MSI GTX 1070 low score (particularly 3D DX9)

[phlupple]

Hello,

First post here and I'm both new to the PerformanceTest software and not especially tech-savvy so please bear with me. I've had this MSI GTX 1070 for about two weeks (still stock) and I'm concerned about my benchmarking results. The G3D baseline I submitted when running my CPU at stock 3.3GHz was 9045 and clocking up to 4.3GHz increased it by about 400 IIRC.

My specs:
Operating System: Windows 7 Home Premium Edition Service Pack 1 build 7601 (64-bit)
CPU Type: Intel Core i5-2500K @ 3.30GHz (currently OC'd to 4.3GHz)
Number of CPUs: 1
Cores per CPU: 4
Hyperthreading: Disabled
Motherboard: P67A-UD4-B3
Memory: 16GB Kingston DDR3 SDRAM
Videocard: NVIDIA GeForce GTX 1070
Videocard: NVIDIA GeForce GTX 1070

• I don't know why it's listing two 1070s. I have one installed and the other PCI-E port is empty, although I did recently swap it from one to the other to see if that would impact my scores (it didn't, or hardly at all). My only other filled PCI port is a sound card.
• Comparing to 5 other similar baselines (same CPU & GPU, with comparable turbospeed clock) it's my DX9 performance that really lags. The average of the other 5 I'm comparing to is about 185FPS, while mine is 119.
• I've checked and I am running the latest version of DirectX.
• I've checked and I am running the latest 1070 drivers.
• I've uninstalled GeForce Experience and all other GeForce software aside from the drivers.
• I've tried running the tests on a dual-display setup with a 42" TV @ 60Hz and a 24" monitor @ 59Hz, and I've tried both as single displays and there's no change.
• I have quite an old 700w PSU and had to use a two-part 6+2 PCI-E cable to plug into the 8-pin socket on the 1070 (my previous card was a 6-pin).
• Using GPU-Z, my core and memory clocks seem to be as advertised.
• I've compared overall 3D scores with a friend who also recently bought a 1070 (EVGA I think). He's running Windows 7 with an i7 @ 3.3GHz stock. Our GPU-Z clocks were almost identical but his PerformanceTest G3D score was around 2-2.5k higher than mine.

Also:
I think I could theoretically return/exchange/refund the card with minimal hassle if I act relatively soon.

Sorry if I have failed to include some vital detail. I've looked over other similar threads and tried to provide everything needed. I have the trial version of PerformanceTest so I don't think I can screenshot my results?

Thanks a lot,
Phil

[David (PassMark)]

I am running the latest version of DirectX

We actually you aren't. Win10 includes DX12 which isn't available for Win7. But this isn't going to effect the DirectX9 result, which is, as you say, slightly low.

I think it unlikley a hardware fault. There are many hardware faults that cause a card to run perfectly fine, but slightly low in one particular test. So more likely it is some configuration issue (driver settings, driver version, 3rd party software, BIOS, PCIe bus speeds, thermal throttling).

Thermal throttling is fairly common on overclocked systems.

See also this post about driver power setting and quality settings. The GTX 1060 we tested bizarrely actually got faster when lower performance (and better quality) settings were selected in the driver for DX9. We assume this effect the same for the GTX 1080

[phlupple]

This is the BIOS download page for my mobo. I am currently running F3, which is pretty old now. Is it worth installing a more recent BIOS? If so, should I just go for the latest one?

The 1070 is plugged in to a PCI 2.0 x16 slot. From what I've read I got the impression it makes minimal difference if a card is slotted into PCI 2.0 or 3.0. Is that true? Could that be the source of the low DX9 FPS?

Is there an easy way to check if thermal throttling could be the culprit? I'm not sure what to look for in the BIOS (assuming that's where I'd find the setting?).

I also noticed that my hyperthreading is disabled. Is that relevant?

[David (PassMark)]

Yes, upgrade the BIOS. Unlikely it will change much, but there is a chance.

PCIe 3.0 is obviously better as that was what the GTX 1070 is designed to support. I agree there shouldn't be much difference, if they are both in x16 slots. I've seen a few people claiming there shouldn't be much difference, but it would be nice to see some actual benchmarks just to confirm that is really the case with the 1060 / 1070 / 1080.

For throttling, you need to check the CPU and GPU temps while the system is max load (so not in BIOS)

[phlupple]

Here's a GPU-Z screenshot after running a Unigine Heaven benchmark. Should I do the same with CPU-Z, or will the Heaven benchmark not be an appropriate load for the CPU?

[phlupple]

I followed the suggestions in the power & quality settings thread you linked and saw no change in the DX9 FPS stress test.

I also noticed that my hyperthreading is disabled. Could that be a contributing factor?