at your memory; it is *not* on the QVL for the board you own (no 16 GB Corsair sticks are on the QVL for your board). Additionally though your motherboard claims support for OCing memory to DDR4 3200 its official support is significantly slower. And additionally, the board tops out at 2133-2400 if you are using 4 sticks instead of 2 to get where you're going. You might try removing the memory from slots A1 and B1, leaving you with 2x 16GB sticks installed in A2/B2 and run your benchmarks that way. The memory frequency and installation chart is on the ASRock page - http://www.asrock.com/mb/AMD/Fatal1ty X370 Professional Gaming/#Specification. You'll also want to make sure you're at least on BIOS version 3.00 on the board (released on 7/14/2017) because that's the last BIOS update to improve memory compatibility on that board, though they've had two further updates since then fixing other issues.
 
LigTasm said:
My first thought - holy crap that's a lot of voltage for 3.9. Have you checked if you are throttling? I would guess your VRM's are roasting.
Click to expand...

OMG, please this. That is WAY high for just 3.9GHz. I got my cousin's computer's 1600X to 4.0 @ 1.375. I could see 1.40V but 1.45 is overkill. Go and touch your VRMs. They will burn your hand. Not fuckin' kidding you. The Ryzen motherboard's VRM cooling are a joke.
 
sinisterDei said:
If your apples-to-apples performance at stock compared to equivalent systems is poor, then that is definitely troubling.

I would be looking at your memory; it is *not* on the QVL for the board you own (no 16 GB Corsair sticks are on the QVL for your board). Additionally though your motherboard claims support for OCing memory to DDR4 3200 its official support is significantly slower. And additionally, the board tops out at 2133-2400 if you are using 4 sticks instead of 2 to get where you're going. You might try removing the memory from slots A1 and B1, leaving you with 2x 16GB sticks installed in A2/B2 and run your benchmarks that way. The memory frequency and installation chart is on the ASRock page - http://www.asrock.com/mb/AMD/Fatal1ty X370 Professional Gaming/#Specification. You'll also want to make sure you're at least on BIOS version 3.00 on the board (released on 7/14/2017) because that's the last BIOS update to improve memory compatibility on that board, though they've had two further updates since then fixing other issues.
Click to expand...

His memory works fine, otherwise his computer would be crashing.
 
I think youre seeing the intel vs amd ipc difference. and yes the others are right, that's too much voltage for 3.9.
 
I leave my 1700x at 3.8ghz with a 1.135 vcore and it is running great could probably get 4ghz with my crosshair 6 and monoblock but see no real need.. If the benchmarks are what your wanting to impove on I would definitely recommend getting 3200 ddr4 tho or 3600 but run it at 3200 with lower latencies.
 
The 1700X should shine with Time Spy pure DX12 easily beating the 7700k in the CPU test. It is not your bottleneck with that test. You need to get your default clocks performing as they should first. It could be bios or microcode update or chipset driver or memory as suggested above for sure it's performing at -9.8% of the comparison you posted. More concerning is your GPU is -11.5% on all graphics test. I see you have two FTW3 try the other one in the AMD system and see how it does. Test your memory and compare to other systems w/similar probably best to use Si Soft Sandra for comparison sake. And absolutely check for bios/micocode updates and driver updates even beta where available what ever it take to get your stock clock system performing in the norm.
 
Gasaraki_ said:
His memory works fine, otherwise his computer would be crashing.
Click to expand...

I was thinking more along the line that in order for his memory to be in and working fine, it may be running significantly under-spec. Obviously it's 'working' or as you said, there would be mass instability. But he could lose a lot of performance because the memory is running at extremely loose timings or significantly slower clockspeeds. His first post says it's running at 2666, his sig says it's 3000 @ C15, so I feel it's at least an avenue worth exploring, since he also stated that his performance problems began at stock speeds for the CPU, which presumably predates the ridiculous voltage.
 
sinisterDei Yeah, I did make my timings much tighter when running at 2666. It says 3000 on my sig because that's their spec, but obviously they don't run at that speed with 4 sticks on this platform.

Gasaraki_ VRM's are actually cool to the touch. :: shrugs :: So I don't know what to tell you. :) I get that people say that's too high but...that's what my system seems to need and it's ok with it.

Here are my temps after running Time Spy

XD6s01c.png
 
Also, my BIOS is fully up to date as are all of my drivers. I did that way before my first post. This isn't my first rodeo. :)
 
just a quick shot.
Have you analyzed if you are hitting a corespeed bottleneck ?
 
Just a quick note, and not trying to be insulting:

Your VRMs should be at least pretty warm under load. Are you sure your checking the right thing?

None of those temps listed on HWMonitor, I believe, actually measure VRM temp.

Your right that CPU temps and watts look pretty good. I would have expected your cpu watts and temps to be a lot higher than what it’s showing there.

Don’t have a definitive answer, but I agree that something doesn’t seem quite right
 
Brian_B You're good. I don't take offense at all. We're all trying to help each other out here right? :) I guess I should say, that they were warm but I could easily touch them (at idle). I didn't touch them at load.

To everyone else: I was able to redo my overclock and get to 3.9 GHz at 1.375v. I had forgotten about the Load Line Calibration setting which is why I needed such high voltage to keep it stable. I set LLC to Level 1 (the highest) and everything is stable. That said, it didn't improve any benchmarks, but did help my CPU temps a little bit.
 
Welp, I'm pretty confused then. If you're still losing by 10-15% in apples/apples comparison at stock speeds to identical hardware, then I'd be swapping out hardware starting with your motherboard. Unfortunately that may not be an option if you don't have another Ryzen board handy. We've definitely moved into the territory that I would call "guess & check troubleshooting"
 
I know this is an old thread, but since I'm the OP, I figured I'd update everyone.

Turns out my issue was with my memory kit. Once I went from 64 GB of RAM to 32 GB of ram, I could run the RAM at the XMP settings and all of my slowness issues disappeared. Everything is running as it should. Guess I probably should have tried that earlier...ha.
 
Just want to clarify for my own knowledge - was it that half the memory was bad, or was it just caused by the lower timings when you use 4 sticks?
 
It wasn't bad, but I had to lower my speed down from 3000 to 2666 which for some reason created a 10% overall performance penalty.

My guess is that the memory controller was having some really big latency issues with 4 sticks as opposed to two.
 
You must log in or register to reply here.
Back
Top