Hi When I run P95 at stock voltages/baseclock it gives me a rounding error within a few minutes on all cores. I think it might be something to do with the memory as I recently added 6GB to take the total to 12GB. All modules are the same manufacturer, with the same rated timings and frequency. I ran memtest86 v4.3.6 but it did not show any errors The spec is as below: i7 920 D0 @ stock 12GB OCZ DDR3 RAM CL9 1333mhz (bios shows memory to be running at 1066, which is roughly 133x8) Gigabyte EX58-UD4 I notice in the motherboard that the stock timings are 7 7 7 16, which appear to be lower than what is printed on the side of the RAM. Should I manually set the timings to something else in the BIOS? 9-9-9 is printed on the side, what should the fourth number be? Any other ideas as to how I can iron out what this issue might be? I want to get it stable at clock before applying a permanent, stable overclock of the CPU to approximately 3.3Ghz help please many thanks All voltages are currently at stock/auto - at a high level these are: Vcore 1.2625 QPI/Vtt 1.175 IOH Core 1.100 DRAM Voltage: 1.500V I have tried upping the DRAM to 1.66V and 1.64V (the RAM is rated for 1.65V), but I still get the P95 fail. UPDATE: P95 failed overnight between 3 and 5 hours on each worker Voltages now: Vcore 1.2625 QPI/Vtt 1.3135 IOH Core 1.100 DRAM Voltage: 1.66V
I think you're right to suspect the RAM timings - that '16' is absurdly low. Try 8-8-8-24, if not 9-9-9-32.
Thanks for the response May I ask how you derived those numbers? Is there a formula, specifically with the fourth number? I'll give them a go now
7-7-7-16 is almost certainly wrong... my corsair stuff is 9-9-9-24... might be worth looking up the SPD timings in something like CPU-Z... even if the board is reading the timings wrong it's a good place to start for setting them manually...
Here are what the SPD and memory tabs are saying with everything on stock/auto in the BIOS Not sure what to make of them to be honest?! Memory says 540Mhz for DRAM Frequency? I thought these were 1333Mhz running at 1066 (133x8)? The SDR tab says a range of numbers in the timing table, which I am unsure of what to make! Any interpretations appreciated! http://imgur.com/JuzmFAl http://imgur.com/ClWINpc
The SPD is reporting the following default timings for 1066 7-7-7-16 @ 533 MHz [1066MHz], 1.5v so I'd set the values to that first and go from there... EDIT: A Quick google points to that kit being pretty temperamental... one of the things that popped up repeatedly and may be worth a try is changing the 'Command Rate' from 1T to 2T
Just to make a note that my gigabyte board doesn't set the voltage to 1.65 either. I have alway's used 1.66 on my corsair with no issues. That kit will need 1.65 and will be unstable @ 1.5.
Thanks for the input so far Update: I have set the RAM timings to 9-9-9-28 2T and upped the DRAM voltage to 1.66 I have also upped the QPI voltage to just over 1.2 P95 has been running successfully for about 15 minutes (which is about 14 minutes longer than it was previously!), so I think I will leave this running overnight to see if it works. Question What is the real-world impact of running this RAM at higher timings? Am I going to lose performance in some way? Many thanks
No real performance hit. Latency used to be a critical factor generations ago. DDR1 and 2 used to see performance gains with low latency, but ever since the move to ddr3, latency is a non issue.
No, those are just based on timings I used to have on my kit. Ah, good point! I forgot about this. Try running 1T first though. As Instagib rightly says, latency timings make bugger all difference unless they are really high. As long as your memory speed is 1333MHz or more, it'll be performing well.
Thanks for your input Update: Overnight I ran P95, however while I was asleep between 3 and 5 hours in 7 of the 8 workers failed again. It seems raising the QPI/Uncore voltage increased the stability, however not quite enough. My strategy is to bump the voltage up a little bit further and re-run the test. Does this make sense? Everything else is held at auto for now, with the baseclock at stock. DRAM is 1.66V and QPI is currently at about 1.3135V Am I missing anything else? Many thanks
Overclocking with 6 DIMMs on X58 is and always will be an absolute pig. You'll struggle to get anywhere near the overclock stability you will just using 3 DIMMs.
I used to run my OCZ Platinum's at 7-7-7-16 with no errors @1440MHz, can easily imagine them running well at 1333. That said it might be worth loosening the timings temporarily just so you can rule that out. Oddly enough I remember trying to run my OCZ's at Cas9 just to see how much of a difference it made from Cas7. Turned out my machine wouldn't boot until I set it back. Is all your ram OCZ?
I think I have it solved! I have just got back and have been running P95 for a total of just over 8 hours, successfully. The increases in the QPI/Vtt voltages is what has increased stability. I think I will tweak the Baseclock to see if I can achieve 3.3Ghz on current voltages, with memory at 1333Mhz (8x166) and see how stable that is. Will try to edge up the Vcore voltage if not. Thank you for all your input fellow bit-techers! Yes its all OCZ 12GB (6x2GB) @ 1333Mhz rated, 1.65V, CL9
I think I have it solved! I have just got back and have been running P95 for a total of just over 8 hours, successfully. The increases in the QPI/Vtt voltages is what has increased stability. The timings are 9-9-9-28 with a command rate of 2T I have also set tRFC to 100 as suggested by Bit_reaper Voltages are Vcore 1.263V QPI/Vtt 1.35V IOH 1.1V Dram 1.64V I think I will tweak the Baseclock to see if I can achieve 3.3Ghz on current voltages, with memory at 1333Mhz (8x166) and see how stable that is. Will try to edge up the Vcore voltage if not. Thank you for all your input, please let me know if I am missing anything
Great, might be worth it to keep going until it's stable for 24 hours but it looks like you are on the right track.
You can run P95 without involving the memory IIRC. I think it was either small FFT or large FFT? something like that. I had a Phenom 9950 with some bad ram (it worked well enough in everything I needed it to) but like you I would fail P95 if I ran a blend test..
Small FTT will keep the data within the CPU cache, which is good for testing CPU stability, but overall you want your whole system to be stable, Including the memory.
Thanks guys, I was using P95 blend which apparently hits the RAM I have run memtestx86 with no errors, so confident about the memory I will do a 24hr run on P95 when I am happy with my final clock speed, aiming to push up the Ghz of the chip I have also got intel burn test and memtest which i will push the system through before signing it off, but raising the voltages on the memory definitely stabilised it on P95. thanks again all