MSI GT80 3dMark Scores, test and tune

Prema's VBIOS is based off Clevo card, and may not be up-to-date if they don't receive new VBIOS from Nvidia/Clevo and they will be outdated.

 

I have noticed with Premas bios a drop in temp. Only like 2-3 c underload, and a little less GPU utilization. but this was in BF4 . thats prettty much the only game i am playing right now other than PVZ-Garden Warfare. ill try and bench it tonight.

 

3Dmark results CPU/GPU tuned stable.
Stock BIOS E1812IMS.10F / vbios 84.04.48.00.29

Model MSI GT80 SLI-263
Test 3DMark
Thermal compound Stock
CPU Tuning 40x cores -20mV, 36x cache +0.9mV
GPU Tuning Stock vBios +110 Core, +400 Memory
Driver : 353.62
Shift Mode Dragon disabled/High Performance Power Plan/Cooling Fan on
Room/Ambient Temperature 70F

Stock settings:
http://www.3dmark.com/3dm/8171349


Tuned settings:
http://www.3dmark.com/3dm/8170934


Update: 3dmark Fire Strike 15909 41x +23mV 36x +0.9 GPU +135 +385
http://www.3dmark.com/fs/5703039


Update 2: 3dmark Firestrike 15911 41x +20mV 36x +0.9 GPU +135 +395
(didn't get uploaded)

 

vbios 84.04.48.00.29

Model MSI GT80 SLI-021
Test 3DMark
Thermal compound CLU on CPU & GPUs
CPU Tuning : stock
GPU Tuning Stock vBios
Driver : 353.62
Shift Mode High Performance Power Plan/Cooling Fan on
Room/Ambient Temperature 29°C

http://www.3dmark.com/3dm/8177940

 

vbios 84.04.48.00.29

Model MSI GT80 SLI-021
Test 3DMark
Thermal compound CLU on CPU & GPUs
CPU Tuning : 4.1Ghz -40mV Dynamic voltage, Processor cache ratio x37 -0.9mV
GPU Tuning Stock vBios @ +135MHz core , +200 memory
Driver : 353.62
Shift Mode High Performance Power Plan/Cooling Fan on
Room/Ambient Temperature 30°C

Max temps : CPU 89°C / GPUs 64°C & 80°C

http://www.3dmark.com/3dm/8178367

 

The performance with Intel Speed Step off, 2.9ghz base speed locked, retaining full GPU OC, results in the same/similar score at stock settings for CPU 38x/37x/37x/37x and GPU for 3dmark Fire Strike.

14429 with Intel SpeedStep disabled, 2.9ghz locked -65mV, 34x cache, GPU OC +135 +400 vs 14292 as posted above for stock speed 38x/37x/37x/37x 36x cache with stock GPU settings.

The benefit? Ultra low peak CPU package temperature: 61c !!

In my shock at the result, I forgot to upload it to Futuremark, but uploaded the next run, 14355:
http://www.3dmark.com/fs/5709128







It's possible the same result can be simulated with XTU setting of 29x on all 4 cores. It's a bit of a pain to need to reboot into the BIOS to disable/enable Intel Speed Step. I will update later after trying this.

FPS drops in games will vary depending on the importance of CPU in the game, initial quick looks show about a 40% drop in frame rate, but at the super high frame rates normally seen, they are all still above 100fps, which syncs nicely with the monitor at 100hz.

IDK if I would actually run this way, but it is nice to have such low CPU temp. It might also help when tuning Intel iGPU - depending less on CPU cores and more on iGPU.

 

Great temps hmscott !

Where do you configure Intel SpeedStep ? In bios ?

I believe we can't OC the CPU in Bios. Maybe an unlocked Bios will release soon

 

gloops, yes, in the Advanced tab, a simple disable. It keeps the CPU at base clock, no higher and no lower, which would seem to be the same as running the High Performance Power plan (CPU 100%/100%) and using XTU to set CPU cores all to 29x.

It's mostly an exercise in trying to reduce the CPU load/power to the minimum needed for the same score.

In the games I am trying, testing shows 100fps+ average, which enough to enjoy the game at 100hz screen sync, while keeping the CPU temp/power low.

It's like tuning a profile for use on battery, but also using it on AC power

 

Good tips, I will try this

But it's strange that the scores are the same with CPU @2,9GHz and @4.1GHz... ^^

 

gloops, the Speedstep disabled 2.9ghz CPU run was with full GPU OC. +135 GPU +385 Mem.

Comparing the 2 scores you can see the CPU performance is lower, but the GPU performance is higher, which helps level the scores to the same result range:



Games do show a definite drop in fps, but not so much as to rule out gaming with SpeedStep off. I guess if I were running in super high ambient temps, this would be a nice thing to try.

The GPU's are already running very low temps, so relying on them to make up performance lost by running the CPU slower - helps keep the same performance while lowering the CPU temp.

Much more testing, and trying out low settings in XTU for battery use, might come up with an even lower power draw / heat generation setting that still has adequate game play.

We do have different CPU's, 5950HQ vs 4980HQ, it will be interesting to see how the 2.8ghz base performs with Speed Step off.

 

I understand, that will be good for benchmarking

I can't OC my GPU's more than +135/+200 (with MSI AfterBurner)

Have you succed to OC more than +135/+200 ?

 

gloops, at first the 980m SLI could only do +110 GPU and +250 Mem.

 

Just for fun, a run of Fire Strike on Intel 6200 iGPU

Model MSI GT80 SLI-263
Test 3DMark Fire Strike
Thermal compound Stock
CPU Tuning Stock CPU
GPU Tuning Stock iGPU
Driver : Intel 10.18.15.4232 (included version)
Shift Mode High Performance Power Plan/Cooling Fan on
Room/Ambient Temperature 68F

http://www.3dmark.com/3dm/8188659

 

geezz thats a funny test between the physics and graphics score but I honestly would have thought they would have done better than that. I don't see how that is even usable with gaming in someones perception but I guess its better than nothing. Just doesn't seem like something to advertise as intel does.

 

GTVEVO, what do you see on your CPU/iGPU?

The 1628 score is about 1/10th of 2 x 980m in SLI, that's not bad... for a little bitty chip on the CPU carrier...

This article promises some good performance in games:

Broadwell: Intel Core i7-5775C And i5-5675C Review
http://www.tomshardware.com/reviews/intel-core-i7-5775c-i5-5675c-broadwell,4169-6.html

 

Never even bothered to test but I will so we can compare for a laugh.

Sent from my SM-G925T using Tapatalk

 

Better than 589 for my HD4000...
http://www.3dmark.com/fs/4719231

 

First let me say that was brutal to watch, but you win. lol

Everything stock cpu and gpu, no offsets.

 

Thanks GTVEVO, that's what I thought when I first watched it, then I began to enjoy the flip-book still images - I saw detail I hadn't seen before in animations that were too fast to catch. I have watched them too many times

1439 is a better score than I expected, I thought the 6200 was supposed to be a much better GPU than previous models, but GPU score is only about 10% faster, 15% faster overall.

I will try some games at some point, on battery as well, to see how it holds up.

Before I rebooted into SLI land again, I thought I would see why I couldn't OC the CPU, when I ran the GPU stress test and enabled all the TDP readings, I saw why.

The CPU Total TDP hit 71w, until it dropped to 47w, and the GPU TDP went from 45w to 30w. The CPU was only at 3% utilization.

There is no additional TDP headroom for a GPU on a 47w CPU package, if you are going to use the iGPU - you won't have much left for CPU, and visa versa.

 

vbios 84.04.48.00.29

Model MSI GT80 SLI-021
Test 3DMark
Thermal compound CLU on CPU & GPUs
CPU Tuning : Stock with Intel SpeedStep Enabled (CPU blocked @2.8GHz)
GPU Tuning Stock vBios @ +135MHz core , +200 memory
Driver : 353.62
Shift Mode High Performance Power Plan/Cooling Fan on
Room/Ambient Temperature 25°C

http://www.3dmark.com/3dm/8191050

 

If you guys limit your CPU to 3.4ghz or so, you will notice you won't hit the TDP ceiling and will basically run at that speed all the time. Firestrikes scores are very sensitive to dome CPU variations but they are much more useful for graphic scores mostly.

 

ryzeki, here are the results of XTU tuning/testing for longest sustained performance with the 5950HQ in the GT80 SLI-263.

35x x4 cores @ -105mV and 35x cache, holds as long as the job runs. I monitored for over an hour before grabbing a snippet of the monitoring.

The average TDP was 45.2w over the hour run. The 5950HQ TDP is 47w.



Update: To try to squeeze out as much throughput as possible, I am now running at 36x cores -100mV 36x cache. The result is an sustained average multiplier of 35.7 with average TDP of 46.855w.

As a test, I did a run at highest OC / peak performance of 41x x4 +50mV 37x cache +1.9v and the sustained multiplier was down to 31x, with higher temps.

Due to the voltage offset difference - starting low at 36x allowed setting -100mV, while starting high at 41x fixed the voltage offset at +50mV, adding heat to the package and reducing the headroom - resulting in a max 31x sustained throughput.

For long jobs its best to align CPU settings at MAX TDP, finding and setting the highest sustained multiplier allowing the lowest voltage to CPU/cache for best sustained throughput.

 

Good testing, i have done the same myself and i concur with your findings. The difference being that I cannot undervolt as much.

My best was 35x at -70mv which worked 90% as it wasnt stable.

The best tool to measure max tdp so far is cimebench multicore test. It is a punishing max tdp test that can bring the cpu turbo to its knees. At 36x and -40mv, my cpu consumes up to 58w. After turbo ends it drops to 31-33x multi.

My personal preference is running the cpu at the max possible sustained multiplier. The reason being that anything that pushes the cpu hard will run at max performance, whereas if something doesnt push the cpu, it probably does not need higher clocks.

Sent from my SM-G925I using Tapatalk

 

ryzeki, cool, I will have to try out Cinebench, I didn't know it had a sustained mode.

I started the testing using Prime95, but wanted to set up FAH and found it was holding a higher TDP than Prime95, so stuck with FAH - doing good work for the future while tuning

Prime95
http://www.mersenne.org/download/

Folding at Home
https://folding.stanford.edu/

The undervolt I was using is stable, but only just. At -105mv it power cycled after 1 hour, and -100mv worked for hours.

But, for everyday use - when I am doing important work I don't want going away, I have it set for -85mV @ 36x x4 cores and 36x cache. No reboots yet

I am using the Balanced Power Plan, for this mode - 100% load sustained work - it works the same as High Performance, and allows cooling during idle times.

I wonder how the Skylake CPU @ 45w is going to fair, will it hold 35x/36x, or will it drop down to 34x/35x due to lower TDP? 2w doesn't sound like much, but in my testing if it has the same power/temperature characteristics as the Broadwell 14nm it's possible it won't hold 35x sustained.

 

Precisely my concern with skylake. In theory it has more performance at the same power consumption. But if tdp is slightly lower (2w is usually 1x multiplier on the 35x 36x range) then we might get still similar performance to haswell.

The trick will be, how much of that tdp is directed at the cores.

With haswell and broadwell we have the power for the cores (IA cores) and the power for NOT the cores (uncore) as the main power consumers.

Uncore is the powrr user by memory controller, graphics etc etc. Anything that isnt the cores themselves. Naturally, the tdp for the cores is actually lower than 47w as that is the total power.

In my experience with haswell, uncore is between 8-10watts under full load, so the cores themselves only have 35-37w.

In the case of my lenovo t540p which uses the intel grqphics, the uncore is larger at up to 20w so my cpu cores pnly run ay 3.0ghz or lower at full load. Not enough power left for them.

Anyways as with any current mobile cpu, TDP is the current limit to beat, provided we get good cooling


Sent from my SM-G925I using Tapatalk

 

First run 3DM13

http://www.3dmark.com/3dm/8453981

 

That's a nice score

 

Hi

 

Not bad, our systems do pack a punch You can go higher my friend.... if you dare! just remember to be careful, we are not responsible for destroyed systems!

http://www.3dmark.com/fs/4716218 you can surpass me, you are the chosen one!

 

It would be great to include a screen shot of where to overclock the GPU . I'm a noob :| I have never overclocked , I only use shift mode.
But It would be great to know how to increase gpu voltage to overclock it

 

http://www.3dmark.com/3dm11/10531400

 

You need unlocked vbios, then flash them, then use nvidia inspector. It's not for the faint of heart and just remember you are playing a dangerous game hahaha

That being said, I yield, brman2000005, I can't get stable enough clocks to outdo you. At best I can match or fall slightly behind. You win the Titan 001 fight

 

Slightly higher than my titan-x results I'm going to have to have another run at some point

 

My setup Titan X

http://www.3dmark.com/3dm11/10444153

 

When I hit 17k in Firestrike I stopped benching lol. Just escaped the great Windows 10 debacle...after reinstalling Windows 8.1 from scratch, and then installing drivers this thing is perfect. Kinda sucks tho, I love 10. Edge browser is legit. Next step is to put four 250 gig SSD's in this thing and see what it can do from a disk i/o perspective.

 

Yeah a full desktop can always just brute force past, but mine is a single titan-x in the GS30 dock

 

Here is my new msi gt80s titan 6qf, completely stock. This is with Battery settings on High performance and msi mode set to Sport. Current Nvidia Drivers that came out 12/21/15.

 

I suspect the GPU core is throttling a bit in these tests, unless I'm wrong about 1126 being the boosted clock of the MSI GT80S 980s.

 

Those things push a sweet score!

 

The best way to check would be a graph of clocks over the benchmark, usually the stated boost clock is exceeded though.

 

Model GT80 SLI-072 (desktop 980s)
Test Fire Strike/3DMark11
Thermal Compound Stock
CPU/GPU tuning Yes - CPU UV -100mv, 40/39/38/38 multi / GPU OC 160 core/269 mem
Shift Mode (Green, Comfort, Sport)
Room/Ambient Temperature 60 F
Comments First attempt at overclocking

ASIC quality 76.3% and 73.6%

3dmark firestrike standard 17507
http://www.3dmark.com/fs/7211735


[/url][/IMG]
3dmark11 19792
http://www.3dmark.com/3dm11/10808990

 

Oh wow, -100mv at 38x? thats awesome. Best I can do is about -85mv at 36x otherwise it is unstable.

 

I finally had some time to play around a bit, and here are my new 3dmark results. I am extremely surprised with how good this thing does with the power limit from EC and the stock thermal paste. I would bet one of the guru's on here could do so much more with this machine.

Model GT80 SLI-072 (desktop 980s)
Test Fire Strike/3DMark11
Thermal Compound Stock
CPU/GPU tuning Yes - CPU UV -100mv, 40/40/40/40 multi / GPU OC 160 core/273 mem
Shift Mode (Green, Comfort, Sport)
Room/Ambient Temperature 65 F
Comments Testing/Benchmarking

ASIC quality 76.3% and 73.6%

3dmark firestrike standard 18234
http://www.3dmark.com/fs/7312801


3dmark11 21183
http://www.3dmark.com/3dm11/10846286


I also tested using dual 330w supplies using the clevo dual adapter. The system did not appear to use any more power which was expected with the EC limitation. I also noted that the same power supply always got warm, while the other was cool. This was no matter which spot it was plugged in to on the adapter. Both supplies are good and net the same 3dmark results by themselves so I am at a loss. I even tested which was was plugged in first, and which one last, but always the same supply got warm.

I can only assume, since it just happens to be the stock adapter, that somehow that was is better and the clevo dual adapter always picks it first. I also assume that the way the adapter works is it won't use power from the 2nd adapter until the first one is maxed out, which of course I can't do due to EC limit.

 

The adapter most likely places them paralell but it has no load balancing whatsoever. You would need extra hardware for that, so that it would draw evenly from each PSU. The thing you have nailed on though, is that you are limited to a single PSU.

 

I knew they wouldn't load balance but I fully expected to be able to swap them around and choose which one it draws from. It always only draws from the same supply though, no matter which one is in which port on the dual adapter, or which one is plugged in first, or which one is plugged in last. I tried every combo and never can get the other supply to be used instead. That adapter works fine on its own so I know it is good. I think it has something to do with that one supply may be slightly higher voltage or something like that.

 

still pretty cool you can do that on one PSU, my 18k+ runs are usually pulling 500 watts plus from the wall with my 980m SLI (P870dm-g). In other words there is no way it would be possible on 1 PSU. I think I made it to mid 16k mark before the adapter would turn off.

 

Yeah it is weird, it's almost like they made the desktop version MORE efficient with respect to a certain performance point, which is the exact opposite of what I would expect. They should be slightly less efficient/more power hungry I would think.

 

Do you know the limit set by the EC?