Lower 3DMark scores after overclocking

[TheRealBeandip]

Im using 2 way SLI EVGA GTX Titan X Superclocked.

Ive added +100 offset to the clock and +400 to the memory clock. Ive noticed lower scores in 3dmark now.
But the overclock isnt artficating and is completely stable.

Any thoughts on this?

 

[skit75]

Maybe the card(s) are thermal throttling? Are you monitoring temps or just going by visual reference?

 

[i7Baby]

Your CPU is bottlenecking.

Your PSU is poor quality.

 

[skit75]

Your CPU is bottlenecking.

Your PSU is poor quality.

How can you know that?

OP: Get some baseline numbers on your rig.

Those cards top out at what 83C? I imagine you are getting to max temp rather quickly with that overclock and then they are throttling themselves causing the lower numbers. You might just need to increase the attack slope of your fan curve and drop the OC numbers a little bit.

 

[i7Baby]

Without info, you have to guess. He hasn't even posted his build.

 

[TheRealBeandip]

Ive got both watercooled.

Max temp was 51 celcius.

Its a LEPA 12 watt gold power supply

AMD FX9590 CPU (will upgrade to skylake)

CPU max temp was 50 celcius

Its reporting a memory frequency of 3998mhz and a core clock of ~1400 mhz during benchmarks.

 

[i7Baby]

Post your complete and detailed build.

 

[skit75]

Sounds like a bottleneck to me then. Can't speculate much more with the limited information.

 

[TheRealBeandip]

Motherboard: ASUS Crosshair V Formula Z
CPU: AMD FX9590
Graphics cards: 2 way SLI EVGA GTX Titan X SC
RAM: 16GB DDR3 1866 Corsair Vengeance (Dual Channel)
Power supply: LEPA 1200w 80+ Gold
OS: Windows 8.1 Pro 64 bit
Storage: OCZ Vector 180 (960 GB), Toshiba 7200 rpm HDD (500GB), ADATA SP900 (512GB and boot drive)
Case: Corsair Obsidian 750D

CPU and both graphics cards are watercooled.
1 360mm radiator
2 240mm radiator
All Noctua NF-F12 fans running at ~1300 rpm

 

[damric]

Maybe the card(s) are thermal throttling? Are you monitoring temps or just going by visual reference?

^^ This is a possibility but only on the core.

Taking VRAM too far also causes FPS to go down as it causes more errors that need to be corrected.

The root of this whole problem is that the OP dialed in a seemingly random overclock amount instead of taking things up gently and testing as he went along.

Also, OC one thing at a time. Start with the core OC. Find your sweet spot, then OC your VRAM.

 

[TheRealBeandip]

Maybe the card(s) are thermal throttling? Are you monitoring temps or just going by visual reference?

^^ This is a possibility but only on the core.

Taking VRAM too far also causes FPS to go down as it causes more errors that need to be corrected.

The root of this whole problem is that the OP dialed in a seemingly random overclock amount instead of taking things up gently and testing as he went along.

Also, OC one thing at a time. Start with the core OC. Find your sweet spot, then OC your VRAM.

Nope, I moved up carefully and got to this point. There was no just random [edited by damric] with numbers. I noticed lower scores throughout. but theyve been consistent. All overclocked scores are roughly the same, and then the stock scores are higher than the overclocked scores.

 

[TheRealBeandip]

http://pcpartpicker.com/p/xjd7FT

Heres the full build minus the watercooling.

 

[skit75]

Seems like a pretty sweet rig. Maybe you've found your glass ceiling and there is no more overclocking overhead. Welcome to diminishing returns land.

 

[TheRealBeandip]

Thanks, Ive certainly enjoyed the rig. Im aware of the glass cieling and diminishing returns, but wouldnt results then flatten out as opposed to reduce?

 

[skit75]

The GTX 980 Ti review hinted at better overclockability compared to the Titan X. I forget what specifically the author referenced in citing that but I remember reading it. Are you developing with the Titans or just gaming? You could always switch to 980Ti's if you aren't getting Titan satisfaction. Have you been able to saturate > 6GB VRAM in your setup?

 

[TheRealBeandip]

im plenty happy with the Titans. i just wanted to see what i could get out of them. Im easily using 6GB of VRAM. I dont do any developing but the Titan X isnt a double precision card like previous titans. Its got worse double precision than previous titans actually. The Titan X is a gaming card and outperforms the 980ti in most situations. The 12GB of VRAM is nice due to the 7680x1440 resolution during games and it translates to easy 4k once i make that switch.

the 980ti outperforms the Titan X in some situations due to its boost clock. The 980ti will boost more than the titan X, even if the Titan X has both power and thermal headroom.

Ive done some testing with the 980tis and found the Titan Xs to give me better performance.

 

[TheRealBeandip]

You guys think 1476mhz boost clock and 7880mhz mem clock is a decent overclock for a titan X?

I havent optimized voltage or power yet. Im considering BIOS flashing with an aftermarket BIOS though.

 

[mp5mafia]

You guys think 1476mhz boost clock and 7880mhz mem clock is a decent overclock for a titan X?

I havent optimized voltage or power yet. Im considering BIOS flashing with an aftermarket BIOS though.

You need to run gpu-z during your 3d mark test and drop us logfile or at least get a screen shot of sensors readings so we can get an idea of what is happening.
If I was to guess without any data I would say it throttles due to either power limit or voltage but can't tell until we see your gpuz data especially Perfcap reason though

 

[TheRealBeandip]

https://mega.co.nz/#!BhIDST6b!6e5q-p8ejWjU3fi_nvF1Y8ob3Jsn97l5L1MkQFpIgLE

GPU-Z sensor log.

I ran 3dmark Firestrike UItra with demo. my lowest score yet as well, 5700. It averages around 6200-6400

Also sorry moderator if posting that link is bad, Couldnt think of another way to provide the logfile.

 

[mp5mafia]

https://mega.co.nz/#!BhIDST6b!6e5q-p8ejWjU3fi_nvF1Y8ob3Jsn97l5L1MkQFpIgLE

GPU-Z sensor log.

I ran 3dmark Firestrike UItra with demo. my lowest score yet as well, 5700. It averages around 6200-6400

Also sorry moderator if posting that link is bad, Couldnt think of another way to provide the logfile.

Wow, I see gpuz does not do great with logfile in sli mode haha
It only gives core clock readings for 1 card but it looks like perfcap is multiplied :/
Can you drop a screen shot when you get time just to verify I read this log right?
Either way it seems you hit power limit couple times , nothing extreme but still.
You should drop your settings a bit to make sure you stay under power limit.

 

[4Ryan6]

http://pcpartpicker.com/p/xjd7FT

Heres the full build minus the watercooling.

Sorry, I have to have the Wow! moment here, but it's more because you invested $2,060 bucks in GTX Titan X cards and decided to power it all with a $250 AMD FX-9590? (One of the hottest under load CPUs on the planet!)

http://www.hardwarecanucks.com/foru...66-amd-fx-9590-review-piledriver-5ghz-17.html

Benchmark performance fall off is usually because the CPU is dropping the ball before the goal line, getting too hot and throttling back to protect itself.

Temperature wise you would have been better off running two loops and isolating CPU and GPU temperatures from each other.

That way you could keep the CPU temperature as low as possible as that is more critical than the combined GPU temperatures, as water cooling is much more effective on cooling the GPUs under load.

So even thought you may think your CPU temperature is OK, they thought the same in the link above but note the Temperature section and the CPU performance fall off.

 

[TheRealBeandip]

http://pcpartpicker.com/p/xjd7FT

Heres the full build minus the watercooling.

Sorry, I have to have the Wow! moment here, but it's more because you invested $2,060 bucks in GTX Titan X cards and decided to power it all with a $250 AMD FX-9590? (One of the hottest under load CPUs on the planet!)

http://www.hardwarecanucks.com/foru...66-amd-fx-9590-review-piledriver-5ghz-17.html

Benchmark performance fall off is usually because the CPU is dropping the ball before the goal line, getting too hot and throttling back to protect itself.

Temperature wise you would have been better off running two loops and isolating CPU and GPU temperatures from each other.

That way you could keep the CPU temperature as low as possible as that is more critical than the combined GPU temperatures, as water cooling is much more effective on cooling the GPUs under load.

So even thought you may think your CPU temperature is OK, they thought the same in the link above but note the Temperature section and the CPU performance fall off.

Im aware of the FX9590s temp problems and powe usage. It is a remnant from my last build and is only temporary. It will soon be replaced with an intel skylake i7 or some other intel CPU. Also I heavily considered a separate loop for the graphics cards but im working in a 750D, so I ran out of room. But soon I may be adding an external 3x3 120mm radiator.

 

[i7Baby]

Check your CPU temps with AMD Overdrive.

But I think Ryan could be right with your CPU. AMD recommend at least 1000W PSU with that CPU. Plus 2 x Titans on top of that. I'd recommend looking at a bigger PSU.

 

[TheRealBeandip]

Ive never seen AMD recommend a 1000 watt CPU with the 9590. And the recommended supply for a Titan X is 600, im running 1200 but planned to get a 1600 soon.

As for the CPU temps I have a thermal margin at load of 20-30 degrees. unfortunately overdrive doesnt give an actual core temp.

Im just waiting for skylake then ill be ditching this CPU.

 

[i7Baby]

https://www.pugetsystems.com/parts/CPU/AMD-FX-9590-4-7GHz-220W-9542
"This level of heat requires liquid-cooling, and AMD also recommends using 1000W or higher power supplies to handle the added electrical draw."

This is also from SR71 Blackbird. But I can't find it on amd.com. I've also come across 700-1000W

Yes - hoperfully Skylake will be worth waiting for.
http://wccftech.com/intels-6th-generation-skylake-processors-scheduled-2h-2015-5th-generation-broadwell-spring-15-updates-2015-2016-mobility-roadmap/ -
"we know that the Skylake processors would be compatible with the latest LGA 1151 socketed boards and Z170 chipset which is part of the new 100-Series chipset replacing the 9-Series “Wild Cat” PCH. The second most interesting thing covered is that Skylake processors would have both DDR3 and DDR4 memory controllers so the different SKUs can be configured to use either of the memory type."