Of SKUs, OCs and what constitutes a midrange part

[DuckThor Evil]

How is it a "fact" that it is an OC?

Well 7970 launches first then 680 launches and is faster, then Ghz edition comes in with higher voltages and core frequency and is now basically too hot or at least very close to it for your reference cooler design, which didn't get any sort of update to handle the increased power consumption. Sounds quite a bit like OC to me. It's a pity that there doesn't seem to be any good blower designs for the Ghz edition. HIS has a pretty good 7950 design, but their 7970 is different, even though that blower should be capable enough.

 

[Albuquerque]

You seem to only understand what the sticker says and are unwilling or unable to comprehend any factors behind the issue.

The ponderings about "what might have been if..." really do not matter. It has nothing to do with "what the sticker says", it has to do with no other higher-capability product was manufactured by NVIDIA in that market space, period. The sticker could have said whatever it wanted, although it would have been pointless to put a "This is a midrange part!" sticker on it if you really want to think along those lines...

Thus, any complaints, statements, pointed fingers or misdirection under the auspices of "oh but it was just a midrange part" are completely misleading and fallacious. Regardless of manufacturer or product, the highest performing part in your production stable is your high end part, regardless of how you got there.

edit: The response to Mintmaster was now that Titan is already out, but also includes my thoughts about the origin of the 680. Obviously it is their high end single GPU card if there is nothing else on the market

No argument there; Titan replaced the 680 as the "high end part". Done and done.

 

[Dave Baumann]

Well 7970 launches first then 680 launches and is faster, then Ghz edition comes in with higher voltages and core frequency and is now basically too hot or at least very close to it for your reference cooler design, which didn't get any sort of update to handle the increased power consumption. Sounds quite a bit like OC to me. It's a pity that there doesn't seem to be any good blower designs for the Ghz edition. HIS has a pretty good 7950 design, but their 7970 is different, even though that blower should be capable enough.

Actually the power is designed to better utilize the design characteristics of previous XT specification, the cooler didn't need changing, but the fact was it was designed to be a drop in for the design the partner had already transitioned to different coolers

But the simple fact is that an OC is something that the partners take and attempt to run it beyond the reference spec and if they ship it out at a higher clock when they hit a higher yield. The closest thing to that from a IHV perspective would be a speed step on a CPU (which isn't even considered an OC). However, in this case not only did the we redo the entire test programs for the ASIC's we went as far as re-profiling the ASIC because we knew that even the relatively short period between when we initially profiled to the XT2 period the process had improved a fair amount. From the ASIC perspective this is more of a "refresh" than it is an "OC", with the exception the fact that we continued to sell the XT definition under it (and all PRO's sold are actually PRO2 definitions, the partners can just choose to run as PRO and PRO2 configurations).

As for the voltage, you can't make a direct comparison because the leakage to voltage steps changed between XT and XT2. I believe you'd find that the XT2's run at the same or lower voltage for the default clock of 1000MHz (which is higher than the XT default) with the higher voltage being utilized for the boost state.

 

[Silent_Buddha]

Yes because that made the most marketing sense given the competitive landscape, still it doesn't change the characteristics of the GK104 chip itself, that's all they needed and now they can sell their actual high end card for $1000. Titan is in now way a more special card than what a 3GB 580 was two years ago, much less so actually if you compare it to something like Evga 580 Classified.

Your logic has holes.

Actually the reality is that their marketing HAD to make it the high end card, they had no choice as they were unable to produce a faster chip than GK104 at the time.

It took them almost a year to be able to get GK110 out in the consumer market to fill a very low volume niche consumer price point at 1000 USD.

The GTX 680 was their high end as they were not capable of launching anything bigger or faster at the time to compete with 7970. It doesn't matter if they originally planned it as their midrange because they were incapable of launching anything bigger or faster in that timeframe. Hence, it is their defacto high end card/chip of the time.

Unless, of course, we go with my original supposition that GK100 was cancelled due to the inability to get good performance compared to GK104. But most people don't agree with me on that.

Basically if they were able to produce something significantly faster than GK104 (15-20% faster) then they would have. And they likely would have charged 650-750 USD for it as it would have been that much faster than the 7970. They didn't because they couldn't. It's not a marketing thing, it's just the reality of the situation they were in.

Regards,
SB

 

[Cookie Monster]

Putting the whole GK100 thing aside (we don't even know if it exists or GK100 is GK110), I think you guys are arguing about different points and actually agree with one another.

Based on nVIDIA's history and past events, the GK104 can be clearly seen as a GPU designed for their "mid range" part initially in terms of its specification, design focus and what not. All of its attributes point to it being suitable for a mid-range part (just look at the ASUS GTX670 mini!!) I don't think anyone can deny that.

BUT because of what was possible at the time, with probably delays of their actual high-end GPU (GK100/110) and higher than expected performance of the GK104 with higher clocks, its just clear as day that they could get away with the a flagship video card based on the GK104 with additional cooling and higher clocks.

So a GPU that was designed as a performance part in mind (not the flagship) just happened to be marketed and sold as their high end flagship due to reasons outlined by SB such as the delay on their flagship GPU ala GK110 or.. GK100? who knows.

And with the HD7970 being unclocked initially, wasn't that a trade off for being a < 300W part hence the conservative clocks?

 

[Albuquerque]

Basically if they were able to produce something significantly faster than GK104 (15-20% faster) then they would have. And they likely would have charged 650-750 USD for it as it would have been that much faster than the 7970. They didn't because they couldn't. It's not a marketing thing, it's just the reality of the situation they were in.

 

[3dcgi]

And with the HD7970 being unclocked initially, wasn't that a trade off for being a < 300W part hence the conservative clocks?

No.

 

[DuckThor Evil]

Actually the power is designed to better utilize the design characteristics of previous XT specification, the cooler didn't need changing, but the fact was it was designed to be a drop in for the design the partner had already transitioned to different coolers

All the tests I saw during the launch gave me a pretty clear impression that the cooler is struggling with the load. Blower type coolers are good if you want to put more than one card in to a case, or your case is a small form factor one. AMD currently don't have an high end option on offer.

But the simple fact is that an OC is something that the partners take and attempt to run it beyond the reference spec and if they ship it out at a higher clock when they hit a higher yield. The closest thing to that from a IHV perspective would be a speed step on a CPU (which isn't even considered an OC). However, in this case not only did the we redo the entire test programs for the ASIC's we went as far as re-profiling the ASIC because we knew that even the relatively short period between when we initially profiled to the XT2 period the process had improved a fair amount. From the ASIC perspective this is more of a "refresh" than it is an "OC", with the exception the fact that we continued to sell the XT definition under it (and all PRO's sold are actually PRO2 definitions, the partners can just choose to run as PRO and PRO2 configurations).

As for the voltage, you can't make a direct comparison because the leakage to voltage steps changed between XT and XT2. I believe you'd find that the XT2's run at the same or lower voltage for the default clock of 1000MHz (which is higher than the XT default) with the higher voltage being utilized for the boost state.

Perhaps OC is not 100% accurate term, but imo it's close enough. It doesn't have to have a negative association, just that you cranked it up. Are there any regular 7970s that can't reach 1 Ghz without touching the voltages. Seems like they all go well past it anyway.

Actually the reality is that their marketing HAD to make it the high end card, they had no choice as they were unable to produce a faster chip than GK104 at the time.

It took them almost a year to be able to get GK110 out in the consumer market to fill a very low volume niche consumer price point at 1000 USD.

The GTX 680 was their high end as they were not capable of launching anything bigger or faster at the time to compete with 7970.
Basically if they were able to produce something significantly faster than GK104 (15-20% faster) then they would have. And they likely would have charged 650-750 USD for it as it would have been that much faster than the 7970. They didn't because they couldn't. It's not a marketing thing, it's just the reality of the situation they were in.

Regards,
SB

Had 7970 been clearly faster at launch and GK104 a bit slower than what it was, (apparently both of those came as a surprise to nVidia) the marketing couldn't have branded the GK104 as a $500 680. You might very well be right about their capabilities of not being able to bring out GK100 or GK110 early, but it could just as well be that it was a better option for them to take full advantage of being able to manufacture and sell more GK104s at a high price. 2 GK104 chips/cards are going for a $1000, that's a lot more than $650-750 and nVidia get's good margins selling GK110s as Teslas. nVidia would have been ok selling GK104s as a 660-670. More than likely the truth is somewhere in between. Last year the manufacturing volume clearly was stacked against huge chips being a high volume part.

It doesn't matter if they originally planned it as their midrange because they were incapable of launching anything bigger or faster in that timeframe.

I think it's an interesting issue and therefore it matters. Time frames tend to change if forced. Some type of GK100/110 card could most definitely have been launched earlier than Feb/March 2013 into the consumer market if it had been seen as a very high priority. Earlier does not equal March 2012 though.

 

[Dave Baumann]

All the tests I saw during the launch gave me a pretty clear impression that the cooler is struggling with the load. Blower type coolers are good if you want to put more than one card in to a case, or your case is a small form factor one. AMD currently don't have an high end option on offer.

The load is what it was designed to handle, however there was a lower variability on power on the XT2 in comparison to XT and they are all closer to the peak loads of the board design. The blower is still the reference design and some partner chose to go with it (i.e. Club3D were one of the first to market.

Perhaps OC is not 100% accurate term, but imo it's close enough. It doesn't have to have a negative association, just that you cranked it up. Are there any regular 7970s that can't reach 1 Ghz without touching the voltages. Seems like they all go well past it anyway.

No, it is not an accurate term at all. Like I said, in doing this we went through the same processes as an initial characterisation, which is not an "OC" process and not even the process of a speed step. Yes, there were ASIC's that failed to make 1000Mhz (especially at lower voltage) because that's what initial profiling had said, except we had quickly moved in to an area where the process had improved and it was clear they were better than that initial characterisation hence we went through the whole process again with fresh wafers in order to understand where those new limits where.

 

[UniversalTruth]

Again, this has nothing to do with anything. There was nothing else they made that was higher than the 680 for an entire year and then some. It was, defacto, their high end card

Hehe, even the "mighty" titanIC is miserable in some applications. Everything depends on the view point

http://www.tomshardware.com/reviews/geforce-gtx-titan-opencl-cuda-workstation,3474-21.html

 

[Albuquerque]

Hehe, even the "mighty" titanIC is miserable in some applications. Everything depends on the view point

http://www.tomshardware.com/reviews/geforce-gtx-titan-opencl-cuda-workstation,3474-21.html

I expect that we would still call the Titan NVIDIA's "high end card", correct? No different than the 680 supplanting the 580, which supplanted the 480. Looks to be faster than the 680 across the board, even if only "mildly" so...

 

[UniversalTruth]

I expect that we would still call the Titan NVIDIA's "high end card", correct?

No, no, no, of course not. That is why I used the word "miserable". Can a miserable product be called real high-end even if the company has nothing better? No, for sure not. Just take the 7970 and you see what a real high-end is

 

[boxleitnerb]

Simply because AMD started out with an underclocked Tahiti chip and Nvidia took advantage of it by overclocking their GK104.

The 7970 GHz edition was 3 months later than the 680, 21% bigger die size and 15% faster. But that's a "win" for Nvidia right?

Same as the 8 month later Titan at 55% bigger and 30% faster than the 7970 GHz was a "win" for Nvidia as well. The double standards is just staggering.

No what you claimed was that the 680 was not high end, while suggesting that the 7970 non-GHz edition was AMD's best effort and the GHz edition was simply an overclocked version.

Stop spinning words, will you?
Compared to the Fermi and Tesla generations, Nvidia gained on AMD in perf/mm2 and perf/W. Fact. I never said anything about "winning".

 

[Albuquerque]

No, no, no, of course not. That is why I used the word "miserable". Can a miserable product be called real high-end even if the company has nothing better? No, for sure not. Just take the 7970 and you see what a real high-end is

I'm guessing you're trolling for something here, because otherwise this statement really doesn't make sense.

If somehow you're angling to make me sound like an AMD fan, you should do a bit of searching for my previous posts. I had a very difficult time deciding between the 680 and the 7970, and have already stated that this will likely be my last AMD/ATI card for a while. Although, I'm also not in the market for anything as pricey as a Titan.

 

[UniversalTruth]

Perhaps I was a lil bit kidding. But seriously tho...

Once again- the simple fact that a company doesn't have anything higher, doesn't mean that its highest product is high end compared to what another company offers. That is your false statement and assumption. A fact.

 

[DuckThor Evil]

I had a very difficult time deciding between the 680 and the 7970, and have already stated that this will likely be my last AMD/ATI card for a while.

Are you running one or two cards? What sort of issues have you had if you don't mind me asking. I'm currently kind of contemplating on going AMD for a change... The gaming bundle is pretty sweet as I haven't played any of those games yet.

 

[jimbo75]

Perhaps I was a lil bit kidding. But seriously tho...

Once again- the simple fact that a company doesn't have anything higher, doesn't mean that its highest product is high end compared to what another company offers. That is your false statement and assumption. A fact.

If AMD's ~215mm2 chip is their midrange what does that make Nvidia's ~215mm2 chip?

If you say "midrange", then it only leaves one segment left for the 680.

The truth is almost certainly that Nvidia had a different strategy planned for this generation and took advantage of some poor decision making and questionable execution (on the driver front) by AMD.

They were not able to launch Titan until they launched it, so the 680 was definitely planned as their top-end GPU for 2012 - they just couldn't believe their luck when AMD underperformed so shoddily with Tahiti at the beginning.

 

[UniversalTruth]

If AMD's ~215mm2 chip is their midrange what does that make Nvidia's ~215mm2 chip?

If you say "midrange", then it only leaves one segment left for the 680

Midrange can point to chips with different die sizes- for example from 200 to 300 mm^2 everything can be seen as midrange. Lower midrange, higher mid range, etc.
Sweet spot, anyone?

They were not able to launch Titan until they launched it

I suspect they intentionally delayed it because there was no need to launch it. They were already quite competitive with the lower tier card

 

[silent_guy]

Midrange can point to chips with different die sizes- for example from 200 to 300 mm^2 everything can be seen as midrange.

Mid-range is generally considered to be from 200 to 293mm2. Everything above is high end.

 

[jimbo75]

I suspect they intentionally delayed it because there was no need to launch it. They were already quite competitive with the lower tier card

At 2x to 3x the price the markets aren't even near crossing each other so there is just no reason to hold it back.