UD3LR 8gb memory crash

[Chike]

Re: UD3LR 8gb memory crash

I seem to be have CAS stable at 7, 6 produces errors in WMD and makes things run a little slow(which seems odd to me). While i tweak these settings should i be increasing any voltage or anything, ill be working on static tRead later.

No need to increase any voltage if they are stable. Lost track what your settings are now, can you post latest settings?

It would be a good idea to run a memory benchmark, everest or other, with CAS 7 and auto to seer the 7 actually made things better and not worse. There is no point to risk stability if there isn't any real improvment.

[milkman6453]

Re: UD3LR 8gb memory crash



WDM is quick to show errors and it is always correct atleast for me.
I also use prime95 and memtest.
So far im pretty stable. Changing static tReads within the hour
what is a memory benchmark i cant find one

[Chike]

Re: UD3LR 8gb memory crash

So this is the best stable you got so far? What the MCH Core and DRAM Votage?
Could you do a memory benchmark (everest or other,) so we can see how it is from performance point of view?
And what is WDM if you don't mind?

[Psycho101]

Re: UD3LR 8gb memory crash

As wierd as it may sound (it's actually logical) your tight settings for sub timings could be effecting the RAM's ability to run CAS 6. This is why I don't advocate using an XMP. It's fine for default XMP timings, but when tweaking, manual gives you control.

As an experement, I would manually set all your 5 subtimings (not the more complex ones, unless you want to) looser. Be sure to test after, to make sure you have an improvement over CAS 7 with XMP. Set 5-5-8-68-5 at 2T and give CAS 6 another go. I only say this, because in my experience (and it's ONLY ime) if a set of modules is run at too low a CAS it simply won't boot. I'm not saying you can make it stable, and also not saying tht increasing those sub timings might reduce the performance gain over CAS 7 with tighter ones, but the fact that CAS 6 boots at all gives hope. Increasing tRAS a little may also help stabilize CAS 6. 6-6-6-21 will definitely out perform 7-7-7-17. Noticeable to you? Maybe, it depends. Improvement in synthetic tests and SuperPi? Almost certainly.

Edit:** WDM = Windows Driver Model. I believe Milkman may be experiencing crashes/errors relating to it. Unstable |RAM can lead to a host of seemingly random errors. Probably memory space allocated to WDM or a driver is becoming corrupt at CAS 6.

[milkman6453]

Re: UD3LR 8gb memory crash

no no no SORRY WDM is windows memory diagnostic its like memtest in 8 minutes :) ill run a bench now

how do i run the benchmark i click it and it goes to some database, im using everest

with current settings stated in above post with vDRAM at 1.6 and Mch Core at 1.2 here is my bench mark
static tRread still at 7 and 7

Latency is 68.2 ns
Memory read is 7558 MB/s
Write is 7084MB/s
Copy is 7387 MB/s

[Psycho101]

Re: UD3LR 8gb memory crash

Windows Memory Diagnostic is WMD.

You may find it more satisfying to use SuperPI to bench with in combination with a synthetic program. It's not exactly the kind of thing you'd use on your machine every day, but it's a little more real world with regards to indicating performance. There's no substitute for testing with what you do most though.

If you game, do some gaming benchmarks. Avoid 3DMark, as those scenes are scripted. there's no AI calculations going on in the GPU tests. Use a game that is predominantly CPU limited to test how it effects frame rates. To see if the CPU speed is a definite bottleneck for the GPU, use a more GPU limited game such as Far Cry 2. Crysis isn't my favorite for benching, it runs like a dog on most non nuclear powered rigs. I bought |Crysis when it first came out, and haven't been able to play it. Can't stand 24 FPS in Max. The problem is, because it's designed to be "photo realistic" anything other than max settings looks crap (to me).

[milkman6453]

Re: UD3LR 8gb memory crash

What should i calculate pi to, for a proper benchmark? i did 1 million digit is avg of 20 seconds
2 million in avg of 44 seconds
when doing 4 million temps went up to 55-65c in 1 minute 49 seconds
i dont know i dont feel safe doing 8m i like low temps

[Chike]

Re: UD3LR 8gb memory crash

Memory read is 7558 MB/s
Write is 7084MB/s
Copy is 7387 MB/s

If taken with everest they don't look bad at all, better than mine at 333FSB.
I had 7167 read 7026 wite 7186 copy (higher than read?) and 70.4ns latency with tDDR-800 4-4-4-12 tRD 6.
I don't know how DDR3 should compare to DDR2.

[Chike]

Re: UD3LR 8gb memory crash

What should i calculate pi to, for a proper benchmark?

1M is used for comparison.

i did 1 million digit is avg of 20 seconds
2 million in avg of 44 seconds
when doing 4 million temps went up to 55-65c in 1 minute 49 seconds
i dont know i dont feel safe doing 8m i like low temps

Hmmm scores not so good, and temps are high.
You should be around 17seconds for 1M I think at 3GHz. No other application is running during the benchmark? Watch with CPU-Z that CPU speed is 3000MHz when the test is running.

[Psycho101]

Re: UD3LR 8gb memory crash

The 1m test is what I use, purely because it's quick. I know then that if it's +-13 sec that I've gained or lost a little. Always run it a few times. 5 runs sits well with me.

Something like the 32M test may vary less in proportion to how many digits are being calculated though.

[milkman6453]

Re: UD3LR 8gb memory crash

static tRead 6

Memory read is 7755 MB/s +~200
Write is 7089 MB/s +5
Copy is 7448 MB/s +~70
Latency is 65.2 ns -3 seconds

yes i was running things when i ran the pi thing here is 1m without running things
16.5 seconds
but core speed was at 1999 for a few seconds in cpu-z then like at loop 5 it moved to 9x multiplier and 3000.
btw im using hyperPI 0.99b

times seem to range from 16.5 to 20 changing the priority from normal to realtime

[Chike]

Re: UD3LR 8gb memory crash

static tRead 6

Memory read is 7755 MB/s +~200
Write is 7089 MB/s +5
Copy is 7448 MB/s +~70
Latency is 65.2 ns -3 seconds

yes i was running things when i ran the pi thing here is 1m without running things
16.5 seconds

Those results look right, maybe if you try what Psycho101 said and manage to get CAS 6 you'd get some more 200MN/s read.

but core speed was at 1999 for a few seconds in cpu-z then like at loop 5 it moved to 9x multiplier and 3000.
btw im using hyperPI 0.99b

times seem to range from 16.5 to 20 changing the priority from normal to realtime

Looks like same results for 3GHz 333FSB but use the super_pi_mod-1.5 that comes with it directly, it seems to slow it down.
The CPU will only go up 3000MHz for the period it is undr load, when it is idle it will drop to x6 multiplier 2000MHz. That's the EIST (Enahanced Intel Speed Step technolofy) causing it to reduce power consumption and reduce heat.
It's perferctly normal, just wanted to make sure you on x9 multiplier.

[milkman6453]

Re: UD3LR 8gb memory crash

CAS 6 is unstable even with lower sub settings i tried 6-6-6-6-17 and 6-7-7-7-17
sub settings were set to physco request i got a BSOD each time page_fault_in_non_paged_area or similar.

I kept the sub settings that physco said but im getting lower benches in everest, could we tweak those now, or is there something else to try for a lower CAS?

also using superPI_1.5 i get about 15.66 seconds

[Psycho101]

Re: UD3LR 8gb memory crash

If using CAS 7, they can definitely be tweaked. Try lowering each one in turn until you regain the performance that XMP gave, then go a little lower to see what happens.

To give an example of how I use SuperPi, I've just done two tests. The first shot is of tSuperPi on a 1M run with XMP enabled, everything else at stock and auto. The second is with a manually selected divider, main and subtimings and with a tRead of 4 set (to match the XMP run). Sorry about two posts for this but it's forum restrictions.

[Psycho101]

Re: UD3LR 8gb memory crash

Second pic, XMP disabled, manual timings. Note I executed SuperPI three times for each lot of settings, and chose the highest result given. Note they are both as said, taken at stock 2.8GHz. Sorry for the lack of CPU-Z in the pic, I forgot to open it.