Haswell Preview Benchmark

[db597]

Originally Posted by ldesnogu

Seeing my 4770K without OC reaching more than 140W for a few seconds reminded me how little TDP means.

Probably this stems from the difference between long and short duration power draw. CPUs are allowed to exceed for a short while, before they start throttling down. Unless your motherboard is loaded with overclocked settings, you should see throttling after a few seconds of 140W TDP.

Concept makes sense under normal usage. Most of the time CPU is idle and running cool. When you launch a program, it turbos up to process the request. Since you're starting from a cool state, the heat generated can be allowed to overshoot the power draw for a few seconds. By then the task is completed, and the CPU powers down.

[kladner]

At the minimum, get a Hyper 212 Evo. They are dirt cheap and way beyond either of your current options. Stock coolers are nowhere near being designed for sustained full load. Anything with 92 mm fans is a least a couple of generations back, and will be loud if it can remotely keep up. I haven't yet gone with water because something which would improve on my current air cooler costs well over $100, which I just don't have.

Still, my system (FX 8350 @ stock, GTX 570, and 580) is hardly quiet (DUH!). The CPU cooler (Hyper 612 PWM) is massive, hard to work around in the case, and has two powerful, non-stock, 120 mm fans in push-pull. I would be better off in several ways with water, but my sequence of acquisitions did not lead me in that direction at the right time.

Still, I kept the CPU under 60 C in all but the hottest (90+ F) midsummer conditions. The OC'd GPUs almost never surpassed 70 C. I don't much care for Midwestern US winter, but even the current cool spell has the CPU hanging around 50 C. As things get cooler, the heat output will be increasingly welcome. and it will get much quieter in my corner.

[Prime95]

Originally Posted by TheMawn

Stock cooler, stock settings, can't hold 100% load? Honestly, WTF? How does that make any bleeding sense?

That's the reason I don't feel at all guilty RMAing the CPU. Intel sold a CPU/cooler that is supposed to run at 3.4GHz and it doesn't.

I fear I have a CPU with poor TIM inside. On various forums, I've read these aren't uncommon. If so, it doesn't matter what CPU cooler I use, overclocking results will be poor. I'm not at all comfortable delidding the thing.

My TX cooler was running with two 92mm fans. Assuming the stock cooler can run at 3.4GHz, I figure the slightly-better-than-stock cooler should run at 3.8 (no turbo).

[Mark Rose]

Originally Posted by Prime95

My TX cooler was running with two 92mm fans. Assuming the stock cooler can run at 3.4GHz, I figure the slightly-better-than-stock cooler should run at 3.8 (no turbo).

The stock cooler can't run sustained at 3.4 GHz. Just mprime LL had my CPU in the 90's, and throttling. I would pick up the Hyper 212 Evo. I paid $25 for it. A friend of mine with an overclocked 4770K uses the same cooler with great success.

[TheJudger]

Originally Posted by TheMawn

Stock cooler, stock settings, can't hold 100% load? Honestly, WTF? How does that make any bleeding sense?

Originally Posted by ldesnogu

Seeing my 4770K without OC reaching more than 140W for a few seconds reminded me how little TDP means.

Are you both sure the CPU runs within Intel specified operating values? My Gigabyte Z87X-UD3H doesn't do so with default settings. It does a four core turbo of 3.9GHz on my 4770K while it should do at most 3.7GHz 4 core turbo as long as it stays within TDP. If current power draw is over TDP for more than a few seconds it should lower the clock rate (turbo). I know that some Intel desktop boards aswell as some Supermicro boards will operate this way. In this case the stock cooler is able to keep the CPU within "save" temperatures. But AFAIK many consumer boards ignore the Intel guides and allow to exceed the TDP aswell as to do "more turbo" than specified. Obviously these boards will run fine for most use cases.

Originally Posted by db597

Probably this stems from the difference between long and short duration power draw. CPUs are allowed to exceed for a short while, before they start throttling down. Unless your motherboard is loaded with overclocked settings, you should see throttling after a few seconds of 140W TDP.

Concept makes sense under normal usage. Most of the time CPU is idle and running cool. When you launch a program, it turbos up to process the request. Since you're starting from a cool state, the heat generated can be allowed to overshoot the power draw for a few seconds. By then the task is completed, and the CPU powers down.

Read above.

Oliver

[AG5BPilot]

Originally Posted by Prime95

That's the reason I don't feel at all guilty RMAing the CPU. Intel sold a CPU/cooler that is supposed to run at 3.4GHz and it doesn't.

I fear I have a CPU with poor TIM inside. On various forums, I've read these aren't uncommon. If so, it doesn't matter what CPU cooler I use, overclocking results will be poor. I'm not at all comfortable delidding the thing.

My TX cooler was running with two 92mm fans. Assuming the stock cooler can run at 3.4GHz, I figure the slightly-better-than-stock cooler should run at 3.8 (no turbo).

If you have a sharp razor blade and a huge pair of cajones, you could try this:

http://www.youtube.com/watch?v=XXs0I5kuoX4

The video is really eye opening if it's not faked (and I don't suspect it is). Replacing the thermal paste reduced temps by 40 degrees. At least some of the IB and Haswell chips seem to have terrible internal heat transfer properties, and probably the best of them are only mediocre at best.

Even if pulling the top off your brand new expensive CPU isn't for you, seeing this video gives you a good idea of what the problem is -- and why no reasonable external cooling upgrade is going to fix it.

[kracker]

razor blades seem risky, you could try vice+hammer+wood
http://www.youtube.com/watch?v=hu03beZlOjw

[ldesnogu]

@TheJudger: it was the same on my ASUS Z87, but I have disabled it. The 140W I got are within spec, it's called PMax in Intel datasheets.

[Prime95]

Originally Posted by TheJudger

Are you both sure the CPU runs within Intel specified operating values?

I googled for quite a while and could not find the Intel "blessed" default voltage.
On "Auto" setting, CPU-Z reports 1.1V and turbo speeds were a modest 3.6GHz.

The motherboard is http://www.newegg.com/Product/Produc...82E16813128599

[TheMawn]

Originally Posted by ldesnogu

Seeing my 4770K without OC reaching more than 140W for a few seconds reminded me how little TDP means.

My GTX 670 and GTX 660 Ti would spike to 650W according to both power draw monitors I was using. This happened right before a crash. The 140W TDP (though you later did mention that this is within Intel's very-short-term spec) MAY not be entirely accurate. CPU's don't necessarily have drivers but whatever is actually driving them might have some limitations in very short term measurements.

Originally Posted by Prime95

I googled for quite a while and could not find the Intel "blessed" default voltage.
On "Auto" setting, CPU-Z reports 1.1V and turbo speeds were a modest 3.6GHz.

The motherboard is http://www.newegg.com/Product/Produc...82E16813128599

The default voltage differs from CPU to CPU. There are two different voltage control methods. The first is the constant voltage setting. You set 1.300V and you get anywhere between 1.250V and 1.350V but it stays there. The difference is this lovely thing called Load Line Calibration which tries to compensate for the fact that the voltage will drop when the power draw becomes higher.


The second method is the offset mode, which is good for power and heat purposes. This is what I use. I can't remember off the top of my head, but with my current load line calibration etc settings, it's something like 0.250V. When the CPU is idle, it runs at whatever frequency (1600 MHz for me as windows has this feature which allows the CPU to downclock itself when under no load) and at a lower voltage. When the CPU revs up, it adds the offset value, which is different than 0.250V for the same reasons mentioned earlier. That's why you do an awful lot of testing, which is neat with any OS-level overclocking like you can get with ASUS AI suite (the rest of the software is crap and I have considered uninstalling it many a time.

Long story short, I run at just a hint below 0.900V (!!) at idle and then up to 1.280V like I currently am, with Prime95 going hard on all four 4.6GHz cores. For me, then, the "default" voltage is the 0.895V or whatever the hell. It's the "lowest" voltage that the motherboard will be physically allowed to feed the CPU. I think the default standard voltage control is some safe amount on top of that, which has shown stability under Intel's own tests. Probably liquid cooled in an air conditioned dust-free zone with air compressors instead of fans, but whatever.

[Prime95]

Which do you think is better:

load FFT value
mul by DWT weight       Latency 5
add carry                     Latency 3

or 

load FFT value
fused-mul-add              Latency 5