Haswell Preview Benchmark

[Prime95]

Originally Posted by kracker

Is 3360K FFT a efficient FFT?

It is for me when running one worker, one thread.

Try doing a benchmark using the options described in undoc.txt

I think this would work:

MinBenchFFT=3360
MaxBenchFFT=3584
OnlyBench5678=0
BenchMultithreads=1
BenchHyperthreads=1

[Ken_g6]

Originally Posted by Prime95

Yes, results are disappointing. I don't know if it is the chip or Apple's OS hampering it in some way.

My Haswell at 4GHz, DDR3-2400 does 4 2880K FFTs at 14.55 ms/iter. A 2880K uses 2880K * 8 bytes of memory, which is read twice and written twice plus about 5MB of sin/cos/weights data. That is, 8 * 2880K * 4 + 5 = 95MB, bandwidth = 95MB/iter * 4 workers / (.01455 s/iter) = 26.12GB/sec.

The Macbook Pro 2.3GHz, DDR3-1600 does 4 2560K FFTs in 16.8 ms/iter. Bandwidth = (8*2560K*4+5) * 4 / .0168 = 20.23GB/s.

One would expect the DDR3-1600 to deliver 1600/2400 * 26.12GB/s = 17.4GB/s. So the L4 cache does help, but much less than I hoped for.
[/CODE]

I missed this the first time around: The Crystal Well cache is slower than DDR3-2400! :surprised

Originally Posted by henryzz

Sorry misread your post.
Seems strange that this is so different to the bandwidth gained in http://www.anandtech.com/show/6993/i...950hq-tested/3

I noticed something in Henry's link:

Originally Posted by AnandTech

Crystalwell is tracked by Haswell’s PCU (Power Control Unit) just like the CPU cores, GPU, L3 cache, etc... Paying attention to thermals, workload and even eDRAM hit rate, the PCU can shift power budget between the CPU, GPU and eDRAM.

So maybe with the new K Broadwell with Crystal Well, if the power limit is turned up to 11 (decawatts if you must know), then the cache might run faster.

[xtruder]

Originally Posted by TheMawn

Hah. 12-phase VRM on a Mini-ITX when the second-best is 6 and the rest are 4. That's why I like ASUS the best. I still filter out non-ASUS boards when I look around.

Iam wondering, how does one find this information? Looking at gigabyte (z87mx-D3H) they do not list it for this particular board.

[kracker]

Originally Posted by xtruder

Iam wondering, how does one find this information? Looking at gigabyte (z87mx-D3H) they do not list it for this particular board.

Just look at a picture of the motherboard or check a review. In that motherboard, it is 4 phase.

[TheMawn]

Originally Posted by xtruder

Iam wondering, how does one find this information? Looking at gigabyte (z87mx-D3H) they do not list it for this particular board.

What Kracker said. This is one of the less exciting features on a motherboard and that is why it is not displayed with all the other information. Essentially, each phase is a capacitor in series with an inductor, which are devices which attempt to maintain a constant voltage and current, respectively, which is critical for a CPU. More phases means better quality of power, but that is only really relevant as far as overclocking is concerned.

[kracker]

Originally Posted by TheMawn

What Kracker said. This is one of the less exciting features on a motherboard and that is why it is not displayed with all the other information. Essentially, each phase is a capacitor in series with an inductor, which are devices which attempt to maintain a constant voltage and current, respectively, which is critical for a CPU. More phases means better quality of power, but that is only really relevant as far as overclocking is concerned.

Hmm. I thought it had something to do with stability as well although that is true: etc what someone of another forum said something like this:

"With 4 phase you can get up to 5GHz or some overclock, but don't expect your motherboard to last long at all"

[TheMawn]

Well, your overclock won't be stable if the power input is not stable. Your margins should of course be larger than 0.01V but if the fluctuation of voltage is too high you'll be having stability problems on one end and component damage on the other. Same idea regarding current.

The idea behind having more phases is that some of them can be off and "cool down". I think my motherboard has 12 but the power control profile (ASUS, man...) I have set up doesn't have them all active at once.

This is where people get into fancy motherboard cooling. My motherboard actually has a built-in waterblock which I will try to incorporate into the loop I will be building soon with school winding down. If I was completely bonkers and into 1.5 volt overclocks on an Ivy Bridge, I would need to set up a performance-minded control profile on my VRM's and use a more aggressive cooling policy (i.e. water).

[xtruder]

Originally Posted by TheMawn

More phases means better quality of power, but that is only really relevant as far as overclocking is concerned.

Well i could be showing symptoms of poor power design. Wish i considered that option when choosing this board.

First, i cant put more than 1.15v on the chip (4770k), which is quite unusual. I cant p95/linpack stress any clock above x40 at that voltage mark, chip just gets unstable. Yet, i can boot 4.7 @ 1.2v and run something lite like cinebench or sp 32M, just nothing heavy. Putting 1.3v wont even post at stock clock.
I had to settle on x4.3 @1.1v which could run latest p95 and and latest LinX builds for days.

Second symptom, are self correcting WHEA errors, event id 19. These only happen during lite loads, usually when running youtube in the background or 1 VM. Those errors slipped by me, because i never had bsod in 6 months and only checked event viewer where the logs are kept for 7 days.
Frequency of archived errors:
http://www.swe-gaming.com/fsx/files/...80039_whea.PNG
http://www.swe-gaming.com/fsx/files/...perational.PNG

I still haven't got a single bsod since i done the overclocking. System is used for Fem analyse, large CAD file conversions along with video encoding.
Perhaps someone experienced anything familiar?

[TheMawn]

If the Prime95 stress test is good for a day or so, you're probably in good shape. The turning point for me is 4.8 GHz on my 3570k and it is so severe that running Prime95 stress test will crash half the processes in task manager within seconds (Prime95 encounters no errors though ) whereas 4.7 GHz is completely fine. I had 1.24V 4.7 GHz perfectly fine and 4.8 GHz 1.4V (just to see, I had no interest in staying there) was still a mess. I can run 1.4V 4.7 GHz fine, but it's too hot. Sadly my BIOS will boot at 4.6 GHz no matter what I tell it, but that's still fine by me.

That does sound like power problems. Depending on your board, you may be able to tweak around with the power profiles. There may be a current limit, you might be in some energy saving power delivery profile, I don't know. But be careful as I would bet that loosening the power limits will only deteriorate your board that much faster for little gain.

This is pushing the limit a bit here, depending on the size of your network of friends, but if you happen to know someone with a 4770k, or even any Z87 processor, you could try swapping processors and see if you experience the same thing. I have a funny feeling your chip is perfectly fine, and it will last forever at such a low voltage.


On the other hand, your stable frequency is at the point where even the most robust of memory systems is going to be hard pressed to keep the CPU fed, so there would be little Prime95 gain in increasing the speed. If you need the speed for other applications badly enough, you might consider a motherboard upgrade...

If you decide to keep your current hardware (my recommended choice), see if you can throttle the voltage back some more. 4.3 @ 1.1 V is a very healthy choice for your chip's sake, but if your power delivery can't handle 1.2 V, I might be a bit concerned about operating 24/7 under load at 1.1 V, but that's just me.

If you have a lot of time on your hand, you could see if you can get 1.08 or 1.06 volts stable. If you have a LOT LOT LOT of time on your hands you could see if 4.2 GHz at something stupid like 1.05 V is possible. It all depends on your confidence in your power delivery. I am likely erring on the side of caution, personally.

Also don't discount the possibility of a bad PSU. If you have a beefy enough GPU, you could try running a GPU benchmark while running Prime...

[kracker]

Originally Posted by xtruder

Well i could be showing symptoms of poor power design. Wish i considered that option when choosing this board.

First, i cant put more than 1.15v on the chip (4770k), which is quite unusual. I cant p95/linpack stress any clock above x40 at that voltage mark, chip just gets unstable. Yet, i can boot 4.7 @ 1.2v and run something lite like cinebench or sp 32M, just nothing heavy. Putting 1.3v wont even post at stock clock.
I had to settle on x4.3 @1.1v which could run latest p95 and and latest LinX builds for days.

Second symptom, are self correcting WHEA errors, event id 19. These only happen during lite loads, usually when running youtube in the background or 1 VM. Those errors slipped by me, because i never had bsod in 6 months and only checked event viewer where the logs are kept for 7 days.
Frequency of archived errors:
http://www.swe-gaming.com/fsx/files/...80039_whea.PNG
http://www.swe-gaming.com/fsx/files/...perational.PNG

I still haven't got a single bsod since i done the overclocking. System is used for Fem analyse, large CAD file conversions along with video encoding.
Perhaps someone experienced anything familiar?

Also, the average overclock for a Haswell is ~4.3 Ghz. Yes, they suck at OC.
Also, Prime95(FMA3) and Linpack(AVX2) is very intensive, so that's normal.

EDIT: Yes.. you probably should have gotten a better MB.

[henryzz]

It would be interesting to see what you could run at with voltage below default(maybe 0.9). Since you will be memory limited there shouldn't me much of a difference in prime95 performance but you should be able to save a fair bit of power. It would be interesting to work out at what clock rate your memory becomes saturated at and then the lowest voltage you can run at that.