Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

[Ken429]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

There you go, my 4790K running with defaults (except memory). Kind of makes you wonder why even mess with OC'ing on this processor. To get an increase of ~14 GFlops it takes a max of 0.056V additional Vcore and generates a max core temp of 11C more heat. But at least the GFlops are more consistent since it jumps to 4.6GHz and stays there. I'll let you guys compute the increased percentages!

[parsec]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

I agree about what is the point of an OC if the return is minimal. But it could be said that your OC to 4.6GHz was not much beyond the standard Turbo speed of 4.4GHz.

Yet another thing to consider is when using stock CPU settings, while stress testing with four cores at 100% usage, Turbo will clock the four cores to 4.2GHz. So the difference between stock and a 4.6GHz OC is not 200MHz, but 400MHz, 10% of the 4790K's base frequency, 4.0GHz.

I'm surprised as well about the variation in the GFLOPs between each run. You thought they are more consistent when running a 4.6GHz, since the core speeds are not changing, as they do at stock settings and Turbo changing the core speeds between 4.4GHz and 4.2GHz.

But consider my Xeon, which is running at stock settings, and whose core speeds change with Turbo too, from 4.0GHz to 3.8GHz. My GFLOP results were not inconsistent as yours were. There must be another reason for the greater inconsistency, unless if you tested multiple times and got the same range of GFLOP values. I'm NOT suggesting you should do that. Something like a background task might have been started, and used some CPU resources.

Apply the core speed ratio and GFLOP ratio thing again with the new '4790K data:

4.6 / 4.2 = 1.095

207 / 194 = 1.067

The core speed difference is 9.5%, and the GFLOP difference is 6.7%. This result ignores the inconsistency, and is a poor pay back on the investment, as you said. But how much can you OC a 4790K? How many owners have them at 5.0GHz? Or 4.8GHz?

[Ken429]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

I let the system run for 10 minutes after booting. Ran LinX two more times and got similar results -except- the reverse of what I got the first time I ran it (previous post) with the default settings?

[parsec]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

That is weird, weird, weird. I have no idea what the cause of that is...

But I have read that, I believe it is HWBot, that collects CPU and other benchmarks, has declared benchmark runs in Windows 8/8.1 as invalid. They don't include them their benchmark lists. That is apparently caused by the timer or system clock used by Win 8/8.1 (IIRC) compared to how it is done in Windows 7. Is this related to what we see here?

I sure don't know, but I don't think this is something to be worried about. It's not a thermal throttling problem, your CPU temps are fine and of course higher at 4.6GHz. Interesting how consistent the inconsistency is, just one run out of range. We would need to see other 4790K results at stock speeds to see if they do that too.

[Ken429]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

Final thought on this thread! Is it safe to say that MaxxMem is ready for retirement or is AIDA64 wrong?

[parsec]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

How in the world could we know which is right or wrong? How would we know who is right? How would we know if both are wrong?

The Intel spec of their Haswell i7, i5, and Xeon E3 processors for maximum memory bandwidth is 25.6GB/s. They say that is for read and write. The G3258 is 21.3GB/s, at 1333.

But, is that at the maximum memory speed spec for these processors, 1600? It must be. Your memory is 2400, mine runs at 2000.

We tend to think these things are simple, but they aren't. We have no idea how each program works. We know MaxxMEM is not up to date with Windows 8 and Haswell processors. AIDA64 is, so IMO it is more likely to be correct. But I base that opinion on nothing more than that.

[parsec]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

Gotta throw one more thing into this, which surprised me frankly.

This is my i5-4670K on my Z87 OC Formula board, with a simple OC to 4.0GHz:



183 GFLOPs peak. I was expecting a greater difference between an i5-4670K and an i7-4790K. Note that my CPU is an i5-4670K, not a DC Haswell processor.

Granted my VCore is higher than the i7-4790K, my 4670K is not a great over clocker, voltage wise, and the VCore is not optimized.

So ~10GFLOPs more on the i7-4790K, which ran at 4.2GHz stock Turbo. I'd ask you to disable hyper threading and run a test, but not important.

[Ken429]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

There you go. I changed the Memory to 4096 for what that's worth. Looks like the 4790K likes to run with Hyper Threading disabled! Also, what do I have to do to LinX to get the bigger display?

[parsec]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

I should of left the memory size the same as your earlier runs for comparison, increasing the memory size increases the GFLOPs.

So we don't really know if disabling Hyper Threading causes the increase in GFLOPs. I'll use 2048 and do another run.

To change the display, note that when a test run is complete, in the lower right corner of the LinX window you'll find Log. You can see it in your last post. Just click it and the display will change. When Log is displayed, the word changes to Table, which is the default display. You can toggle back and forth between the two.

Are you aware of the Save Screenshot option in the File option on the upper left side of the LinX window?

[parsec]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

Hmm, so much for that theory...



For some reason the HWiNFO power usage information is crazy low, after enabling higher power in the UEFI.

I just noticed you did do a run a 2048, at the top of page 3, so disabling hyper threading increases the GFLOP results. That might indicate why some gamers say they prefer Intel i5 processors, which do not have Hyper Threading. Actually, they should be disabling Hyper Threading on i7 processors.

Still, the processor speeds are not the same, I should do a 4.2GHz run...

[parsec]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

Now a run of 4.2GHz:



So results are almost identical in GFLOPs to the i7-4790K at 4.2GHz with Hyper Threading on. The similarities end there...

First the core and CPU temps. I'm using a Thermalright True Spirit Power 140, with the Thermalright TY-143 fan. A great CPU cooler in general, but Haswell processors challenge the best. The 99C core temp is so close to thermal throttling, but none apparently occurred. The CPU fan was maxed out at 2500 RPM. I have the board in my CM HAF X case, a more open air type case.

The main thing to notice is my VCore, max of 1.352V! That is Adaptive voltage doing its thing, which can be fixed... somewhat. Is LinX running AVX2 instructions? My CPU Input voltage is not drooping much, and my PSU is rock steady with that load. I need to work on the VCore.

As usual with non-DC Haswell processors, heat and VCore are the limiting factors. Ken has better cooling than I do, plus his VCore is much lower. The low VCore is key IMO, and is what makes DC processors better than the others.

I don't get the inconsistency that you did earlier, actually my results are very tight in min - max range. Your CPU Input varies much more than mine does, you should check your CPU Load-Line Calibration, I have mine set at 2. My VRM temperatures never broke 36C, so the board is not the limitation, nor my PSU. I know my 4670K is not a great over clocker.

[profJim]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

An old but useful LinX guide: How to run LinPack stress test(LinX/IBT)properly-an explanation(maybe a guide)

You can test your system with an older, pre-AVX version of the Linpack libraries, probably an early 0.6.4 version or the 0.6.3 version from https://forum.lowyat.net/topic/892376/all
You should also be able to download and temporarily replace the 0.6.5 Linpack libraries in the LinX folder with an older version of the pre-AVX libraries from 2009 or early 2010.

An ideal minimum to maximum GFlops range is within 0.05 to 0.10 GFlops during each test run.

As I recall, with a Sandy Bridge or newer cpu that supports the AVX instructions, the GFlops scores would be about 60% - 75% higher when you tested with the Linpack libraries that support AVX.

[parsec]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

Oops, sloppy terminology on my part. I should have wrote AVX2 instructions, which are unique to Haswell processors.

The next question is do the linpack libraries in the 0.6.5 version have AVX2 instructions. The dates on the libraries are later than the release of Haswell processors. Since Intel provides those libraries, without further research, I assume they do.

Since I normally don't run these stress tests much, and running AVX2 instructions is realistic for Haswell processors IMO, I won't be changing to the earlier libraries. Also to maintain apples to apples comparisons, using the same libraries is essential.

Your idea is certainly valid and would work.

AIDA64 allows AVX2 instructions to be enabled or disabled. Prime95 pre-AVX2 versions can be found, or may even have an AVX2 switch by now.

That LinX/IBT guide is... interesting. It does implicitly make a good point, there are many variables that affect benchmark results, and every PC will be different in widely varying degrees. Comparing benchmarks and then being worried when ours are worse than others is a waste of time.

Running these benchmarks in Windows Safe Mode would be the simplest way to get a more equivalent environment from PC to PC. Not perfect, but improved.

[profJim]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

Below are the results when I tested my Q9550 @ 4.20GHz in windows safe mode.

This peak GFlops score was 7.5% higher than my posted 4.20GHz results in post #18 where the peak score was 55.4540 GFlops and the number of processes and threads using safe mode was much lower.

[parsec]

Re: Windows Only Sees Sata_0 and Sata_1 on ASR Z97X Killer MB?

Nice, and there we have it, proof that fewer background processes allow a CPU to score higher in benchmarks. Not that this is a surprise IMO.

We can also see in the Safe Mode test that the time of each iteration and the GFLOPs results are very consistent and their range is smaller than in the normal Windows mode. The normal mode test shows almost a two second difference (1.951 seconds) in the shortest and longest iteration, while in Safe mode the difference is 0.261 seconds.

Ken, I am 90% sure the variations you saw in your first stock speed test were simply caused by Windows processes using CPU resources. The time difference you had there is only 0.563 seconds, so not that much anyway.

The speed of the processor will affect the length of each iteration, with slower processors taking longer. That affect caries over to the range of speed variations.

A very rough comparison with the Safe Mode processes and threads seen above, my Win 10 TP with Firefox and HWiNFO running shows 59 processes and 787 threads.