New GPU-Z Feature: ASIC Quality

[Ryu Connor]

GPU-Z is now able to present an ASIC quality value embedded in the chip at fabrication. This only works on Fermi and Southern Island GPUs.

What does it mean? Not a clue. There is no context available for the value reported from the chip. It is purported to represent how leaky the transitors are, but even that lacks a way to quantify against the value reported.

People are sharing their results. I figure we can join in on the fun.

To access the feature right click on the green adapter card icon in the title name bar of the GPU-Z program. With the context menu now open you should be able to find the ASIC Quality option. If that's unclear here is a picture:



My results | eVGA Classified 590:

[Forge]

Single eVGA GTX 460 1GB, recently acquired. 102.6%. I'll update when my second RMA arrives.

[Ryu Connor]

eVGA Classified 560ti 448:

[CampinCarl]

I'm more interested in how, exactly, they would test for leaky transistors via Windows software?

[Forge]

Campincarl - this data is burned into the GPU by nvidia. GPU-z is not testing anything, just reading a register.

[CampinCarl]

Forge: Ohhhh. Well, I guess that makes sense as part of the QC process. Thanks. Guess I get to look silly for not having entirely read the OP

[Ryu Connor]

No more of these to share? Would love to see more, especially if you OC your card.

[morphine]

eVGA 570, stock: 78.3%

[Crayon Shin Chan]

eVGA GTX470, ASIC quality: 69.4%
This can't be right! It overclocks just as well as the rest of them! 770MHz at 1087mV and 840MHz at 1150mV, the temperatures go through the roof when overvolted though.

[Forge]

eVGA GTX470, ASIC quality: 69.4%
This can't be right! It overclocks just as well as the rest of them! 770MHz at 1087mV and 840MHz at 1150mV, the temperatures go through the roof when overvolted though.

ASIC quality might be a bit of a misnomer. So far it looks like its a mixture of at least two metrics, one being lithographic clarity (chip quality) and the other being transistor leakage. The leakier your GPU, the better it'll clock, but the more voltage it'll need to do it, and the more voltage your chip dumps as heat.

[just brew it!]

eVGA GTX470, ASIC quality: 69.4%
This can't be right! It overclocks just as well as the rest of them! 770MHz at 1087mV and 840MHz at 1150mV, the temperatures go through the roof when overvolted though.

Leakier transistors would result in more heat when the device is overvolted, so this sort of makes sense.

[Forge]

Just judging from the *entirely* informal reading/collating I'm doing, it looks like the overclocking sweet spot is between 60% and 80% for Nvidia cards. Lower than that, the chip quality is iffy and leakage is insane. Mostly seeing very low percentages on downbinned SKUs, too, like the GTX465 and 470 from GTX480, and the GTX560Ti448 and GTX570 from GTX580.

The very high quality numbers correspond to late runs, I have a 100%+ GTX460 that was made recently, and all the other 100%+ cards I've seen were late in the production cycle.
My OCing seems average to good, but not exceptional, my stock voltages are very low, and my heat is pretty much the same at stock or high OC, and doesn't rise sharply when I overvolt.

[Compton]

My GTX460 that I bought for a uATX rig weighs in at a hefty 125.1%. Which just sounds impressive. I hope a higher percentage is better.

It's a newer GTX460 that I bought in December for a 2500K/Maximus IV-Gene Z travel rig.

[Forge]

My GTX460 that I bought for a uATX rig weighs in at a hefty 125.1%. Which just sounds impressive. I hope a higher percentage is better.

It's a newer GTX460 that I bought in December for a 2500K/Maximus IV-Gene Z travel rig.

I think that might be a new record. I think the highest I'd seen until now was about 116-118%

[Compton]

My GTX460 that I bought for a uATX rig weighs in at a hefty 125.1%. Which just sounds impressive. I hope a higher percentage is better.

It's a newer GTX460 that I bought in December for a 2500K/Maximus IV-Gene Z travel rig.

I think that might be a new record. I think the highest I'd seen until now was about 116-118%

Here is a pic:

It's a base model EVGA too. I don't even think the warranty is one year on those models.

[Rand]

My Gigabyte GTX 460 purchased just after launch has an ASIC quality of 43.1%, one of the lowest I've seen.
VCore is a pretty typical .975V, it does overclock fairly poorly though 820MHz. So that could be indicative of the lower quality ASIC.

Counter intuitively, it does run extremely cool however compared to results I've seen around the internet. This is tough to compare though, likely more indicative of case ambient temperature and stock cooling on specific cards then anything else though.
Still, even during the heart of summer my card was running much cooler then I'd have expected.

[Forge]

My Gigabyte GTX 460 purchased just after launch has an ASIC quality of 43.1%, one of the lowest I've seen.
VCore is a pretty typical .975V, it does overclock fairly poorly though 820MHz. So that could be indicative of the lower quality ASIC.

Counter intuitively, it does run extremely cool however compared to results I've seen around the internet. This is tough to compare though, likely more indicative of case ambient temperature and stock cooling on specific cards then anything else though.
Still, even during the heart of summer my card was running much cooler then I'd have expected.

At 102%< I default to 0.9325V at max 3D, much less at idle. Clocked to 800 core effortlessly, did 850 unstable. Flash SSC, got 850 solid with 0.95V 3D. Can do more, but temps ramp sharply at even 0.95V.

[Compton]

My 125.1% 460 had never even been overclocked before. I stopped after hitting 855MhZ, but I did throw a lot of voltage at it just in case. The 460 is pretty beastly at those clocks... but above 900 is really good.

Correction: It's stable up to 925 if you want to cram a lot of voltage in it. I actually think the VRMs are probably holding it back. I have another GTX460, so I'll have to check to see what the ASIC % is on that one.

Are we certain ASIC quality isn't just for easily determining individual chip voltage? The 125.1% has a stock V-core of .937 and stock clocks of 763MhZ. My other, older 460 is sure to be much lower.

[Ryu Connor]

Are we certain ASIC quality isn't just for easily determining individual chip voltage? The 125.1% has a stock V-core of .937 and stock clocks of 763MhZ. My other, older 460 is sure to be much lower.

We are going by simply what W1zzard (the author of GPU-Z) has presented. I presume his sources are good, but only NVIDIA, ATI, TSMC, or one of the board partners (EVGA, MSI, etc) could tell us exactly how to quantify it. I know we have a few students on this forum going through engineering courses for microprocessor design. Not sure if their classes discuss a concept like this at fabrication or not.

[TurtlePerson2]

I am really confused as to what the ASIC quality number actually means. I read the linked post and I still am not sure what it actually refers to. "How leaky the transistors are" shouldn't vary by as much as people are reporting, a Vt shift of 3x isn't really possible without the circuit failing.

[Forge]

Compton: More or less what Ryu said. We don't *know* anything, W1zzard found a register with some interesting data in it, made a quick percentage conversion, and put it up for what it's worth. Everything beyond that at this point is guesswork. Some of it's very educated guesswork, and we can partially test and validate it by comparing, but only Nvidia knows what Nvidia means by it, and same for AMD. I wouldn't even expect a flat statement of it's purpose from Nvidia to match precisely with AMD's interpretation of how it should be used.

[Compton]

Compton: More or less what Ryu said. We don't *know* anything, W1zzard found a register with some interesting data in it, made a quick percentage conversion, and put it up for what it's worth. Everything beyond that at this point is guesswork. Some of it's very educated guesswork, and we can partially test and validate it by comparing, but only Nvidia knows what Nvidia means by it, and same for AMD. I wouldn't even expect a flat statement of it's purpose from Nvidia to match precisely with AMD's interpretation of how it should be used.

You're right. I should have said, "is it possible"? Every GPU get's it's own independent voltage settings at the factory, right? Maybe it's just a way to facilitate that. It's cool, whatever it is.

[ShadowEyez]

62.0% - I got my 460 w/1 gig ram soon after the card, so I'm assuming early production run with the board itself. I assume the 100%+ stuff was during a later production run. Card still runs well in games, though I really don't overclock it much for long periods of time.

[vargis14]

Well the 2 evga 560ti superclocks i have one is 76.4% and the other is 86.3% they both hit 1020 mhz with a voltage bump.One is slightly warmer then the other by 2-5c

[Forge]

Well the 2 evga 560ti superclocks i have one is 76.4% and the other is 86.3% they both hit 1020 mhz with a voltage bump.One is slightly warmer then the other by 2-5c

Grab Nvidia Inspector and take a look. I'm betting the 2-5C cooler one is the 86%, and runs a slightly lower voltage by default. Might even be the same voltage, but slightly less leakage.

[Hallucin8]

Doesn't work on my work box Quadro NVS 295. "ASIC quality reading not supported on this card."

[Forge]

Doesn't work on my work box Quadro NVS 295. "ASIC quality reading not supported on this card."

Emm, no way to say this delicately, so I'll be blunt.

Did you think your card was Fermi, or did you think it was an AMD 7000 series?

[Hallucin8]

Doesn't work on my work box Quadro NVS 295. "ASIC quality reading not supported on this card."

Emm, no way to say this delicately, so I'll be blunt.

Did you think your card was Fermi, or did you think it was an AMD 7000 series?

Bluntly back, I didn't see anywhere in the post that it required such. Now that I read the GPU-Z revision history, I see that. You win at the internet!

[Forge]

Doesn't work on my work box Quadro NVS 295. "ASIC quality reading not supported on this card."

Emm, no way to say this delicately, so I'll be blunt.

Did you think your card was Fermi, or did you think it was an AMD 7000 series?

Bluntly back, I didn't see anywhere in the post that it required such. Now that I read the GPU-Z revision history, I see that. You win at the internet!

Fair enough. Ryu bolded and enlarged it for the benefit of other latecomers. I recommended blink tags, but TR (wisely) filters those.

[just brew it!]

And in his defense, I would've probably tried it anyway, just to see what would happen!