It's also difficult to cast blame on a product using a means that leaves nothing but conjecture as to why the failure occurs. By mistunng VCORE, VTT or NB voltage I can produce a failure in Linpack just as fast as bad memory will. And then there is overall system stability and proper memory timings.
Ouch man, ouch.Quote:
Originally Posted by Cronos
Ok, now I'm in business!
I bought an eVGA 780i board and finally, the 8G are stable!
Now only that, on my previous 680i I could never get an FSB above 1400. This board, is a piece of cake with no playing around with voltages at all (except CPU), I can boot into windows at 1800MHz and run a bit of prime 95 with no problems so far.
Of course I have alot of work ahead of me to ensure stability.
I will do the linpack in honour of Cronos!
But the memory now passes memtest at 1000 MHz 5-5-5-15 2.1V
I'll keep everyone posted, and maybe make a new thread on 8GB of this ram.
ok guys, I have a question...
I just migrated to vista x64 so I can now access all my available memory and I'm considering the move to 8gb.
My question is which will perform better:
4 gigs at 1200+ MHz (which my tracers do with ease)Or
8 gigs at 1000 MHz
???
What do you think? 8 gigs of ram for under $250 sounds like a screaming deal to me!
oops
Keep what you got, results vary. I cant get much more than a 1000 with these, but its probly my 650i. I had to back down to 437x9 and lose 100mhz cpu to get these to run at 1000. I'm going x38 soon. 8gb is an e-peen thing. I run games servers on Vista at times and I think I've seen usage go as far as 1.5g free in sidebar once.Quote:
Originally Posted by sonofander
Great, keep us posted.Quote:
Originally Posted by tomb18
By the way, i am very interested in 780i memory performance, i still can't find any information if there is any improvement over 680i.
Try to run Right Mark Multi-Threaded Memory Benchmark (RMMT), which is part of
RMMA suit
http://cpu.rightmark.org/download.shtml
Compare single-threaded BW with multi-threaded.
Don't forget to set at least 30MB per each thread, default 1MB is only good for CPU cache read benchmark.
Cronos, I've been playing around with different FSB's, multipliers and linked vs unlinked etc. I've been looking at bandwidth as displayed by everest. Some of the FSB's and settings give really high latencies like 74ns, other 54 or so.
It seems when I up the FSB to 400 MHz (CPU=3.6 GHz) , memory performance seems to drop alot. Like 8500mb/s in everest (read). On the other hand, I now have the system with an FSB of 360 (CPU = 3.6 GHz) and at a memory speed of 1080 MHz I'm getting 9900 MB/s. Quite a difference.
I haven't quite figured it out yet.
I'll try the right mark.
Ok, The multithreaded version gives an average on 4 cores of 6130MB/s using a 50Mb per thread.
On a single thread the average is 8400 MB/s
At what memory speed?Quote:
Originally Posted by tomb18
I forgot to mention, set Read W/Prefetch and prefetch distance=256.
There is some improvement as i see, 680i multi-threaded BW used to drop almost two times vs single-threaded. On the contrary, P35 improves about 20%, having almost 100% advantage over 680i.
Ok, with all threads running and the prefetch set on 50mb sets, the average read is 6226MB/s. It's 8400 MB/s on a single thread.
This is with 4 threads.
This is at 540 MHz (1080) 8GB
I have ~10 000MB/s /4 threads on P35 at 1000MHz
As you see, P35 has almost two times more multi-threaded bandwidth. both 680i/780i are not suitable for serious memory-intensive multitasking.
If someone thinks results showing super_pi 32m equate to 100% rock-solid stability, then they are looking to be disappointed. Most reviewers don't have a month worth of hours to perform a review to show 100% stable overclocks, so we do the best we can with the time we have, and super_pi generally fits that niche.Quote:
Originally Posted by Cronos
Hey guys,
Thanks for all the info posted in this thread, lots pf good info here.
I'm using this RAm with a GA-X38-DQ6 and Q600. I can't seem to get them stable with an hour test on OCCT. Has this been an issue for anyone else and should I just use prime 95 for stabililty testing? My CPU seems rock solid with Prime 95 small FFT's for 10 hours. Whats the best way to go from here? And is it worth getting rid of these for the PC2-8500 2 x 2 GB?
Thanks for the help
I'm surprised more people aren't testing their memory bandwidth and latency. I see many configurations posted here that are slower than DDR2-800 speed with a lower set performance level (tRD).
Has anyone achieved <58 latency in Everest or >9000 in memory read with this G.Skill on a P35 board? If so, what were your settings?
I have on a 780i board. Highest stable speed so far is around 9200-9800 MB/s with a letency of around 54ns. This is at 5-5-5-15 2.1V and speeds between 1000 and 1033. Speed on this platform so far seems mostly dependent on the FSB. FSB's under 1600 give much beter performance. Once over, speed drops to under 8500 MB/s
All reported by everest.
This is with 4x2GB.
This is the best I was able to do at stock voltage with a Gigabyte DS3R + E8400.
2000FSB vs 1800FSB
http://img127.imageshack.us/img127/1...ckvoltaws1.png http://img91.imageshack.us/img91/264...ckvoltaoo0.png
Memset screenshots for 2000FSB and 1800FSB.
The problem with P35 is that you cannot increase the FSB without increasing the tRD (known as performance level in Memset), and tRD has by far the most significant impact on memory performance.
Has anyone been able to attain significantly better results on a P35 board with this memory?
Use 6:5 divider with PL=7.
What PL are you on with these two settings?
I linked to the memtest info below the screenshots. The PL is 6 for the 1600FSB and 8 for 2000FSB.Quote:
Originally Posted by Spyrus
I'll try as you suggested.
Edit: I updated the previous post with 6:5 divider screenshot (right).
Well, I found that 8GB of this ram is stable at 1000MHz 5-5-5-15 and 2.1V with 200 iterations of Linpac. As Cronos mentioned, this is a uber stability test. It makes my quad hit 74C on water (prime 95 never gets more than 60C).
However, to do so, I had to lower my multi on my QX6700 down to 9 from 10. With an FSB of 1400 I'm taking a hit down to 3.15 ghz whereas before with 4G and a 680i board I ran stable at 3.5 GHz.
I haven't figured out why yet but is it normal that with 8G you would have to lower the cpu speed? I would have thought that I would have had to lower the ram speed.
The problem is, if I raise the FSB much mnore then memory perfomance really takes a dive.
My congrats!Quote:
Originally Posted by tomb18
Just one more question, to clarify the matter, it passed 200 iterations with 30k problem size with all residual equal to the last digit?
Cronos,
Opps, I started it again and it overwrote the file!
I've brought up the fsb to 1380 to try and get back some of my speed.
We'll see how it does.
Temps are hitting 75C!
I'll let you know how it goes.
tomb18,
Just want to remind you, bluescreens&hangs are rare and extreme events, meaning your configuration is far from being stable.
Error-free operation is what you should be after. Check the Residual(Norm) values, they should all be equal down to the last digit.
You can open the *.bat file in any text editor and set any desired name for the output file for your convenience.
On Asus maximus formula the 450 fsb @ 540-5-5-5-x works easily and withQuote:
Originally Posted by Ken_F
low nb voltabe ~ 1.35 v with performance level = 7. Latency is ~53+ ns.
So your bandwidth is accepted with those settings.
If you can push memory to 550 they will have same latency as with tRD=PL=7@540
You can try to tighten these timings:
tWR,tWTR,tRRD,tRTP = 6-3-3-3
Quote:
Originally Posted by Cronos
Well I did a quick set of 20 and the residual norm values give two sets of different numbers, half are identicle and the other half are identicle. Should all of the residuals be the same?