Need a volunteer to confirm some test results for me....

[PiLsY]

I need to waste an hour of some kind souls time out there .

If anyone with the following spec is reading this and fancies lending a hand plz could you pm me .

Nf-7S v2.0
Bios 10 (original shipping)
Non vmodded board (or mods disconnected)
Capable of running 200FSB plus at a 12x multiplier with dual channel.
Knows how to use memtest, sandra, 3dmark.

It should really take less than an hour - ive done all the groundwork and really only need another set or 2 of reported results. Nothing complicated .

Thx,

PiLsY.

[Nohto]

Pilsy I have a v1.2, but not a 2.0. I'm headed out the door right now, but I will be back in about 4 hrs and I could run it for you then.

[b0bd0le]

well, im planning to buy a rev2 board this week so in about a weeks time i can help ya out

just curious, but what for?

[PiLsY]

Nothing groundbreaking, but im not willing to spill the beans until I know im right .

Nate - that would make it 5am here when you get back - ill be ordered off to be by she-who-must-be-obeyed far before then m8 . Im really after rev 2.0 owners, but sure if you want to try then it would give me a chance to see if what ive found is specific to the rev 2.0 or happens across other NF7 revisions as well. I could really do with confirming on someone elses NF7 v2.0 to make sure what ive foundisnt just specific to my board though. All very vague and mysterious, but you know me . Wont speak at all until I know for sure that im not speaking out my arse .

PiLsY.

[PiLsY]

Thx Mickey .

PiLsY.

[SPQQKY]

I've got nothing better to do right now.

[Jupiler]

PiLsY :
check your PM.

[bito]

PiLsY

Drop me an e-mail/PM and i'll get to it.....
NF7-s V2.0 .... two of them actually.
All above criteria ok.

Cheers

[beatnic]

hmm.. i've got a v2.0 but with bios11

[SAE]

My rev2.0 is being rma-ed cause I could not do same cpu as on my rev1.1, what means 250Mhz less.

And then testing samsung CTL dual (ran 230 2,5-5-3-2 at 2,8V stock on 1.1) didn't indeed led me to over 200 stably.

So this board must be exchanged (no answer though from rma dep)

Oh I am so bored...

[PiLsY]

Thx for all the offers guys . Ive pmed jupiler with instructions, so we shall see what he comes up with. If his results confirm mine ill create a new post outlining what ive been playing with and the pm I sent him so you can all give it a try .

Thx again,

PiLsY.

[BrainStorm]

PiLsY,

Even if Jupiler's board doesn't show it, what if it's something that does affect some boards but not others? Shouldn't you just let everyone know so you can get results from a larger sample size?

Even if it doens't affect anyone's board but yours, I don't think anyone minds if you end up, as you say, talking out your arse...

[WxChaser]

Even if it doens't affect anyone's board but yours, I don't think anyone minds if you end up, as you say, talking out your arse...

Offtopic - that just cracked me up bigtime here, BrainStorm!

[PiLsY]

lol fair enough and a point well made .

Pasted below is my PM to Jupiler from earlier today . Try it if you like, report results up if you do.

Paste below :

======================

I think ive found something weird with the mem timings and just need it checking out. Ive got all the important bios settings written down - i'll copy them in below so you can set to match.

Memory in dimm slots 2 and 3. Twinmos w/twinmos chips.

Softmenu 3

12 x 202FSB
66mhz AGP
1.7vDD
1.8vcpu
2.9vddr
1.5vagp
FSB/Dram 6/6
CPU Interface enabled
Over temp 100c

Advanced Bios features

Apic enabled and MPS version 1.4 set.

Advanced Chipset Features

System bios cacheable : disabled
video ram cacheable : disabled
AGP Aperture size : 128mb
FSB Spread spectrum : disabled
AGP Spread spectrum : disabled
CPU Thermal throttling : disabled
Enhance PCI performance : Enabled
AGP Transfer rate : 8x
Fast Writes : Enabled

Integrated Preipherals

Onboard IDE : All Auto/Enabled
Onchip PCI : USB Disabled / Audio Enabled / Lan Enabled / Boot Rom Disabled / IEEE1394 Disabled
Onboard PCI : SATA Disabled
Init display First : AGP
FDD Controller : Enabled
Serial Ports and Parallel ports all disabled

Power management

ACPI - S3
Button : Instant Off
PCI Wake : disabled
Ring wake : disabled
Alarm wake : disabled
Power On : Button Only
AC Power Loss Recovery : Power Off

PCI/IRQ Resources

Update ESCD : Disabled

Resources to manual
IRQ Resources all set to PCI

PCI/VGA Palette snoop : disabled
Assign IRQ to VGA : Enabled
Assign IRQ to USB : Disabled
PCI Latency Timer : 32 Clocks
PIRQ_0 : 7
PIRQ_1 : 5
PIRQ_2 : 3
PIRQ_3 : 11

PC Health
CPU Shutdown : Disabled
CPU Warning Temp : 80c/176f
Shutdown on fan fail : Disabled




Now I dont doubt most of those are already set as I have them set, but I have to be thorough with this really. All I really need are memtest bandwidth results, but sandra buffered/unbuffered and 3dmark would be helpful too if you have time. The memtest stuff is the important one anyway.

Now what ive found in testing is that lower Tras isnt always better. Infact when youre set to cas 2 /trp 2 / trcd 2 settiong TRAS to 5 drops your bandwidth in memtest by some 150mb/s compared to higher values.

If you could set your ram timings to 2/2/2, then start at Tras 4, and work all the way up to tras 15, booting into memtest and writing down its reported memory bandwidth that would be enough to confirm my results. My results are below for compare.

Timings 2/2/2 @ 12 x 202. Variable Tras.

Tras 4 : 1508mb/s
Tras 5 : 1508mb/s
Tras 6 : 1558mb/s
Tras 7 : 1583mb/s
Tras 8 : 1583mb/s
Tras 9 : 1583mb/s
Tras 10 : 1610mb/s
Tras 11 : 1610mb/s
Tras 12 : 1639mb/s
Tras 13 : 1610mb/s
Tras 14 : 1583mb/s
Tras 15 : 1583mb/s


Ive found out that where the bandwidth peaks is down to many, many factors. Settings affecting where your peak bandwidth arrives are :

Vcore
vDD
vDDR
CPU Multiplier
FSB
Dimm slots in use.
timings of other settings (trp, trcd and cas)

Lots that affect it. Once youve sorted the results out for me have a play and see if you can grab a bit of extra bandwidth. Sandra doesnt report much difference (30mb/s unbuffered and buffered), but 3dmark does. Ive seen a 300 point spread between tras 4 and tras 12.

Anyway, get back to me if ive confused you on anything, im anxious to hear what results you get .

==================


PiLsY.

[CodeRed]

PiLsY,
I noticed that too (Tras bandwidth effect), but didnt record the results

I could simply be that the mem timings are more "matched" to the instruction sequence (and prefetch pipeline depth) that memtest uses.

Setting the Tras low defintely gives better performance on real world apps ... just try pifast @ 3-2-2-2 & compare with 7-2-2-2 timings.

I think that the importance of bandwidth measures are really overrated. Latency is far more important for application performance.

[Ewok]

i posted in another thread earlier the same thing but my test used sandra unbuffered. http://www.xtremesystems.org/forums/...313#post134313

i found i scored higher with 8 than i did with 4. didnt bother to test anything else. this was at 219fsb DC 222.

[PiLsY]

CodeRed - pifast times go up, seti times go down, folding times go down, 3dmark goes up, dvd ripping time goes up, super pi stays the same.

Its very much a trade off. Apps which are relying on pure memory speed tend to go down, whereas apps relying on constant data feeds benefit. The like sof seti, 3dmark et al rely on a constant rate of data, rather than speed of delivery. Stuff like pifast and dvd ripping is all about speed of delivery rather than quantity. I suppose its best to match it to what your pc does most of.

PiLsY.

[BrainStorm]

PiLsY, very interesting. Do you really think it's confined to NF7-Ss? Have you tried with an Epox board, etc.?

[BrainStorm]

Originally posted by WxChaser
Offtopic - that just cracked me up bigtime here, BrainStorm!

I was just repeating what PiLsY said in his second post in this thread...

[Swiftdeathz]

I've also noticed this on my 8RDA+.... When testing the stability of my system with Memtest86/Sisoft2003 my bandwith score would drop when I went CAS 2-2-3-5 over CAS 2-2-3-6....

I figured it must have been my RAM crapping out and not liking that tight of a timing but guess I was wrong... :p

[BrainStorm]

I screwed around briefly with my 8rda with the following results at 203x10:

4-2-2-2........3079/2888
6-2-2-2........3082/2893
8-2-2-2........3082/2900
10-2-2-2......3089/2905
12-2-2-2......3089/2905
14-2-2-2......3085/2904

So at least in sandra, 10-12 did the best. Too tired to try other benchies tonight.

[Jupiler]

I'm still benching and testing. I hope to have results by this evening. But it takes a lot of time.

[Holst]

I couldt be arsed testing many settings, but 13 was higher in sandra than 5.

Just shows what a POS sandra memory is :P

[PimpJack]

what effects ram stability the most. I'm having problems with 225fsb and 6-3-3-2 timings and maybe this will be able to help me without loosing performance. Is it tras or the other two values and how do they effect performance. I will take a look at some results using v1.2

EDIT: I just did some tests but using my bios settings and 2.1v at 220fsb

tras(4-13) 1754mb/s for each one and a drop occured above 13.
14-1722
15-1692

Sorry with my board and settings this theorey diddn't work out. Or maybe the diff was so small that it didn't show up. I was hoping for some nice scores though

[MrLavender]

Hmmm, I've just tried this on my v1.2 and I think you want to be careful about Memtest86 bandwidth results. I saw no difference in tRAS settings 4-12, but sometimes I would get a result of 1551mb/s and other times 1612mb/s (@ 200 DC) with the SAME tRAS setting...