LN2 @ vantage = Fail session

[Splave]

Okay so I have 3 different vantage runs here.
Temps: -105c to -110c constant tonight
(last night I thought I was t -135 -140? now it freezes when that cold)

24357 1370/1200 1.5v 4.96ghz x16 uncore (last night)

24216 1320/1200 1.45v 5.18ghz x20 uncore (tonight)

24039 1330/1200 1.5v 5.21ghz x18 uncore (tonight)


I thought it was maybe overvolting or error correcting last night so I lowered the volts on the gpu and clocks and the scores are worse. Makes sense right?

CCC
CAI advanced
Triple buffering unchecked

Perhaps I just need to reinstall windows? is dual channel killing my score?
No sinks on the vrms an issue? :blush:

[Neuromancer]

Dunno about 5000 series but my 4890 VRMS got hot hot at lower volts then that until I fixed the heatsink... I would have sinks on the VRMs before I put them on the memory.

But I also do not know how well the cold radiates off the GPU pot.. VRMs are back kinda far from the GPU though I thought... couple inches at least.

[Gutterz]

ya u need sinks on the VR's I was told. Nice scores by the way!

[Kal-EL]

I would think that Vantage would like triple channel better since it responds so well to bandwidth but I've never tested that. The other side of the coin is higher cpu clocks with less triple channel load on teh imc which would net higher scores me thinks.

What temp reader/probe you using? Sounds like it sketches out under -100c. Have you tried 06 to see if the card clocks are funked?
Shammy always makes it point to get sinks on vrms.

[RomDominance]

Has to be the triple channel situation Splig. Your cpu scores are a full 1k higher than mine.

[ocgmj]

Mem & uncore look low. Images are hard to read thou.

[Splave]

uncore was 4200 OC

kal temp probe was the standard one that came with the extech. Im thinking its freaking out because Im using both inputs?

I will try to get some sinks on the vrms, and 3x2gb kit of hypers are in transit to rule that out.

[Splave]

106k in 03 at 1330/1200 1.45 so thats seems pretty on point no? Cpu 5.28

[ocgmj]

Down right sick brother.

[Kal-EL]

Nice, you nudged out your romulan buddy but its all good in the oca hood ;)

[HITandRUN]

106k in 03 at 1330/1200 1.45 so thats seems pretty on point no? Cpu 5.28

Sweet clocks mate! :thumbsup:

[crustytheclown]

You're running your pic-e x8 mode man...
bump up the pci-e freq a bit to hit x16 and your scores will be just fine...

[thebanik]

hmmm...it can very well be vrm related, I was told a hot vrm would lead to ....., hmmm, whats the word, aah, I am missing the word, basically the card will downclock automatically...

[Splave]

You're running your pic-e x8 mode man...
bump up the pci-e freq a bit to hit x16 and your scores will be just fine...

Epic find :) thanks man. Im in the 4th slot, guess I should move up to the third :cool3: what a relief!
Thanks for all the help everyone!


and banik I heard about that error correction as well when you clock too high it doesnt crash but gives you worse scores weird stuff. I would rather it freeze up!

[ReverendMaynard]

error detection is on the memory brother, not the gpu so those clocks and volts shouldn't activate it.

EDIT: Just looking into it a bit more, some interesting stuff came up. I need to read a bit more, but this is the jist.

GDDR5 supports a function call "Interface training". Upon power up, the controller and memory train each other for bandwidth and frequency.

Just a 5 second theory of mine, but could you try setting your clocks to boot at the intended bench speed? Maybe the interface training will make some adjustments on a fresh power up that wouldn't happen otherwise?

[Splave]

interesting indeed

[DrNip]

What OS you using? Try Vista 32 if your not.

[Splave]

using win 7 x32, vista x32 faster?

[ReverendMaynard]

I absolutely hate Vista 32b for benching. Kills my cpu score by 1000 points in 06.