ASIC Quality

[JCNiest5]

How do you guys interpret a GPU-Z's ASIC Quality? I realize on the same window where it shows the ASIC Quality number, there are some descriptions below it, but I still don't understand it.

My GTX 760's ASIC Quality is: 77.7%.

What does that mean?

 

[Dreamerbydesign]

How do you guys interpret a GPU-Z's ASIC Quality? I realize on the same window where it shows the ASIC Quality number, there are some descriptions below it, but I still don't understand it.

My GTX 760's ASIC Quality is: 77.7%.

What does that mean?

It's a decent card, is ASIC scores matter to you. Many people do not care.

 

[dragonstongue]

ASIC score is guesstimation nothing more. had 2 7870s both have the exact same score, 82.9 and yet the one I kept overclocks and overvolts, underclocks and undervolts by a significant margin more, if it is actually doing what they say the quality does this would not be the case.

 

[bloodypulp]

High ASIC score = High voltage leakage, Less voltage needed for Clock
Low ASIC score = Low voltage leakage, More voltage needed for Clock

A card with either score can still be a good Overclocker, or a bad Overclocker.
The score is just a guideline to adjust the voltage appropriately.

 

[xp3nd4bl3]

I haven't been able reliably map any graphics card characteristics to the ASIC score. And I've tried. To me it means absolutely nothing.

 

[Liger88]

High ASIC score = High voltage leakage, Less voltage needed for Clock
Low ASIC score = Low voltage leakage, More voltage needed for Clock

A card with either score can still be a good Overclocker, or a bad Overclocker.
The score is just a guideline to adjust the voltage appropriately.

The more you know. For once a higher number isn't better. I was under the impression that it was opposite. Guess that helps explain why my GTX 780 @ 73% OC's pretty damn well. Probably could take it higher but I haven't really bothered as 1188Mhz is good enough for me.

 

[dragonstongue]

it is a generalized number, if they actually had a very specific build list of parts used for that card in question then asic quality would matter much more. high leak may need less volts but typically is not as stable and runs much warmer due to this higher volts, low leak well the volts do what they need to do easier but also need much more volts to hit a higher clock but also general stay cooler in the process.

I think it matters more that you get a good card as in MSI or whatever have gotten a nice core or on average they clock up nicely with lower volts decent temps. It matters little unless you say take a 780 as mentioned and they average a score of say 73% and all are within x volts x temps at given clocks then the score matters a good deal more then random score that can vary extremely widely when it is just a "random" number.

The actual chip if binned properly matters more and this can be seen from average clocks with the specific model in question more then the random numbers.

 

[Nenu]

High ASIC score = High voltage leakage, Less voltage needed for Clock
Low ASIC score = Low voltage leakage, More voltage needed for Clock

A card with either score can still be a good Overclocker, or a bad Overclocker.
The score is just a guideline to adjust the voltage appropriately.

Other way round paritally.
High ASIC = LOWER voltage leak, less voltage needed.

Voltage leaking is current leaking to ground. This raises the silicons temp and reduces the current capacity from the voltage supply to where it is needed, requiring higher voltage to compensate.
This is why less voltage is needed with lower leakage.

 

[dragonstongue]

low leak generally do not have the high speeds that high leak can hit, but as you pointed out nenu it also means low leak need a higher volt to compensate to get those higher clocks, but there is that upper limit of course where voltage given will not allow speeds to scale higher for various reasons, this is where said high leak comes in, higher temps but also higher speed capable and because it is leaking the voltage it needs more average voltage to hit above spec speed and will get quite hot in doing so. Really depends on the design in question however, more modern chips are smaller so they do not seem to scale extremely high in speed nor require lots of power but they also get extremely hot when attempting to do so though they of course have higher specific performance in question at those clocks. Older chips are the best way to see the differences really as the required speed to clocks to temps could be a very large discernible difference.

The overclocking champion chips generally are very high leakage, part of the reason they need wicked voltages and in turn need extreme cooling but also hit extreme speeds. Low leakage chips such as used in cellphones or mobile chips simply do not hit the same sheer clocks but also get much lower average power use as well as temps.

 

[Armenius]

The more you know. For once a higher number isn't better. I was under the impression that it was opposite. Guess that helps explain why my GTX 780 @ 73% OC's pretty damn well. Probably could take it higher but I haven't really bothered as 1188Mhz is good enough for me.

It would probably be better if they scaled ASIC quality on an integer basis, like acidity/alkalinity. When most people see a percentage they will automatically assume that higher=better.

 

[dragonstongue]

but who makes the numbers :O
would make be easier to understand though as in a -/_ type bar graph or something.
Above +x need less power but will not hit the same speeds but will also average lower temps relative to other chips in this design architecture.
Below -x% more power needed though results in higher speeds as well as increased temps relative to other chips in this design architecture.

Show the number on the scale say its 50 on the scale it will be an average card/chip in regards to possible speeds vs required power and given temps.

 

[Unknown-One]

but who makes the numbers :O

That would be Nvidia

They test each GPU and determine how much leakage is present. Chips with leakage issues are fed more voltage in order to assure stability.

That's all ASIC quality is a measure of. High ASIC chips will tend to run cooler and quieter (due to lower default voltage), but will perform identically to low-ASIC chips.

Show the number on the scale say its 50 on the scale it will be an average card/chip in regards to possible speeds vs required power and given temps.

I've never actually seen a chip with ASIC quality that low.

I assume it follows the A, B, C, D, F grading scale, where anything below 60% is an insta-fail that would send the chip back for re-binning with additional SMX units disabled.

 

[dragonstongue]

Question in this would be, why would Nvidia or AMD bother supplying this info for consumer use, for makers I can understand as in "batch A is all high grade low leakage that as such are a higher base price" or some such thing, the makers themselves sell us the card as a "premium" models such as MSI Hawk, lightning and so forth that usually use binned chips which are more then likely purchased as such.

More then likely the chip is "paged" for certain figures such as base VID and so forth and this forms the "asic quality" Although it is neat for consumers to compare various cards and reverse calculate the numbers for various reasons, it simply has very little bearing on things, the only way you know this number is if you use gpu-z for this purpose, beyond that, its just a gpu that can clock low or high use a little or a lot of voltage and produce little or a lot of heat, the grading is not really done for our benefit as consumers, it is done for the makers so they can upsell and HOPEFULLY use the best chips in their better offerings

 

[HOODedDutchman]

I have checked it on multiple cards and high 90%+ cards overclock like crazy. Actually owned 5 different 670s and actually scaled to asic. 2 ftw sig2 that were 89% and 91%. The 89% did high 1200s and the 91% did just over 1300. Had an amp that was 98% and did just under 1400. Then had 2 wf3 that were 72 and 78 I believe. 72 would even do 1200 and the 78 did about 1202.

 

[LordEC911]

Question in this would be, why would Nvidia or AMD bother supplying this info for consumer use, for makers I can understand as in "batch A is all high grade low leakage that as such are a higher base price" or some such thing, the makers themselves sell us the card as a "premium" models such as MSI Hawk, lightning and so forth that usually use binned chips which are more then likely purchased as such.

More then likely the chip is "paged" for certain figures such as base VID and so forth and this forms the "asic quality" Although it is neat for consumers to compare various cards and reverse calculate the numbers for various reasons, it simply has very little bearing on things, the only way you know this number is if you use gpu-z for this purpose, beyond that, its just a gpu that can clock low or high use a little or a lot of voltage and produce little or a lot of heat, the grading is not really done for our benefit as consumers, it is done for the makers so they can upsell and HOPEFULLY use the best chips in their better offerings

AMD does not...

AMD supports W1zzard with specific registers for things relevant to the performance specifications of the GPU. An ASIC score would not be relevant and would also show too much insight into their binning process and potentially yields.

Wavey never specifically mentioned ASIC score relevance in your quote...

Nothing like completely misrepresenting what Dave was trying to say.
If you are going to quote him at least spell his name correctly...


http://forum.beyond3d.com/showpost.php?p=1808073&postcount=2130

ASIC quality is completely misleading since it is assigning a percentage to a specific register file without understanding what it is.

Any sort of "ASIC quality" also gets thrown out the window with features like GPU Boost 2.0 and PowerTune 2.0 bringing about new binning techniques.

It is possible that there is some relevance there for Nvidia GPUs but that would be somewhat unlikely.

 

[dragonstongue]

It is meaningless numbers, Nvidia supplies makers with chips as does AMD, they do internal testing and possibly sell high grade chips for premium products at a higher price but lets go by averages, the average card out of a specific line will hit roughly the same average speed with the same average power with the same average temperatures, some like the FTW cards are "binned" much tighter supposedly and will generally clock higher as will Asus TOP based cards and so forth.

If it was a very specific grading say you got a FTW card that had a grade A+ chip one could expect extremely low voltage or temps and/or possibly much higher then average overclock ability, but am quite sure Nvidia nor AMD and especially not Intel would be telling consumers that it could very much castrate sales, they would rather just sell the damn card and if your maker (asus or whatever) happens to sell you a golden card at a premium price that shows well what they are trying to do awesome, if not well its the luck of the draw as always.

The quoting by lordECE911 is very good in this. "ASIC quality is completely misleading since it is assigning a percentage to a specific register file without understanding what it is"

Without knowing where or how these numbers are derived, they might as well say all chips are grade a+ or 95% and not all chips are the same this would be just as true and just as out of context. 99% or whatever to us might be a far far different meaning to Nvidia such as "every chip used in the GTX280 is a 99% cause if it is not, we don't use it for one. there is and always will be variances this is why we bin them and designate them as a specific model"

 

[Unknown-One]

I have checked it on multiple cards and high 90%+ cards overclock like crazy. Actually owned 5 different 670s and actually scaled to asic. 2 ftw sig2 that were 89% and 91%. The 89% did high 1200s and the 91% did just over 1300. Had an amp that was 98% and did just under 1400. Then had 2 wf3 that were 72 and 78 I believe. 72 would even do 1200 and the 78 did about 1202.

It's still not a fool-proof indicator of overclockability, not by any means.

My GTX 780 reports an ASIC quality of 80.2%, which isn't fantastic... yet, it can maintain a stable 1300MHz overclock, with the stock cooler, and a core temperature of 85c. Totally insane overclock for such sub-optimal conditions.

I run 1200 MHz for day-to-day, and have swapped-up to an ACX cooler (temperatures at 1200 MHz are much more manageable). I'm really curious how far it would go if it were water cooled, though...


Edit: And then you see threads like this where people are hoping and praying that they'll get a card with ludicrously low ASIC quality, because they believe it'll overclock better: http://forums.evga.com/tm.aspx?m=2121101

 

[HOODedDutchman]

Kingpin is a card meant for ln2 tho. So low asic is much better. I don't think anyone should be hoping for it unless they are on chilled water or ln2. On average high asic is much better for air.

 

[xcamarox]

This

 

[limitedaccess]

I'm not sure why Nvidia and AMD wanted to open up this issue though by revealing ASIC numbers.

ASIC does effect Nvidia cards at stock. Assuming you are not limited by factors such as temperature, ASIC affects what value the card will boost to. The difference though is not huge, each boost bin is 13mhz for Nvidia, so you would need a 4 bin variance to account for a ~5% difference.

As for ASIC and overclocking, it is one factor. This mean's a higher or lower ASIC is not the only factor for OCing but just variable. ASIC value by itself does not mean whether a chip is better or worse.

 

[QwaarJet]

can't mean much if Gigabyte puts this in their ghz edition card which are supposedly cherry picked, right??? right??

 

[dragonstongue]

not very cherry picked then huh, course the other side to this, should overclock like a bat out of hell provided cooling is fantastic.

Also asic quality refers to the chip, and to me, reference design. So if a maker decides on using extreme cooling, or way upping(or crapping out on) Vreg design, chokes and so forth it will end up meaning very little, it is a very specific number for a vastly variable thing in the real world, probably meant just for a "my card is l33t because" type thing.

 

[seiyafan]

My Evga GTX 770 ACX is 85.8%. I can overclock it to 1293 Mhz, RAM at 8000 Mhz (2000Mhz)

 

[dragonstongue]

so to come back to this thread after a very long while cause was curious looking at my power use with my good olde 7870 (the one I kept of the 2 I owned) the quality is now 84.1% instead of the old number way back when of 82.9%....it seems you WANT low quality if you have the proper cooling because they will clock like crazy, BUT you want high quality i.e low leakage for "normal" cooling, operating temperatures and the like.

Suppose it is something along the line of, if you want to chase crazy speeds, you NEED that leakage so produces more waste heat which needs to be cooled in a timely fashion (heat is the enemy) but I suppose if the transistor is "looser" i.e leaky, you can keep cooling in check allowing the much higher than average clock rates...the "tight" or not leaky one will be less power to "drive" it up to a certain upper threshold the design hits that "edge"

(kind of like modern Ryzen they are not built for wicked clock speeds their upper limit seems to be at most from what I can tell ~4.3Ghz no matter the sheer amount of cooling that it is given (air/liquid/exotic) a "decent one" will hit the 4.0-4.14Ghz (instead of maybe maybe able to "touch" 4Ghz (barely) not need a bunch of raw volts to "power it" where the ones that can hit higher say 4.2-~4.3Ghz need more raw voltage, the cooling factor is not quite as prevelant but the ability for the volts to allow ths slightly higher than normal clock limit seems to be part of it, so, normal is "higher asic quality" or they would not make the cut at all, and the other ones that maybe run a bit warmer or whatever they seem need/want/able to use the extra volts to get that extra 100-250Mhz "redline" (not sure on the normal lower voltage limit i.e "standing voltage" no load like we seem to be able to measure the GPU that we "downvolt" to normalize clocks/reduce idle temps type thing.

--------------------------------
-------------
pure throwing numbers at a wall example, lets say average Radeon 7870 to do 1300 on the core (average higher end overclock ability seems ~1400) needs 1.31v, the "low leakage" one may only require 1.05v instead to hit/maintain those clocks and be much cooler in process and maybe needing only 1.37v for that 1400 but it almost reached the limit of speed towards redline, where the "high leak" one needs instead a starting voltage of 1.4v jumps to 1.43v for 1400 but given the proper cooling can top out say 1480-1520 with 1.52v,(numbers are way off to lunch am just using as an example) so more power to drive, more cooling required, higher ceiling achievable, whereas the lower leak one can hit the end of its redline sooner without needing quite as much raw power nor extra cooling potential, and can maintain its default clocks even if you reduce the from factory voltage.
-------------------------------------------------
-------------------------
-----------
I know years ago the micron D9 ram was similar, higher leakage, could up the volts a fair amount and it would clock WAY higher then comparable "standard" memory, BUT, you had to make bloody sure you kept it quite cool so it was not a for sure suicide run.

I suppose the long and short of it is (to my current understanding and at least a bit more knowledge) with great power comes great responsibility..

bad analogy...
liquid fueled SR71, crazy leaky design BUT because of it being leaky the way that it is, it also becomes very much able to do wicked high speed and chews a boat load of fuel and produces a an amazing amount of heat that has to be countered or a very serious risk it just melts in the process.

compared to a standard "jet engine" which is by far less leaky and by far less mach 4 level speed demon, still is hit running but far far cooler as result as well though, in the first example that big leaky fast moving jet will never be a lean cool running machine, it basically cant whereas example 2 that not so leaky beast will never be able to hit the same sheer speeds it is not leaky enough to "enable it"

so SR71, high leakage "low asic quality"
standard "jet engine" low leakage "high asic quality"

I think my good old 7870 for all the reviews I have seen of them and various asic quality results, am kind of the very happy medium with an 84%, can get a really nice low voltage and maintain standard clocks or put the volts almost as low as can and still load some graphic intensive things (150/300 core mem is .826v, I can do .852-.856 for 525/600, standard clock when 3d load is 1050/1200 at 1.219v, I can do 1175/1382 1.176v or if drop a bit to say 967/1325 @ 1.062-1.132