LinX & IBT

[Tullphan]

Testing my overclock on my i7 920 (D0) @ 3.5GHz.
I ran LinX 25 times using half (3072) of my available memory (ran for 57m 6s). No problems.
I then ran IBT using Standard stress level for 20 times. No problems. Immediately afterwards, I changed the stress level to "High" & ran it 20 more times. No problems.
What next to makes sure my OC is ok?
BTW, temps never went above 75c for all tests (if that matters)...room ambient was in mid 80's, low 90's.

 

[nevbie]

Sounds fine to me.

LinX seems to find memory instability faster or as fast as HCI memtest, so it seems to be a good test for CPU/NB/RAM alone. I don't think I'll ever need more than LinX and FurMark for stress testing purposes.

 

[terentenet]

You can also Prime for 24h... if you want to make sure you are stable. But that's time consuming and I don't really think you need JUST another CPU test. Your CPU seems stable as it is... or 95% stable.
So... do a PSU test for a change. Get Furmark, run in Stability testing, 1280x1024, 2/4xAA. Not fullscreen. Get in Process Manager, set Furmark Affinity to a single Core and not Core 0 (i set on core 5), Priority on High.
Let Furmark on, now start Prime95 and Blend test for a couple of hours. I did for 24h. My previously presumed stable CPU needed a notch more voltage to complete this test. My IOH voltage also needed a small bump.

This test should create a worst case scenario for the system, being under stress for a long period of time. Your PSU may act if it's not up to the job. My APC UPS shows a 1280W power usage in this test.... but I also have the monitor connected to it.

My system:
i7 975 D0 4GHz w/ HT on @ 1.27v
12Gb Corsair Dominator 1600
Asus Rampage II Extreme X58
3 x Zotac GTX 285 AMP!
X-Fi Titanium Fatal1ty Pro
Asus Xonar D2/PM
2 x OCZ Vertex 120Gb in ICH10R RAID0
2 x Seagate 1Tb ICH10R RAID0
another 2 x Seagate 1Tb ICH10R RAID0
Enermax Revolution 1250W PSU
Everything is watercooled in 3 loops, 5 x 480 rads, 20 Noctua fans on the rads, 7 DDC Ultra pumps (cooling should draw about 200W on it's own).

 

[Tullphan]

Originally posted by: terentenet
So... do a PSU test for a change. Get Furmark, run in Stability testing, 1280x1024, 2/4xAA. Not fullscreen.

"Furmark"? Isn't that a GPU stress-test?

Get in Process Manager, set Furmark Affinity to a single Core and not Core 0 (i set on core 5), Priority on High.

Where do I find process manager & how do I set that program to a single core?

My system:
i7 975 D0 4GHz w/ HT on @ 1.27v
12Gb Corsair Dominator 1600
Asus Rampage II Extreme X58
3 x Zotac GTX 285 AMP!
X-Fi Titanium Fatal1ty Pro
Asus Xonar D2/PM
2 x OCZ Vertex 120Gb in ICH10R RAID0
2 x Seagate 1Tb ICH10R RAID0
another 2 x Seagate 1Tb ICH10R RAID0
Enermax Revolution 1250W PSU
Everything is watercooled in 3 loops, 5 x 480 rads, 20 Noctua fans on the rads, 7 DDC Ultra pumps (cooling should draw about 200W on it's own).

Nice setup!

 

[soccerballtux]

OP-- your overclock is stable. Next up is to see if you can push it further. What about 4.0Ghz?

 

[k0rnh0li0]

funnie question is why did he make this thread if he had all that going. i would have to think nothing would go wrong maybe he can probably even OC more without having problems. 20-25 tests for everything. SOMEONE HAS HELLA TIME

 

[MagickMan]

OC further, keep going until IBT fails on 5 passes, then work with voltage a bit more. Repeat until you reach voltage limits you can live with, and watch your temps.

Once you feel that you're done, run IBT for 20 passes and follow that with 24H of Prime95. If it passes all that, play some games and enjoy.

 

[Tullphan]

Originally posted by: k0rnh0li0
funnie question is why did he make this thread if he had all that going. i would have to think nothing would go wrong maybe he can probably even OC more without having problems. 20-25 tests for everything. SOMEONE HAS HELLA TIME

Because I respect most others opinions on here.
Perhaps I was missing something...perhaps I wasn't running the tests long enough...perhaps I needed to change the settings. I could've missed something.

BTW...going on 12 hours w/P95 & Furmark running simultaneously. No problems thus far.

 

[Rubycon]

Run Prime95 small FFT + OCCT PSU tester (Linpack on the CPU test part of that) and your system will pull max amps guaranteed.

 

[Tullphan]

Originally posted by: Rubycon
Run Prime95 small FFT + OCCT PSU tester (Linpack on the CPU test part of that) and your system will pull max amps guaranteed.

I installed OCCT, but it said DirectX 9 wasn't up to date & it wouldn't support the PSU test & GPU test and that my graphic card didn't support CUDA & GPU Memtest would be disabled.
Is this some kind of bug? Will this program not work w/DirectX 10?
I had an older version of OCCT installed & it did the same thing.
Ideas?

 

[terentenet]

Prime95 small FFT + OCCT PSU test will not draw as much as possible, since the OCCT PSU tester is basically Linpack + GPU test. If you run Linpack + GPU test + Prime95 small FFT, the GPU part will have no clocks left to calculate geometry on. So, the GPU's will not be that stressed. I have tried the most combinations, Prime95 small FFT + Furmark draws the most. I have even tried Intel Burn Test (linpack) + Furmark but again, I don't have enough free CPU cycles left to calculate for Furmark and the GPU's will not be stressed enough.
Go for Prime95 small FFT + Furmark if you want max amps drawn. Prime95 blend + Furmark for overall stability, as it tests memory and everything as well.

Assuming you have both programs (Prime95 and Furmark), that's how you do it.

Start Furmark. Set to Stability testing, select 1280x1024 resolution, 4xAA, NO Fullscreen. Hit Go!
http://i27.tinypic.com/ajumo9.jpg

To start process manager, hit CTRL+SHIFT+ESC. This will appear, I'm sure you've seen it before. Go in the Processes tab, find the furmark executable (furmark needs to be running and heating the GPU's at this stage).
The file you're looking for should be etqw.exe . Right click it, you will see Set priority, select High.
Then, right click etqw.exe again, click Set Affinity..., select one core only, different form Core 0 (most windows processes and services run on that one). So, untick Core 0, select a different one. For example, I select Core 5.
http://i27.tinypic.com/5lt7x3.jpg

Now, you are ready to start priming. Start Prime95, select Blend mode, set the number of threads equal you your processor cores (4 if HT off, 8 if HT on).
http://i26.tinypic.com/21d3div.jpg

Now you can leave the system running the test as much as you want. I left it for 24h to ensure stability.
When I first did this test I thought I was completely stable.
I was 24h Prime95 blend stable.
20 passes of Intel Burn Test (Linpack with GUI) stable at Maximum stress, all 12Gb of RAM tested - Stable (takes time, about 3h here).
It even passed 20 passes of IBT + Furmark, allthough I knew something was not right, the GPU's were not getting hot enough.

Prime95 + Furmark crashed first time at 8h, BSOD. I increased CPU PLL one notch. Crashed even faster, 45min. Decreased PLL back to previous value, increased Vcore one notch. Stable for 15h, then BSOD.
Increased IOH Voltage one notch, 24h 100% stable.

 

[Rubycon]

Originally posted by: Tullphan

I installed OCCT, but it said DirectX 9 wasn't up to date & it wouldn't support the PSU test & GPU test and that my graphic card didn't support CUDA & GPU Memtest would be disabled.
Is this some kind of bug? Will this program not work w/DirectX 10?
I had an older version of OCCT installed & it did the same thing.
Ideas?

You need to install directX 9.0c libraries for this to work.

Originally posted by: terentenet
Prime95 small FFT + OCCT PSU test will not draw as much as possible, since the OCCT PSU tester is basically Linpack + GPU test. If you run Linpack + GPU test + Prime95 small FFT, the GPU part will have no clocks left to calculate geometry on.

Actually it does and I've verified this with a circuit that had Dranetz power monitoring hardware. Linpack on OCCT only loads to 50% because it's borked. The PSU test with Furmark+Linpack was pulling a nice load however when P95 was also engaged it went about 90W higher. I suppose one could play with combinations of these along with process priority and squeeze every watt out of the system. Don't forget if you have lots of mechanical hard drives and an intelligent host to stress those with IOmeter as well!

 

[Tullphan]

Originally posted by: Rubycon

Originally posted by: Tullphan

I installed OCCT, but it said DirectX 9 wasn't up to date & it wouldn't support the PSU test & GPU test and that my graphic card didn't support CUDA & GPU Memtest would be disabled.
Is this some kind of bug? Will this program not work w/DirectX 10?
I had an older version of OCCT installed & it did the same thing.
Ideas?

You need to install directX 9.0c libraries for this to work.

OK..did that. Working now. Still don't know about the "CUDA" thing.

Originally posted by: Rubycon

Run Prime95 small FFT + OCCT PSU tester (Linpack on the CPU test part of that) and your system will pull max amps guaranteed.

OK...did that @ 3.6 & 3.7GHz for 3 hrs each. I'll try for 3.8GHz overnight. All this is w/stock voltage (which I was hoping would be less).

 

[faxon]

i just tried OCCT PSU tester + p95 4 threads on my Q9650 and it showed the same if not lower draw than just running the PSU tester (they were within 5w of each other), so i doubt it makes much of a difference. food for thought my 9800GTX 512mb, 5 Fans, 2 Spinpoint F1s, 2 7200.10 Barracudas, a Vraptor, my Q9650@4ghz with 1.296Vcore, my UD3P, 2 keyboards, and a mouse, all in all drawing about 400w (475w @ ~83% AC-DC efficiency) as measured with a kill-a-watt.

 

[Rubycon]

Originally posted by: faxon
i just tried OCCT PSU tester + p95 4 threads on my Q9650...

Could be possible since my results came from 8 threads.

 

[terentenet]

I am telling you, OCCT PSU tester + Prime95 doesn't make any sense, it's Linpack + Prime95 + a GPU test that doesn't stress the GPU's as much as Furmark does. Prime95 + Furmark all the way... max amps drawn this way, because the GPU's will draw as much as they can. I measure the watts from the UPS providing current to the PC. It clearly draws most with Prime95 and Furmark. Goes up to 1350W during small FFTs.
With OCCT PSU tester, max I've seen was 1180W... and the GPU's don't get as hot as with Furmark, constantly 5-6 degrees under Furmark.

 

[Rubycon]

What GPU and how many of them do you have? What CPU and how many do you have? I was pulling over 8 amps at 240V. :Q

Does furmark support SLI? The one in Occt does and pushed mine pretty hard. I may have an outdated version. I don't get to update my software as much as I should.

 

[terentenet]

I have 3 GTX 285's and an i7 975 D0 @ 4GHz with HT on (8 threads). My rig also has 4 HDDs, 2 SSDs, 1 optical unit, 2 sound cards, 7 DDC Ultra pumps, 21 x 120mm fans, 2 x 60mm fans (on the memory), 3 x UV led strips, 1 x UV Neon kit. I've seen it pull max 1150W on itself (only PC connected to the UPS), 1280W with monitor connected as well.
I have Furmark 1.70, it supports SLI, it scales nicely with the 3 GPU's and heats them all up.
What PSU do you have that can sustain 8amps @ 240V? That's like 1900W.

 

[Rubycon]

Ok I definitely need to get a newer more update version of Furmark as I have 1.50 and it was not supporting SLI.

The power supply is an Ultra X3 1600W. I was incorrect - the 8A figure was at 120VAC not 240VAC which I am usually running on. I don't have that system running any more but it had 15 HDDs 10 which were 15K drives. That's why I mentioned (in humor) IOmeter as it makes a difference when you thrash 10 15K drives. I currently use four SSDs and almost (yeah almost!) miss the gentle sound of sleet hitting a metal roof sound when working hard! :laugh:

UPDATE: Latest version of fur mark I could find is 1.70 released just this month.

 

[Tullphan]

Originally posted by: terentenet
Prime95 small FFT + OCCT PSU test will not draw as much as possible, since the OCCT PSU tester is basically Linpack + GPU test. If you run Linpack + GPU test + Prime95 small FFT, the GPU part will have no clocks left to calculate geometry on. So, the GPU's will not be that stressed. I have tried the most combinations, Prime95 small FFT + Furmark draws the most. I have even tried Intel Burn Test (linpack) + Furmark but again, I don't have enough free CPU cycles left to calculate for Furmark and the GPU's will not be stressed enough.
Go for Prime95 small FFT + Furmark if you want max amps drawn. Prime95 blend + Furmark for overall stability, as it tests memory and everything as well.

Assuming you have both programs (Prime95 and Furmark), that's how you do it.

Start Furmark. Set to Stability testing, select 1280x1024 resolution, 4xAA, NO Fullscreen. Hit Go!
http://i27.tinypic.com/ajumo9.jpg

To start process manager, hit CTRL+SHIFT+ESC. This will appear, I'm sure you've seen it before. Go in the Processes tab, find the furmark executable (furmark needs to be running and heating the GPU's at this stage).
The file you're looking for should be etqw.exe . Right click it, you will see Set priority, select High.
Then, right click etqw.exe again, click Set Affinity..., select one core only, different form Core 0 (most windows processes and services run on that one). So, untick Core 0, select a different one. For example, I select Core 5.
http://i27.tinypic.com/5lt7x3.jpg

Now, you are ready to start priming. Start Prime95, select Blend mode, set the number of threads equal you your processor cores (4 if HT off, 8 if HT on).
http://i26.tinypic.com/21d3div.jpg

Now you can leave the system running the test as much as you want. I left it for 24h to ensure stability.
When I first did this test I thought I was completely stable.
I was 24h Prime95 blend stable.
20 passes of Intel Burn Test (Linpack with GUI) stable at Maximum stress, all 12Gb of RAM tested - Stable (takes time, about 3h here).
It even passed 20 passes of IBT + Furmark, allthough I knew something was not right, the GPU's were not getting hot enough.

Prime95 + Furmark crashed first time at 8h, BSOD. I increased CPU PLL one notch. Crashed even faster, 45min. Decreased PLL back to previous value, increased Vcore one notch. Stable for 15h, then BSOD.
Increased IOH Voltage one notch, 24h 100% stable.

Looks like mine ran for 8 1/2 hrs. before I got the BSOD (started it @ 6:20pm, PC was in sleep mode when I got up @ 4:45...best I could tell by my P95 log it lasted till 2:50am).
So my best bet will be to increase CPU voltage a notch? Looks like my IOH is set to "AUTO".

 

[Rubycon]

Running both simultaneously does not really prove you have instabilities at the previous settings. It shows that your voltage settings are slightly too low and the extra draw causes sufficient sag to cause an error or BSOD. Upping it just a little does no harm and would ensure under the "worst operating load scenario" that you do not get an error or bring the system down. (Providing your radiators stay dust free!)

 

[Tullphan]

Radiators?
I don't have no stinking radiators!

 

[Rubycon]

Originally posted by: Tullphan
Radiators?
I don't have no stinking radiators!

Sure - the fins on heatpipe towers when clogged will drive temps way up.

 

[Tullphan]

Ahhhh...THOSE radiators. hehehehe

 

[terentenet]

Tullphan, increase the CPU voltage by a notch. See if that helps. If it still fails, when does it fail? Much sooner or later? If it fails under 5 hours, go back to the previous setting and increase IOH instead. If it keep on going for more than 12 hours, it means it helped. Don't go too high on the voltages. Make sure your cooling can keep up. When I do this test, I do it with Everest started and I keep an eye on the temps.
Also, IOH might help as well, it helped me. Can you see your current IOH voltage, on the Auto setting? Set it manually at that. If the CPU voltage doesn't help, or doesn't bring the stability you want, increase the IOH by a notch.

PS. 8 1/2 hours is nothing to laugh at. That's considered stable in my book. You will most likely never get to stress the system like that under normal usage, for such extended period of time.
Go for 24h stable only if you don't have anything else better to do.