GTX 980m Limited by CPU....

Well for giggles, I took the review sample that PowerNotebooks provided me of the MSI GT72 with GTX 980m and i7-4710HQ and thought of a way to check if the CPU was limiting the performance of the 980m. My findings are more or less, yes, and depends on the game.

I ran 7 game benchmarks while running the CPU at 2.5GHz, 3.0GHz, and 3.5GHz using Intel XTU and was looking for a trend of FPS improvement with faster CPU speed. The CPU provided with this machine is an i7-4710HQ with a top single core boost speed of 3.5GHz with four cores at 3.3GHz. This chip can be boosted an additional 200MHz using Intel XTU (and likely Throttlestop) so I could achieve my 3.5GHz speed.

The seven game benchmarks run were:
Bioshock Infinite
Crysis 3
Grid 2
Resident Evil 6
Sleeping Dogs
Middle Earth: Shadow of Mordor
Tomb Raider

They were all run at 1080p and their highest detail presets.

For those results that the FPS trended upwards, I also did the same test using an i7-4810MQ and GTX 860m on my Clevo W230SS to see if the same trend occurred, basically to eliminate any chance of it being game related. Of the three games that FPS trended upwards, the GTX 860m remained flat. This tells me that the CPU was limiting the performance of the 980m.

The vBIOS is still clock locked at +135MHz for GPU and so far haven't been able to flash anything, since nvflash hasn't seemed to be updated to work with this specific GPU. Prema hasn't been able to get it working yet either. In any case:

980m Stock = GPU: 1038/1128MHz, vRAM: 5000MHz
980m OC = GPU: 1173/1263, vRAM: 6000MHz
860m Stock = GPU: 1200, vRAM: 5400MHz (Ok not stock, but slight OC)

See results of the testing below:
















The difference is subtle but it is measurable and apparent, and at which clock speed it is no longer limited is not known, but I guess advisable to choose a faster CPU if you want to get the most out of your 980m.

 

I don't see a CPU bottleneck. 4/7 tests were within margin of error. In the other 3, yes, there is a tiny increase of a few percentage points but look at how high the FPS already is. The games weren't GPU-bound and the 980M was probably yawning, hence the limiting factor turning to the CPU. I bet if you added DSR to those 3 games, they'd fall within margin of error just like the other 4.

So all in all, no meaningful impact on performance regardless of 2.5 GHz or 3.5 GHz that I can see.

An XM chip that could be pushed really high and really low would've been great for this sort of testing. But great job nonetheless. :thumbsup:

(Damn it, still can't give you rep.)

 

The games that trend upwards do so 8-15% difference 2.5 to 3.5GHz, but 860m was less than 2%. It may not be meaningful, but it is limiting it to some extent.

I did a couple runs lowering detail and that gap widened even more. So I guess moral of the story is, especially with soldered chips, better to buy a faster one, although I don't think XM or i7-49x0MQ is needed. And might as well put your 980m to work at higher details so it isn't limited by the performance of the CPU overall. But considering the 980m can push pretty much any game at reasonable frame rates, it isn't much of a concern (for now).

 

Can you run those 3 games with DSR and then see what kind of performance scaling you get at different CPU clock speeds? I bet it'll flatline.

Because 860M was already GPU-bound and FPS was a lot lower.

 

It will flat line because then the GPU will be the bottleneck trying to crank out frames at such a high resolution. I'll give it a go when I get a chance later.

 

Exactly. I can make my 650M SLI "limited by the CPU" if I run everything at 720p Low and get 100+ FPS. It doesn't mean I'm bottlenecked by my CPU, just that my graphics settings are too low. My GPU doesn't have enough work to do, so naturally the burden falls onto the CPU.

I think the moral of the story is to run reasonable graphics settings for your GPU. Don't turn the eye candy down too low for a fast card, and don't crank it up too high for a slow card.

As long as you have at least a quad-core i7 with that 980M, you should be fine for 60 FPS in the majority of games. But if you're chasing that 120 FPS mark with a 120 Hz screen or have SLI, where CPU bottleneck is the #1 enemy of scaling, then get the fastest CPU you can afford.

 

I'd be curious to see what the results look like in SLI.

 

More CPU-limited, that's for sure. If I had 980M SLI with a 120Hz screen, I'd want an overclocked 4940MX.

 

all the more reason to wait for Broadwell, thanks mate!!

 

Umm, Broadwell may not even have socketed chips, and the soldered H chips aren't due until 2Q 2015.

 

If Broadwell is the single digit percentage performance improvement typical to die shrinks, then nope.

 

well, it is not only the performance, but efficiency that bottlenecks laptops, as higher TDP chips cannot go into laptops, I believe Broadwell will be an improvement in efficiency, therefore we will have better cpus in laptops

 

Actually, it's mainly the heat that limits Haswell in notebooks. I don't have too much hope for Broadwell. It'll do more work but also run hotter at the same TDP.

 

That kind of test shows what I think is normal. But here's a better test I wish you could perform Sir Wingnut. I would like you to run 3.2GHz constant and then boost to 3.5GHz constant. We've seen 4-5FPS in games that use the CPU more heavily going up in increments in 0.5GHz, but if the difference is between a 4710MQ and a 4810MQ, I'd like that to be more apparent (for people who would not or can not overclock their CPU). I made the point with Cloudfire that a +200MHz bonus on CPU does almost nothing in relieving CPU limitations when he stated how the 4710HQ was likely limiting 980Ms' performance substantially. I'd honestly like to do a lot of testing on my own with one of these bad boys, but alas that isn't going to happen anytime soon and if it ever does, it'll be irrelevant by then XD.

 

Honestly, I think that would be a waste of HTWingNut's time. We've already seen how small the difference 500 MHz makes, it doesn't take a genius to deduce that 300 MHz would be an even smaller difference.

 

Truthfully that's why I said I wish, and that's why I then said I wish I could try it myself. I was more wondering if the 3.3GHz to 3.6GHz bonus would have been past some sort of cutoff line for benefits, so to speak.

 

The thing I don't understand is how the CPU is only 47W TDP, has a similar cooling setup to the ~ 90W TDP GPU, yet still runs hotter than the GPU at load.

The advantage of the i7-4810MQ over i7-4710MQ/HQ is that you can overclock by an additional 400MHz, so if your laptop is up to the task you can overclock it to 4GHz with all four cores, and 4.2/4.1GHz for one/two cores. 500MHz can make an appreciable difference at those speeds. 500MHz effective improvement. If I were running SLI I would run an i7-49x0MQ or XM CPU.

 

Because Intel CPU's have much higher thermal density than GPU's, hence are much harder to cool even at a lower TDP.

 

That was my one mistake I made, getting the 4800MQ I think. A 4900MQ should have been what I went for, but I didn't see the 100MHz bonus being worth the $120 or so, since I didn't expect to be able to OC the thing (and I can't yet either XD)

 

I would really be interested to see 980m sli results

 

Feel free to buy me a 980M SLI notebook for testing. I will write a bible on it.

 

Preferably without a 4710HQ...

 

The 4800MQ will be thermally limited in your system to 3.9ish GHz which the chip can be set to.

 

THERMALLY limited? Whot?

 

The cpu cooler is relatively small so with fine tweaking you can keep it happy up to about 3.9ghz. Any higher and it may throttle in games.

 

hmmm, the bottleneck is there, subtle, but there. The 980m is basically 2.3 times (accounting for clockspeed differences) an 860m so the true performance with not CPU bottleneck should be in the 110 FPS ish range. I guess an XM chip in the 4.5ghz range is mandatory for SLI setups otherwise there will be very limited scaling.

 

I say it depends on the game and settings. If you run every demanding game at 4K, it will scale nearly perfectly across the board.

 

Seems like the key for 980M SLI might be to crank up settings just enough to shift the bottleneck back to the GPUs. Or overclock a 4940MX if you have the dough and your laptop can take it.

 

Make the game GPU-bound, that's always been rule #1 for optimal multi-GPU scaling.

 

Just seems to me that for the first time, one will actually need a better-than-base-model processor to tap the full potential of 980M SLI at 1080p. Or grab one of those 3K/4K screens :laugh:

 

It's because mobile GPU's are getting more powerful relative to desktop cards. CPU bottlenecking and poor scaling has been an issue on extreme multi-GPU desktops for years. SLIKnight, the legendary Nvidia tweaker over on the 3DCenter forums, runs 780 SLI with a 3.7 GHz Nehalem i7-940 and never gets 99% GPU usage unless he adds really high levels of AA.

As for 980M SLI, just get a 3K/4K screen or add SSAA and you won't need anything above a base-model quad-core as you'll always be bottlenecked by your GPU.

 

Can't get any worse than Skulltrail right?

Was recently taking a trip down nostalgia lane and when I read up on Skulltrail I chocked a bit. That's $3000 in processing power alone for 2 QX9775 chips and another $600 for the D5400XS. We haven't even begun to look at other things. And the QX9775 was a 150W part. Holy farting batman. And seems like the choice to go with FB-DIMMS really held back performance, although Kingston did release a low latency kit which I don't know if it helped any. Just as a kick in the balls, for all that money Intel didn't even have the decency to include a northbridge fan, which apparently was a must if you intended to overclock. And apparently the board itself kind of sucked for overclocking anyway. Skulltrail sounded like it was an epic(ally expensive) fail.

To bring this back on topic: if quad SLI is wonky now, I can't even imagine what quad SLI with 9800GX2 would've been like on Skulltrail.

 

The CPU 4860 might go better than the 4710?

 

No, it's a waste of money.

 

if I take a asus G751 with 4860, a bit of difference there seems to be.

 

Whatever floats your boat.

Mobile Processors - Benchmarklist - NotebookCheck.net Tech

 

then put any processor, the differences will be minimal

 

Any CPU short of a significantly overclocked one, yes. Right now you're comparing two chips with only 100 MHz difference.

 

but there is no alternative for now for Asus g751

 

So get the cheaper 4710HQ and take the free +200 MHz OC.

 

How much oc did u do on the 980m? I request to bench on 4940xm at 4.5 ghz if possible please

 

Not happening, his CPU is soldered.

 

I think Intel's definition uses words like "average heat dispersion in the range of..", etc.

 

If you go with i7-4860HQ you can run at 4GHz. How much more this will help who knows. There is thermal headroom in this specific laptop, however there is not much TDP headroom, and the i7-4710HQ constantly enters TDP throttle state when at 3.5GHz and haven't figured a way around it, even with Throttlestop. Intel needs to stop gimping their CPU's, remove stupid Boost, and let us OC how we see fit, just use a thermal throttle limit only (like 98C).

BTW, thanks to LPC-Digital, should have an SLI setup in the next few weeks. I don't know how I'll find time to do all this, but will do what I can.

My whole point in this exercise really was to see if running at 1080p and at max or near max detail, which is realistic for users with this type of laptop, if there was any affect to performance based on a base level CPU. While minor, it is there.

I wish I could swap out CPU's to test, but with a soldered CPU, no chance.

 

Hmmm if 4710HQ is already TDP limited, what's the point of overclocking 4860HQ by 600 MHz if it's not gonna be able to turbo that high under load? They're both 47W chips.

Sounds like an unlocked BIOS to raise the power limits is in order...

 

I'll fiddle with it some more.

I updated the first post stating what the clocks were for 980m stock, 980m OC, and 860m Stock (not stock, slight OC).

 

Interesting... the 4860HQ is the minimum thing to get if integrated O_O

 

As i said before:

http://forum.notebookreview.com/gam...ficially-announced-nvidia-18.html#post9799400

 

There's a difference between a small increase and a "bottleneck". A small increase is OCing = raising FPS in small increments. A bottleneck is where there's a huge performance lockdown due to CPU. Most games, no matter the CPU, benefit from OCing slightly. Is it a HUGE benefit? Nope. Is it a reason to OC a CPU that's at 3.5GHz normally? Not really.

If a 1GHz OC doubled FPS, that'd be indication of a bottleneck. But 10fps up from an average of about 85? With a 1GHz OC? That's... not much. Plus, his CPU is limited. His best bet is to attempt it with a 4910MQ at 4.3GHz versus an artificial 3.3GHz using a single 980M or something, and THEN we'll see what it's really like. But if the improvement drops off after like 3.6-3.8GHz, then it's not all that much of a bottleneck at all. Ironically, the one game where CPUs are always a bottleneck for powerful systems is CoD: Black Ops 2. An improvement of 3.4GHz to 4.5GHz on a i7-4930K yields an extra 150 FPS at max graphics 1080p on two GTX Titan Blacks. ONE HUNDRED AND FIFTY FPS EXTRA. I've never seen such a benefit in any game ever. That there is what a bottleneck looks like.

 

Can someone explain how "Turbo" works compared to desktop CPUs? Does this mean as long as we set our power settings properly we can expect to stay at the Turbo frequency without overclocking? Or does the frequency go up and down while it's under load?