PCMark Vantage World Records

[Anvil]

My 5770 had no issues @119.
I haven't really tried pushing the 9260/11 yet, all I know is that it works at 104.
I once forgot to reset the pcie from 115 and needless to say, the 9211 did not show up.

What do you normally set IOH to?
I've been using up to 1.24 while doing pcie > 115. (bios F7j)

[mbreslin]

I never touched it before and just started at 1.24 10 minutes ago, 107mhz was fine, 110 no 9260

Any idea max safe ioh?

Went all the way to 1.34 didn't seem to matter (though everything else shows up which is cool) 9260 won't come alive after 108 mhz, still a few more than max I was able to do before 104.

[Anvil]

I'm not sure about the max IOH, could be that the PCIE needs to be pushed up a few notches as well.
I've havent really tried pushing it yet but I expect 1.60 would be safe for the PCIE. (been using 1.54)

[Computurd]

those voltages sound about correct.the most i can get from the 9260 is 107 as well....

[mbreslin]

111k hdd suite score.

Done with that subtest, on to the next.

[GullLars]

Hi guys, this is slightly off-topic, but i'm posting a notice here in hope more people will notice i'm doing this.
I'm working on a little project to map storage sollution scaling by QD for 4-32KB blocks in the QD 1-32 range, wich are mostly the relevant block and QD ranges for random reads on desktop and most types of workstation. I'm hoping as many people as possible will contribute by running my IO config (takes about 20-25 minutes). I'm looking for results for any type of storage, from a 10 year old 5400RPM PATA drive, to a first gen JMicron SSD, to a velociraptor, Intel SSD, and 2-8(+) SSD/HDD arrays. Basically the more diversity i get the better. 1-N scaling of arrays would also be great (although i know this would require breaking them up, so i don't expect too much).
I already have a bunch of results for 4KB random read for SSDs, but this is a much larger project to better understand storage scaling, and illustrate it with graphs for all interrested users on XS and diskusjon.no (i'm getting results from both forums).
Link to the post with the config and more info: http://www.xtremesystems.org/forums/...0&postcount=39

[mbreslin]

31252

push uncore to the brink
go with 5870 only super oc'd
I think I could go 4.9-5ghz with my chip

With these 3 things I honestly believe I'm leaving at the very most 1.5-2k on the table. As this is still under mike (and well under the 40k he says is needed to truly beat him) and close enough to steve that he would pass me in short order I see no reason to continue. I very much wanted to pass corsair, as it stands now I got very close with 1ghz less cpu clock and 6 less drives (the sweet spot for me with 9260+fp for pcmv seems to be 5 drives). I can live with this.

If the 1880 turns out to be a massive performer AND I can sneak it past the wife I may give it another shot. (I would say this is split 70/30 right now 70 being the unlikely) For now back to 24/7 settings!

I spent about 8 hours a day the past few days tuning. Every time a run finished and I would get 30k and change I would think more and more just what a massive score 35k is and he says he can get 40k. Amazing.

As for 8 c300s it's anecdotal but I do seem to be getting a better general use experience with a 5r0 array, there was some microstutter I didn't notice before and only realize it now since everything is so perfectly smooth. I guess I will shelve 3 drives for a bit until I find a use for them.

I went from a pentium 4 2.4ghz and 120gb ide spinner to 4.8ghz 980x and 2GB/sec sequential/600mb/sec random r0 array. Pretty big leap!

I learned a ton from all of you and got ideas I never would have thought of on my own. Getting every ounce of performance out of my hardware turned out to be great fun and something I never tried before. Now I wouldn't even think of NOT doing it!

Our second child is coming in a few weeks so I'm likely done spending gobs of time at xs, just wanted to give a tip of the hat to everyone who gave me advice/ideas. Thanks!

That was a long rant!

[SteveRo]

I can confirm a couple of your observations - 5870 (1) super oc'd I found to be best also.
So far - much to my chagrin - I also find that 5 drives are best - interesting coincidence since you are on 9260 and I am on 1231 but I am still playing with allocation (cluster) size - i might yet find a better sweet spot.
Did you take enough time to investigate permutations of drive/stripe/allocation size - so many possible combinations!

[GullLars]

Steve, are you logging your results? If you do, you can set up graphs and 3D diagrams showing 1-2 variables changing at a time.
If you do stripe and allocation size as 2 variables, holding one at each step while scaling the other, you can make a 3D diagram. With relevant stripe and allocation size range, i'd guess you'd end up with 20-30 runs in total. And if you need to test # of drives at each of those runs, you could easily end up with >100 runs. If OC'ing changes the best spots, you would end up with a possibility of several hundred runs.
I think i would settle for spot-testing and then check the with all best spots while changing one variable at a time one step up and down, and possibly another good spot just to make sure the best is selected. That way you could narrow it to a fraction of the number of runs.

BTW steve, care to contribute to my little project i linked to?

[mbreslin]

@Steve: I tried all combinations of stripe/allocation BUT I did this with 8 drives, I suppose it could have been different for 5 drives but I didn't care to do it all again.

[SteveRo]

OK, Mike and Planet - wake up guys - time to go again! - http://service.futuremark.com/compare?pcmv=335925

I found a better sweet spot for the oc and the partition - seems like it took forever!

[mbreslin]

Well done as usual. The question in my mind is how far over 40k will you be able to go with your cascade. Will be fun to follow sir. Great job.

[Anvil]

Congrats Steve,
The bar is raised, again

35x136 was better or just easier to get stable?

[pcnazz]

King of the Hill Steve , great job your doing .

[Computurd]

wow! i knew it was coming, just figured it would be with the cascade though, nice job!

[dinos22]

whoa 5x Acards eeeek

im getting the adaptec 5805Z for a bit of a play but all i have are about 8 WD raptors.....i'll see what that does...obviously i got no chance against SSD and acard guys but itll be fun

[SteveRo]

Thanks guys!

@ Mr. mbreslin - yes, i am expecting over 40k too - assuming I have a good subzero chip, we will see.

@ Mr. Anvil - 35x136 needed a little lower v's so lower temps even though the ambient is probably 10C higher!

@ Mr. dinos22 - yes 5xR0 but actually 3acard "boxes" set to Ro via jumpers so each box has 2SATA connections and then using only 5 of the 6.
4 or 5 drives seems to be the optimum number,
believe me i have tried many combo's of # of drive/stripe/cluster.

One interesting observation to pass on -
I could not get pcmv to complete (got n/a) anytime i used other than the default 4k cluster (allocation) size.
I have no idea why other allocation sizes shouldn't work (8k, 16, 32, 64).
FYI - to do this you need to pre-create the partitions prior to installing windows (200MB, 4k partition and then XXGB partition at whatever allocation size you want). Then select the XXGB partition as the install location.

[Computurd]

sorry steve i may be being a bit dense here but i lost you on the part about partition allocation...so if the only size it runs at is 4k isnt 4k the default allocation size anyway? why would you need to pre-create it?

[Planet]

Hey Steve congrats. I am out of the game as I no longer have a controller card ATM. Maybe in the future I'll go back again.

[SteveRo]

@ Mr. CT- for example - I found I was able to get much higher PCMV HDD numbers using 4k stripe with 64k allocation - pcmv suite test would abend on 2 of the subtests - don't remember which 2 - this happend on all but 4k allocation size - don't know why.

@ Mr. Planet - thanks and hope to see you back in it soon - if you had used a 1231 last time I might still be chasing you.

[TheKarmakazi]

Holy . Your builder moves fast. It has taken nearly 4 months for my cascade and he says it will be another month or two yet.

Your not the only one getting a run around from him

[SteveRo]

Paul - Are you making progress with your SS?

I am finding subzero oc to be much different than air or water - sometimes you need to lower voltages to get better oc.

My cascade progress so far -

http://www.xtremesystems.org/forums/...d.php?t=255686

http://www.xtremesystems.org/forums/...&postcount=711

I need to look into this further - for those using or thinking about using Gigabyte boards - how about this - adjust the oc in realtime for each subtest -

http://www.xtremesystems.org/forums/...d.php?t=255513

[Anvil]

Great results Steve
When can we expect a new pcmv run?

Both Cloud OC and HotKey OC looks interesting, downloaded

[SteveRo]

@ Mr Anvil - thanks, regarding a new pcmv run - it is taking me much longer than I expected - it might be a day or two or it might be a month!
This sub zero stuff is harder than I thought - cpu behaves differently - I thought I could just crank the voltage and be done - wrong.
It's really been different from air or water so far.
Sometimes you do better if you reduce voltages!

[Computurd]

@stevie- well it is going well, I have it all together i have just been busy as hell at work, salaried work is a . needing to make time for it.
the GB utility you linked above it what i have been harping about for so long! not the GB specific one , of course, but overclocking on the fly, during the benchmark. i was thinking more along the line of evbot or eleet of course.
that is especially helpful for me though, actually, as i have a GB board on the phase right now so i really appreciate the link! I am also having a learning curve in the little I have done, this GB board is a lot different than the evga 1366 boards I have been using, and going back to 775 suxors as well, I have to brush up on my 775 skills! delaying everything. experience counts! unfortunately i lack it with this setup....but in time....