Minimum Haswell CPU clockspeed for 3k 60fps gaming

Hello all,

I'm currently in the process of undervolting and underclocking my I7-4720HQ processor in my P35X V3 with GTX 980M to reduce the CPU temps, as currently I'm constantly hitting 95C in several games and throttling, causing stutters. I have not yet repasted but i will do so in the near future with Gelid GC extreme, which I expect to bring down temps up to 5C. I want to underclock it to the lowest possible and leave it at that, as to keep the temps below the throttling range and prevent stutters.

Currently, I'm undervolted dynamic CPU voltage by -60mV and running at 3.2Ghz turbo on 4 cores and 3.1Ghz cache. (seems like a weak undervolt compared to others sadly. -70mV crashes in daily use. I have seen users with 4710hq's undervolting by -90mV, when running 3.3ghz on 4 cores. I expected the 4720 to perform better, since it's a later revision and Intel's manufacturing process would be more refined, thus creating better silicone overall).

When playing War Thunder ground forces at 3k ultra, i seem to manage 60fps with these settings, but when running any lower than 3.2ghz the cpu is bottlenecking it and fps is significantly lower.

I'd like to know your findings on minimum cpu speeds needed in order not to bottleneck a 980M in games.

Kind regards,

Arthedes

 

When I think of it, the factory voltage of the 4720 might be lower than the 4710's. Whats your voltage when running 3.2ghz?

Mine's 1.014V @3.2GHz -60mV

and 1.034 V @ 3.4GHz -40mV

 

You mght want to repaste.. with lower temps, your 4720HQ will run at higher clocks and not throttle.. Also if you are undervolting increase the core current limit to prevent current throttling...

 

I can't control current limit since i have an HQ chip sadly, but luckily I haven't seen XTU reporting current limit throttling at all.

I do notice power limit throttling in some stress tests, but that's due to the HQ chip being hard limited to 47W according to D2Ultima appearantly. I can't imagine it going above 47W in real world applications like gaming or rendering though but that has yet to be seen...

Definitely going to repaste BTW, gonna use Gelid GC Extreme. How much temp decline can I expect from that? I find it hard to say since its throttling anyway.

 

It depends on how good thermal grease you use now and whether the application was good .. Normally + - 5 degrees difference between a good paste and a medium paste.

 

I never said you couldn't adjust current limit due to using a HQ chip. I said your motherboard probably does not allow it.

Next, the clockspeed depends on the game. GTA V, 3.5GHz 1080p60fps is not enough if you don't force a GPU bottleneck.
1080/60 BF4 max graphics shouldn't be a bottleneck with 3.2GHz.
1080p/60 Mass Effect 3 probably wouldn't even be an issue at 2.5GHz

Please note: higher resolutions INCREASE CPU USAGE overall. It's not by much, but it does. People don't often notice, because it MUCH more likely forces a GPU bottleneck. But with that 980M you got there and a TDP locked CPU... well....

 

Thanks for your reply.

How come my processor can run at 58W sometimes, although not sustained, when the motherboard does not allow it?

 

Your 58W is your short power boost. Your 47W is your long power boost. You can always run at your long power boost. You're actually lucky. Some machines cannot even pass their long power boost. The short power boost is designed to trip off after a while, as I pointed out in my guide when I said "benchmarks betray the power limitations as they are often under the time limit allotted to the short power boost". Machines like the GT80 Titan can draw infinite power on their CPU for that 2.5 minutes, so you could overclock it to 4.2GHz and run Firestrike if you wanted. It'd pass with flying colours. But if you sat and ran BF4 at 125fps, you'd start stuttering like crazy after 2 minutes as your chip downclocked by over 1GHz (with the voltage needed to achieve 4.2GHz, its TDP would skyrocket).

 

So, any clock speeds you guys found to be too low for certain games? Whats the minimum you can get away with?

 

For me in some games? I realllly want that 4.3GHz

 

4.2GHz is plenty however with the games becoming more CPU intensive even a 4930MX @ 4.3GHz won't be enough for 4K@60Hz by next year... Then again, who knows what will happen in a year?

 

Nah, I pretty much guarantee GPU will be the bottleneck at 4K if we're talking laptops

 

Wow, that seems like a steep goal. I never imagined the CPU needing to be clocked THAT high

 

3.5GHz is a bottleneck for me in some games. Granted, I'm not forcing a GPU bottleneck and I have a 120Hz screen, but if I could find SOME WAY I would keep my CPU at like 4.5GHz 24/7

 

I could do 4.5GHz constant if I didn't have this stupid TDP throttle.. Damm Alienware

 

I use i7 4710MQ at 2.5 GHz at FHD resolution and I have not experienced bottleneck in games. Sure, I lose a few fps but not instant drops like from 60 to 10 fps which is pretty common with bottlenecks. I use it due to lesser heat and fan noise. Actually 3.0 GHz is the best in terms of balance between heat, fan noise and performance for i7 4710MQ but still, I prefer lesser heat and fan noise for sacrificing little performance. I can't even imagine overclocking my i7 4710MQ to 4.0 GHz. That would be horrible when Haswell CPUs are already running too hot. Maybe someday I will enjoy 4.0 GHz CPU speed with no heat or noise problems after I will sell my laptop and buy a desktop.

 

i7 -4710MQ to 4.0 GH ? Perhaps with all 4 cores? What.... LoL

 

@Dufus got his 4700MQ to like 4.7 GHz XD

 

I know... But it is no others who have managed this Whatever 4.7ghz does not work with Wprime 1024

 

FWIW I'm not the only one to achieve 4.7GHz on a 4700MQ but perhaps it's a mute point. Maybe if I had a fine machine such as yours Papusan I could achieve a bit more. Also ambient temps where I live of over 30C most of the year round don't help. Better if I lived in a country like Norway where just opening a window can achieve ambient temps of -40C hahaha. Well, it felt like -40C during the times I've been over there.

What's your best Wprime 32M? I see Mr. Fox ran at 4.9GHz for just over 5 Seconds, should be under 5Secs at 4.9GHz (~4.7Secs) so maybe some throttling or it just wasn't run optimally. The 4930MX is IMO a better binned chip so results should be better but running heavy loads at desktop CPU speeds with a much lower pin count is probably not such a good thing for the CPU.

 

LoL... Yes its cold home but not -40C... My wife love to have fire in the fireplace home so it's hot anyway we don't need aircondition home if you understand. I run normally my 4930mx ok 24/7 on 4.3ghz so I haven't tested with higher clock speed. I run more Wprime 1024 and Cinebench benchmark tests because I like the test run a longer time and compare with Hq processors ... You know this is not a BGA processor with turbo boost that can run only 28 sec before the clock speed go down. I have tested only with 4.3 and get a score 5.351 in Wprime 32 and 167.560 in Wprime 1024. Many laptop processors(also bga crap) can run with an o.k Score in short benchmark test but not if the test go over a longer time.

@Dufus have you noticed that many of the i7-4940mx processors have more problems with an ok overclock than lder 4930mx ? I and Mr Fox think maybe they are poorer binned. What is your conclusion on that?

 

High CPU clock speeds are needed for high refresh rate gaming rather than high resolution gaming. For 60 fps 3ghz should do fine.

 

I don't see why it is so funny to you, it is not a BGA after all...

 

Would the CPU run at a lower clock speed if it is not needed? For example, in game A you would get 40 fps with 3.5 ghz and also 40 fps when clocked at 2.5 ghz, would it run at 2.5 ghz all the time or stay at 3.5 ghz?

I havent noticed this really.

EDIT: I just ran AC:U in a busy area on 3k and I get 35 fps with either 3.0ghz and 3.4 ghz on 4 cores. This indicates that in this case the GPU is bottlenecking the situation. However, I find it strange that the CPU would run at a higher clock speed and a higher temperature when the framerate would stay the same. With 3.0ghz the cpu hits 85C and with 3.4ghz the CPU hits the thermal cap of 95C.

 

I thought CPU bottleneck only occurs when one or more cores is 100% loaded. However, in Borderlands 2, when running at 2.6GHz I get less FPS than when running at 3.4 GHZ (even though at 3.4 it will start thermal throttling after a few minutes), even when the cores are not fully loaded (I know this because I run msi afterburner, showing the load of all my 8 individual cores, and not one of them is at 100%, more like 40%)

How do you know your cpu is bottlenecking or not?

 

Turn off Hyper-Threading then check load

 

It's not possible to compare pitiful Bga processors that throttle in thin laptops and socket processors. No Bga processors = No thin laptops with pitiful cooling . With a bios that is functioning properly so is a i7-47xxmq processor the best choice anyway. I do not bully socketed processors (the best choice). There are only Hq aka BGA processors who is known as garbage... See what is written in my my Custom Title under my Avtar.

 

Why?

 

No, it'd likely clock up quite a bit more. It usually clocks up more than necessary, probably to compensate for CPU load spikes.

Higher clockspeeds (and voltage) automatically mean higher temperatures™ even if the load is the same (not 50% CPU vs 50% CPU, but 50% CPU at 1.5GHz and 25% CPU at 3GHz, etc). If you are worried about temps and you find that game X is GPU-bottlenecked™ then you might as well lower the clockspeed via Throttlestop or something.

No, this is not the case. Bottlenecking can be a real annoying thing to discover, and regardless of what JayZTwoCents™ says, is present almost all the time, in some form or other. It USUALLY is a GPU bottleneck of sorts, but even with an absolute GPU bottleneck, raising CPU speed can still sometimes improve FPS. CPU bottlenecking however can be hard to determine, for various reasons. Here's some ways in which you might be seeing "extra" power available, but overclocking/underclocking greatly affects the game still:
1 - Games don't like to use "100%" of a single core when hyperthreading is enabled. This usually tops out at ~95% if it's heavily stressed, but to your eyes, more power is available.
2 - Games sometimes are designed for x# of cores. If they see more, they can split the load, and your CPU will appear to be quite un-stressed, but it's actually using all that it is coded to use, and thus clockspeed bumps get a nice boost to FPS.
3 - Windows 8. Some games or programs have a distinct limit of the CPU power they're allowed to use. Windows 8 has screwed up by making "100% CPU" mean somewhere between 75% and 80% CPU. For example: HERE is me rendering with Vegas at Win 8's 100% CPU. And HERE is me using TS8's benchmark, which actually uses 100% CPU. So sometimes Windows 8'll think it's at 100% and a game will stop using the CPU or something, even though there's much more power available. HOPEFULLY Windows 10 fixes that, or you can use Windows 7 if you aren't interested in the benefits that Win 8 or Win 10 bring.
4 - Games might just be single-thread heavy. Mass Effect 3 kind of pissed me off like that. I had set it to lock to 125fps and it had trouble maintaining that because 3.5GHz on a single core with all the graphics goodies I turned on via NVI wasn't fast enough (see previous statement about 4.3GHz or faster CPU being wanted by me =D). This one however, should be pretty easy to tell.
5 - Games could be made by Bohemia Interactive, and for some reason have a minimal load on both CPU and GPU and still run like a sloth with broken legs. For reasons yet unknown, overclocking seems to work.

With hyperthreading off, as I touched on earlier, you can see the full, proper load that the game is placing on your CPU. The hyperthread cores are extra CPU performance, yes, but with them active, loads will never look like they hit 100%. Except for Final Fantasy XIII that game somehow manages to hit 99% on a single core and 0% on the hyperthread.

 

Because unless a game utilizes multi-threading effectively (hint: not UE3 DX9 games like BL2 which are basically single-threaded), you are CPU-bound at 50% load on a Hyper-Threaded CPU. That's why you don't see a single thread getting maxed out even though you are clearly CPU-bound. Disable Hyper-Threading and you'll get a real picture of your CPU utilization.

 

Ahh so that might explain my cores being loaded only to ~40%. Would that mean that with HT on, 50% load is 100% actual load?

Wait, come to think of it, ive seen loads of 80 ish sometimes, how could that be explained?

 

Thanks for your post, it is very educational.

So pretty much the aim would be to clock the CPU as high as possible.

After tinkering last night, my optimal setting is 3.0 GHz. This way, I'm constantly on the brink of thermal throttling, so the cpu stays at a constant 3.0GHz. If I clock to say 3.4 GHZ, thermal throttle will occur and most of the time it will clock below 3ghz and then up to 3.4 again.

I prefer the constant 3.0ghz to the inconsistent throttling.

Now its time to order some GC Extreme and remove my crappy stock TIM application.

EDIT: would I benefit from disabling HT? Most games dont seem to utilise 8 threads even in this day and age and I heard HT only brings temps up.

 

While in theory it should be 50% load, HT is only ~30% of benefits at best. As I did say, I've seen single threads being maxed before. I thought I had a great example, but this is the best I can get on short notice with my overlay on. FF XIII is the culprit; look at that core #0 load =D

There is no benefit to be gained from killing HT on quadcores. Even if the game doesn't use it, other programs can and do, like voice chat or music players, etc.

 

An app which is not multi-threaded properly is CPU-bound at 50% of overall load on a Hyper-Threaded CPU is what I meant. 30% ideal perf gain from HT is separate from the CPU load, otherwise HT would double throughput (+100%) which it obviously does not.

That's an example of a predominantly single-threaded game, like this one. Most last-gen console ports are dual- or tri-threaded since that is the number in PS3 and XB360, with a large main thread due to DX9.

Yes, if the reduced temps from disabling HT allow your CPU to clock up higher, you will get better performance in most games, particularly single-threaded ones like BL2.

 

I'd like to try that out. I'd imagine a higher clock with the same temps and 'only' 4 cores would achieve better results.

The option is not in my bios though...

 

Any other findings that could be interesting?

 

Guys, check this comparison between processors and their relationship with resolution.

http://www.tweaktown.com/tweakipedi...80-sli-vs-gtx-980-sli-at-2560x1440/index.html

By their findings, increasing the resolution does not increase the CPU load.

 

That doesn't show CPU load. If anything, increasing resolution decreases CPU load. Higher resolution = lower FPS = lower CPU load.

 

What my understanding is, is that increasing resolution increases CPU hit per frame, but decreases CPU load overall by forcing a GPU bottleneck.

It's like... if you took a 4.5GHz haswell i7 and two 980Ti cards and ran something at 720p then at 1080p, you're probably gonna find you get less FPS without actually maxing the cards at any point.

The increased CPU hit is so far negated by the reduced CPU load (due to GPU bottleneck) that people don't consider it.

 

In what way? Increasing resolution is an entirely GPU-bound operation.

 

Doesn't the CPU still require drawing those "thin frames" first? Shouldn't a larger "thin frame" help? I've seen more CPU leeway by dropping resolution before, where I was CPU bound (and not GPU bound) at 1080p I would drop to 720p, keep the same GPU util and get lower CPU usage.

 

What in the world is a "thin frame"?

 

I don't know the name of it. But you know what I'm talking about. The CPU renders a very basic frame and pushes it to the GPU to fill out. If the CPU can't render fast enough to utilize the GPUs, then you hit a CPU bottleneck. If the GPUs can't fill out all the frames the CPUs are pushing, then you hit a GPU bottleneck.

 

Well the CPU handles all the game's simulation and draw calls for each frame. It doesn't actually render any of the pixels, that's all on the GPU. So resolution is entirely GPU-dependent. But if the CPU needs to update its simulation more times per second because the GPU is able to output a higher FPS due to lowering resolution, then CPU usage increases.

 

I really need to do some more testing of that, then. I need to find a game that's CPU-limited AND locked to say... 60fps (and dips below 60 due to CPU bottleneck) then drop it to 720p and see if it still dips below 60.

If higher resolutions don't use more CPU power at all, then regardless of resolution, a CPU bottleneck is a CPU bottleneck. The parts where I drop from 60fps should still drop at 720p, because the CPU is the bottleneck, not the GPU. Does that make sense? If I don't get CPU limited at lower resolutions the same as at higher ones, then it should prove higher resolutions use more CPU power.

 

Yes

 

Okay, when I stop being lazy I'll check with some game or other XD

 

Okay, so I stopped being lazy and I checked using GTA V. Seems there's a VERY SLIGHT improvement between 1080p and 720p (using windowed mode). At 720p the GPU util plummets and I get ~2-3 frames more on average. I was just standing at one spot for quite a while. Minimum FPS was 62 and max was 68 at 1080p, and minimum FPS was 64 and max was 71 at 720p. It was so slight it was very little to be concerned about, but I'm not sure what the jump to 4K would be like. I was looking for one of those instances where I got locked to about 44fps, but apparently they patched it and it runs better now. I was legit amazed at the performance I was getting; I didn't go under 60 when I was trying to find a place to go under 60 XD

 

So this proves you're CPU-bound in GTA V

 

Oh I knew I was the whole time, which is why I said 3.5GHz is not enough for GTA. But those underwater features where I was using 15% of the CPU? No.

Also, funnily enough... GTA V can use legit 99% of the CPU (and I do mean ~98-100% of all 8 threads; throttlestop saying C0% is 99%) when first starting up. After it loads its title screen etc it seems to get locked to ~62% in-game. XD