NVIDIA Maxwell Speculation Thread

I'm a little surprised that there isn't a GTX 960 out yet. Maybe they haven't cleared out enough stock on 7xx series, but I would have thought they would have wanted one out before the holiday shopping season.

NVIDIA GeForce GTX 960 Allegedly Postponed To Q1 2015 Due To Strong GeForce GTX 980 and GTX 970 Sales

http://wccftech.com/nvidia-geforce-...2015-due-strong-geforce-gtx-980-gtx-970-sales

NVIDIA’s next generation budget-friendly GeForce GTX 960 graphics card may have been postponed to Q1 2015 as reported by GDM.OR.JP. The Japanese based retailers which are highly accurate with dates and specification of upcoming products have revealed that due to strong sales for the GeForce GTX 980 and GeForce GTX 970, both cards with remain the flagship GeForce 900 series products for this year while the more budget friendly and sub-$300 US cards will hit the market next year.
 
Possible rumours ....

Said shipping data has already confirmed the taping out of GM200 INT0 and GM 200 CS samples. Meaning both the prototypes and the customer samples have already taped out.

GM200-Customer-Sample-TITAN-II-635x222.png



If we look at the time it took for GM204’s customer sample to tape out and appear on shelves, we will see that this timeframe has already passed for the GM200. Which leads us to two very obvious conclusions: 1) either the GM200 GPU is landing very soon, and by soon I mean within this month or the next, or 2) Nvidia is deliberately holding the GM200 chip back until AMD comes up with any competition or 970/980 sales dry up.

Some of you might remember that Nvidia has already accidentally leaked out the theoretical and expected raw performance efficiency of the GM200 chip which is 25 GFlops per Watt. While they have given the performance efficiency of the GK110 chip (12 GFlops) , they did not give the efficiency of the GM204 chip. No worries though, we can approximate it ourselves. The GM200 is approximately 208.33% faster than the GK110 based on these numbers. Now this is where it gets tricky. The GM204 is locked at 1/32 DP so we cant really measure efficiency (it requires DP numbers). We can however estimate the relative performance per watt by measuring pure single precision GFlops of the GK110 and the GM204.

The GK110 is roughly 550mm^2 and the limit of TSMC is at roughly 600mm^2. So can the GM200 exist on a 28nm Node? Absolutely, yes. Will it? Well, the consumer samples taped out a long time back, and they are sure as hell not on 16nm FinFET. Recent reports seemed to indicate that TSMC has canceled plans for 20nm and while AMD can theoretically revert to Global Foundries for 20nm GPUs, Nvidia cannot. But those reports could very easily be inaccurate. It is also interesting to note that our calculation coincides easily with the latest rumor on Big Daddy Maxwell we received a while back. Needless to say that these GPUs could be ported to 16nm FinFET once its ready, since we have already heard rumblings about the same. Well, there you go folks, everything of interest that happened at the earnings call.


Read more: http://wccftech.com/article/generation-graphics-prospects-nvidia-big-daddy-maxwell-16ff-ports/#ixzz3IteWLnac
 
Last edited:
Next week, at SuperComputing 2014, would be a very appropriate time to announce (and perhaps ship) a GM200 based Tesla M20.


Well, as far as I managed to find out, there won't be any Maxwell-based Tesla soon (meaning: not in the next 6 months).

Instead, there will be a launch this month for a dual chip Kepler based Tesla SKU. As strange as it sounds..
 
It could be the Tesla K80, which is rumored to be a dual-chip part.

If this image is real then the K80 (GK210-DUO) would have ~520 MHz core clock and ~5.0 Gbps memory, assuming all SMXs and memory interfaces are enabled.

NVIDIA-Tesla-K80-GK210-DUO-635x356.jpg
 
I've been wondering about GPU-Z's 'ASIC quality' on the 9xx chips. There's still a perception in the casual overclocking market that a higher ASIC value tends to indicate a higher overclock. As I understand it, ASIC quality is more or less a proxy for 'default operating voltage', and a higher ASIC value means that the chip's default operating voltage is lower. I gathered from reading Dave Baumann's posts on this forum that for AMD Hawaii at least, a chip with a lower operating voltage is one that has high leakage as it needs to run at the lower voltage in order to stay within the TDP target, and I'm assuming that similar is true for NVIDIA's 9xx.

In one of the threads somebody speculated that perhaps we see comparatively fewer 'high ASIC' (high leakage) chips 'in the wild' because most of them probably fail to run at the lower voltage they require and so are discarded or binned into lower parts. With 9xx being relatively voltage-constrained by NVIDIA (ie you can't increase the core voltage much without bios mods) and most chips generally responding well to voltage, will these high ASIC/leakage chips tend to be better overclockers purely because there is more headroom to play with while staying within NVIDIA's constraints? Obviously there is still individual variance but I wonder if there is a general trend.

I also started wondering about the rarity of the high ASIC/leakage chips - if it is the case that comparatively few of them pass binning (because of the above voltage and TDP constraints), can we assume that as the process matures the foundries will tend to produce less high leakage and more low leakage chips to increase yields? Perhaps you are most likely to get a high ASIC chip if you order a reference card very shortly after launch of a new GPU, as this is the time when the process has not been refined as much, and reference cards are usually the first out of the gate.
 
you mean eg 9xx being released on what is now a very mature 28nm process? i assume there is still some level of refinement for a 'new' chip even on a mature process?
 
I also started wondering about the rarity of the high ASIC/leakage chips - if it is the case that comparatively few of them pass binning (because of the above voltage and TDP constraints), can we assume that as the process matures the foundries will tend to produce less high leakage and more low leakage chips to increase yields?
There is also a possibility that some are being stockpiled for a higher clocked 980 should AMD release a competitor that falls between the current 980 and the GM200 (which I'm sure Nvidia would prefer to sell as a Quadro/Tesla given the prices), a dual GM204 (GTX 990) card, and Quadro's.
Perhaps you are most likely to get a high ASIC chip if you order a reference card very shortly after launch of a new GPU, as this is the time when the process has not been refined as much, and reference cards are usually the first out of the gate.
Has the ASIC quality changed with 0.8.0 ? Full support for GM 204 was lacking in previous builds of GPU-Z. It isn't beyond the realms of possibility that the ASIC quality calculation method prior wasn't completely correct.
 
There is also a possibility that some are being stockpiled for a higher clocked 980 should AMD release a competitor that falls between the current 980 and the GM200 (which I'm sure Nvidia would prefer to sell as a Quadro/Tesla given the prices), a dual GM204 (GTX 990) card, and Quadro's.

Has the ASIC quality changed with 0.8.0 ? Full support for GM 204 was lacking in previous builds of GPU-Z. It isn't beyond the realms of possibility that the ASIC quality calculation method prior wasn't completely correct.

Not as far as i know, reads the same for 0.7.5, 0.7.9 and 0.8.0 for me (the version I have to hand). Pure speculation on my part, prompted by my personal experiences and anecdotes on forums. I just figured that as the foundries refine the manufacture of a given design, they'll be aiming to increase yields by targeting the middle of the range, so we'd see more in say, the 65-85% range as GPU-Z sees it, and less outliers in the 65- and 85+ range.
 
I've been wondering about GPU-Z's 'ASIC quality' on the 9xx chips. There's still a perception in the casual overclocking market that a higher ASIC value tends to indicate a higher overclock. As I understand it, ASIC quality is more or less a proxy for 'default operating voltage', and a higher ASIC value means that the chip's default operating voltage is lower. I gathered from reading Dave Baumann's posts on this forum that for AMD Hawaii at least, a chip with a lower operating voltage is one that has high leakage as it needs to run at the lower voltage in order to stay within the TDP target, and I'm assuming that similar is true for NVIDIA's 9xx.

In one of the threads somebody speculated that perhaps we see comparatively fewer 'high ASIC' (high leakage) chips 'in the wild' because most of them probably fail to run at the lower voltage they require and so are discarded or binned into lower parts. With 9xx being relatively voltage-constrained by NVIDIA (ie you can't increase the core voltage much without bios mods) and most chips generally responding well to voltage, will these high ASIC/leakage chips tend to be better overclockers purely because there is more headroom to play with while staying within NVIDIA's constraints? Obviously there is still individual variance but I wonder if there is a general trend.

I also started wondering about the rarity of the high ASIC/leakage chips - if it is the case that comparatively few of them pass binning (because of the above voltage and TDP constraints), can we assume that as the process matures the foundries will tend to produce less high leakage and more low leakage chips to increase yields? Perhaps you are most likely to get a high ASIC chip if you order a reference card very shortly after launch of a new GPU, as this is the time when the process has not been refined as much, and reference cards are usually the first out of the gate.

I cant speak for the Maxwell 970-980. But in general the rules was (( but it is absolutely not warranty ):

Low Asic = can take high voltage, good for LN2 and subzero due to high leakage.
High Asic quality = Low voltage, dont suited for high level voltage, aircooling... Low leakage.

I suspect for Maxwell, that if you see so much low, average Asic quality is with the low TDP and high turbo speed, they can pass more chips in the wild, hence the high numbers of thoses used chips on non retail, overclocked gpu's.
 
Not sure where exactly to post this since there isn't a thread for GTX970/980, but I'm happy to report that my MSI GTX970 has no coil whine at all. My GTX670 would get pretty whiny on loading screens and under certain loads, but this 970 is dead silent. Also the fan doesn't even spin until the card gets hot!
 
I cant speak for the Maxwell 970-980. But in general the rules was (( but it is absolutely not warranty ):

Low Asic = can take high voltage, good for LN2 and subzero due to high leakage.
High Asic quality = Low voltage, dont suited for high level voltage, aircooling... Low leakage.

I suspect for Maxwell, that if you see so much low, average Asic quality is with the low TDP and high turbo speed, they can pass more chips in the wild, hence the high numbers of thoses used chips on non retail, overclocked gpu's.

I think that's the complete opposite of what Dave Baumann said, but perhaps I misunderstood. I think high leakage means low voltage (and thus high asic quality), because higher leakage means higher current and thus more power consumed at a given voltage. The voltage on a high leakage part has to be lower because otherwise it will breach the target TDP. Perhaps it's not so much that the high leakage parts can run at a lower voltage, it's that they have to. The ones you see in the wild are the ones that were stable at the lower voltage. Does this also mean that a given low leakage chip might run at the same lower voltage too, it just doesn't because it's bin didn't require it?
 
I think that's the complete opposite of what Dave Baumann said, but perhaps I misunderstood. I think high leakage means low voltage (and thus high asic quality), because higher leakage means higher current and thus more power consumed at a given voltage. The voltage on a high leakage part has to be lower because otherwise it will breach the target TDP. Perhaps it's not so much that the high leakage parts can run at a lower voltage, it's that they have to. The ones you see in the wild are the ones that were stable at the lower voltage. Does this also mean that a given low leakage chip might run at the same lower voltage too, it just doesn't because it's bin didn't require it?

Dont compare AMD chips binning ( described by dave ) and the Nvidia one .

This said my experience is based on overclocking chips for Asic, not stock one.... But with turbo and highly overclocked retail gpu's i think it is valid as binning chips for a low TDP chips like Maxwell . ( not for official one, but it is possible to find some Nvidia official gpu's `? )

When we overclock under subzero, what we want is the lower Asic possible, for a reason, we control the leak by the cooling and thoses chips can take any voltage you want to throw at it, they are less willing to provide problem with cold bug .. and they can take any voltage ( for a golden chips ).
 
Last edited:
Dont compare AMD chips binning ( described by dave ) and the Nvidia one .

This said my experience is based on overclocking chips for Asic, not stock one.... But with turbo and highly overclocked retail gpu's i think it is valid as binning chips for a low TDP chips like Maxwell . ( not for official one, but it is possible to find some Nvidia official gpu's `? )

When we overclock under subzero, what we want is the lower Asic possible, for a reason, we control the leak by the cooling and thoses chips can take any voltage you want to throw at it, they are less willing to provide problem with cold bug .. and they can take any voltage ( for a golden chips ).

maybe someone else will chime in, but surely the principles are the same? they're both made on TSMC's 28nm process, the behaviour isn't going to be the complete opposite just because one is AMD and one is NVIDIA. The chip binning as described by Dave B makes sense:
  • the PCB is designed to deliver and handle a certain amount of power, the cooler is also designed to dissipate a certain amount of power
  • a 'high leakage' chip leaks more current and so uses more power at any given voltage
  • it follows that in order to keep the power the within the designed limits of the PCB and cooler, high leakage chips must be run at a lower voltage than low leakage chips
Now it's pretty well established that low ASIC quality means higher voltage at stock (ie where the board manufacturers have designated that a particular chip will run), whereas high ASIC quality means lower voltage designated by manufacturer. This means that low ASIC quality should correspond to low leakage, and high ASIC quality to high leakage. The anecdotal evidence from all the overclockers like yourself is that the behaviour according to ASIC quality is as you describe, all I'm disputing is that the classification of high/low ASIC quality in to low/high leakage is wrong. It must be the opposite, as I've described: this would serve to narrow the TDP range within which the PCB and cooler have to operate.

If, on the other hand, what you say is true then within a given product family, in the wild, we would have a much wider range of operating TDPs: low leakage (high ASIC quality) chips running at low voltage and using little current (and thus power) and high leakage (low ASIC quality) chips at the complete opposite end of the spectrum, using high voltage and high current and exponentially more power. This makes no sense.
 
it follows that in order to keep the power the within the designed limits of the PCB and cooler, high leakage chips must be run at a lower voltage than low leakage chips
Why is it that low leakage chips cannot be run at low voltage, provided they are stable?
 
That's what i'm not sure about, I think they can but perhaps the manufacturers don't bother. I assume they don't test individual chips, they sample a few in a batch and characterize the leakage based on that. Given X leakage, we can run these chips at Y voltage. I don't think they bother testing to see if they'll actually run at a lower voltage.
 
Back
Top