SLI'd 680s - different clocks.

[phil4]

I've two EVGA 680GTX SC's.

I've just recently installed EVGA's precision X tool, and noticed that one GPU is running slower than the other. I need to set the GPU Clock Offset to +40 on the lower of the two, +9 on the higher to get the same speed.

So, I just wondered from those with a little more experience and wisdom than me, how come two cards, same make model, and purchase time have differing clock speeds?

Thanks for any info.

[kompukare]

So, I just wondered from those with a little more experience and wisdom than me, how come two cards, same make model, and purchase time have differing clock speeds?

Thanks for any info.

Well, like I said when Kepler came out: Boost is basically overclocking (far more so than Turbo clocks found in Intel and AMD CPUs) and as always, overclocking is YMMV (that is Your Mileage May Vary)

So basically, all silicon is not created equally. Which is where the term binning comes in. For instance take an Ivy Bridge CPU. Say Intel only make one die for dual core IB's and then bin them, fuse parts off etc. As well as that they also bin parts of voltages - so AFAIK an ULV i3 starts off on the same die as any i3, Pentium, Celeron etc. On any wafer there will be dies which vary: you could have some bad ones which fail totally (rejects), some which fail at certain speeds and some which are able to work fine at voltages a fair bit below the rest (those ULV CPUs in Intel's case). In effect those perfect dies have lower leakage etc. and we can call those 'golden samples'.

Nvidia's boost is actually fairly clever in that it monitors voltages, load, temps etc. (arguable cleverer than the boost in HD7950 GHz etc. but wait for the caveat...). Problem is that with each die potentially varying (and remember silicon power consumption varies with temperature: a hotter chip (poor cooling etc.) consumes more power at the same speed and voltage), so it's quite possible (almost expected) that two Kepler cards will perform differently.

Nvidia being Nvidia it would not surprise me if cards shipped to reviews were 'golden samples', but a greater problem is that unlike Intel's and AMD's Turbo cards are likely to be throttled from their maximum boost a lot more easily since NV Boost varies a lot more with smaller increments so if someone's case doesn't cool that well (or even winter vs summer) the performance will vary. So even someone who has the exact same spec PC as a review used is very unlikely to score the same as the reviewers did.

[Bonebreaker777]

Sorry, wrong post.