Fermi based Tesla will be available in Q2 2010.

[Idontcare]

Well there's "TDP specs so we can technically claim our product conforms to PCIe specs, etc" and then there is "TDP specs because shit melts if it gets any hotter than this".

What you see there is the 5970 engineered to elevate the overhead to the latter, but it will no doubt be "stock clocked" such that if conforms to the thermal budget alotted by the former.

Same way all those CPU overclocking folks manage to overclock their 140W TDP processors to clockspeeds that require the power-consumption to then be in excess of 300W.

Of course you aren't going to do it with your $15 bargain budget 250W sparkle PSU on clearance at Frys either, the whole power-distripution setup has to be able to handle the loads.

 

[SlowSpyder]

Charlie has a story now stating that Nvidia has 'massively' missed clock speed targets with Fermi. He states that earlier Fermi DP performance numbers were 768GF of DP performance, now Nvidia is stating 520-630GF of DP performance.

That 768GF of DP performance number does stick out as something I read earlier, and AMD used it in one of their slides. So, now that number is 18% lower. So, if that 768GF of DP performance was infact given by Nvidia at an earlier point, then I guess something did change, and at this stage in Fermi's planning/pre-production work I'd guess clock speed would be the most likely change.

But I didn't see much when I searched for where that 768GF of DP performance number came from. Like I said, that number does stick out to me as something I had seen somewhere, but I don't know if it was more than AMD's slide. Tempered81 has a pic of the slide in earlier post in this thread. I know I've seen that before, but I *think* I've seen the 768 number referenced other places, but maybe not... really can't say for sure.

Charlie's story: http://www.semiaccurate.com/2009/11/16/fermi-massively-misses-clock-targets/

 

[faxon]

Well there's "TDP specs so we can technically claim our product conforms to PCIe specs, etc" and then there is "TDP specs because shit melts if it gets any hotter than this".

What you see there is the 5970 engineered to elevate the overhead to the latter, but it will no doubt be "stock clocked" such that if conforms to the thermal budget alotted by the former.

Same way all those CPU overclocking folks manage to overclock their 140W TDP processors to clockspeeds that require the power-consumption to then be in excess of 300W.

Of course you aren't going to do it with your $15 bargain budget 250W sparkle PSU on clearance at Frys either, the whole power-distripution setup has to be able to handle the loads.

frys doesnt sell sparkle

 

[Idontcare]

Charlie has a story now stating that Nvidia has 'massively' missed clock speed targets with Fermi. He states that earlier Fermi DP performance numbers were 768GF of DP performance, now Nvidia is stating 520-630GF of DP performance.

That 768GF of DP performance number does stick out as something I read earlier, and AMD used it in one of their slides. So, now that number is 18% lower. So, if that 768GF of DP performance was infact given by Nvidia at an earlier point, then I guess something did change, and at this stage in Fermi's planning/pre-production work I'd guess clock speed would be the most likely change.

But I didn't see much when I searched for where that 768GF of DP performance number came from. Like I said, that number does stick out to me as something I had seen somewhere, but I don't know if it was more than AMD's slide. Tempered81 has a pic of the slide in earlier post in this thread. I know I've seen that before, but I *think* I've seen the 768 number referenced other places, but maybe not... really can't say for sure.

Charlie's story: http://www.semiaccurate.com/2009/11/16/fermi-massively-misses-clock-targets/

This is another one of those stories where perfectly rational business decisions are completely conflated with needless speculation.

Before, Nvidia had no real idea where Cypress would debut at. They had their estimates but until AMD actually released the 5870 Nvidia would not know what the clockspeeds would be.

Now that they know what Cypress' DP specs are they have a more realistic "budget". Speed costs money by way of lowering parametric yields. Why toss out all the otherwise perfectly functioning chips that won't clock high enough to hit 768Gflops DP if you can sell a lower-clocked SKU that will still beat the competition?

Somehow this standard business practice went from "binning based on competition so you make more money" to "NV sucks so badly they have no choice but to make lower bins! the idiots!".

Look at what Intel did with Kentsfield versus Phenom...was Intel making those low clock speedbins because Kentsfield horribly missed clockspeed targets? Charlie's story is just as needlessly ludicrous.

This is straightforward business management, and once again its the kind that every company in the industry practices but only Charlie would find a way to convolute the story to make it seem like maximizing gross margins by binning to your competitions offerings is somehow indicative of a fatal flaw in the design and engineering of the IC itself. lulz

Had Cypress debuted at clockspeeds that enabled 750Gflops of DP you can bet that Nvidia would have kept to binning Fermi SKU's to high enough clockspeeds that they hit their original upper-limit estimate of 768GFlops of DP. But now they don't have to, so why throw away sellable silicon?

 

[MrK6]

This is another one of those stories where perfectly rational business decisions are completely conflated with needless speculation.

Before, Nvidia had no real idea where Cypress would debut at. They had their estimates but until AMD actually released the 5870 Nvidia would not know what the clockspeeds would be.

Now that they know what Cypress' DP specs are they have a more realistic "budget". Speed costs money by way of lowering parametric yields. Why toss out all the otherwise perfectly functioning chips that won't clock high enough to hit 768Gflops DP if you can sell a lower-clocked SKU that will still beat the competition?

Somehow this standard business practice went from "binning based on competition so you make more money" to "NV sucks so badly they have no choice but to make lower bins! the idiots!".

Look at what Intel did with Kentsfield versus Phenom...was Intel making those low clock speedbins because Kentsfield horribly missed clockspeed targets? Charlie's story is just as needlessly ludicrous.

This is straightforward business management, and once again its the kind that every company in the industry practices but only Charlie would find a way to convolute the story to make it seem like maximizing gross margins by binning to your competitions offerings is somehow indicative of a fatal flaw in the design and engineering of the IC itself. lulz

Had Cypress debuted at clockspeeds that enabled 750Gflops of DP you can bet that Nvidia would have kept to binning Fermi SKU's to high enough clockspeeds that they hit their original upper-limit estimate of 768GFlops of DP. But now they don't have to, so why throw away sellable silicon?

Except that doesn't seem like it's the case here. Despite the moronic way he writes, Charlie brings up a good point: you don't produce a part that is merely comparable to the competition six months down the road and not hang your head in shame. While reducing clocks to increase yields is a great business strategy, it only works when you're already on top and can actually sell said parts. You can bet NVIDIA would love to come out with guns blazing and for Fermi to be a monster card. However, all current reports/speculation/FUD/crying points to the fact that it won't even be worth looking at. Therefore, I'd lean more towards the fact that the chip isn't coming out nearly as well as NVIDIA would have liked, and they therefore running through revision after revision to find a feasible solution.

 

[Janooo]

This is another one of those stories where perfectly rational business decisions are completely conflated with needless speculation.

Before, Nvidia had no real idea where Cypress would debut at. They had their estimates but until AMD actually released the 5870 Nvidia would not know what the clockspeeds would be.

Now that they know what Cypress' DP specs are they have a more realistic "budget". Speed costs money by way of lowering parametric yields. Why toss out all the otherwise perfectly functioning chips that won't clock high enough to hit 768Gflops DP if you can sell a lower-clocked SKU that will still beat the competition?

Somehow this standard business practice went from "binning based on competition so you make more money" to "NV sucks so badly they have no choice but to make lower bins! the idiots!".

Look at what Intel did with Kentsfield versus Phenom...was Intel making those low clock speedbins because Kentsfield horribly missed clockspeed targets? Charlie's story is just as needlessly ludicrous.

This is straightforward business management, and once again its the kind that every company in the industry practices but only Charlie would find a way to convolute the story to make it seem like maximizing gross margins by binning to your competitions offerings is somehow indicative of a fatal flaw in the design and engineering of the IC itself. lulz

Had Cypress debuted at clockspeeds that enabled 750Gflops of DP you can bet that Nvidia would have kept to binning Fermi SKU's to high enough clockspeeds that they hit their original upper-limit estimate of 768GFlops of DP. But now they don't have to, so why throw away sellable silicon?

Isn't Fermi supposed to go against Hemlock (~1Tflops in DP)?

 

[Genx87]

Except that doesn't seem like it's the case here. Despite the moronic way he writes, Charlie brings up a good point: you don't produce a part that is merely comparable to the competition six months down the road and not hang your head in shame. While reducing clocks to increase yields is a great business strategy, it only works when you're already on top and can actually sell said parts. You can bet NVIDIA would love to come out with guns blazing and for Fermi to be a monster card. However, all current reports/speculation/FUD/crying points to the fact that it won't even be worth looking at. Therefore, I'd lean more towards the fact that the chip isn't coming out nearly as well as NVIDIA would have liked, and they therefore running through revision after revision to find a feasible solution.

Why? Because you have some idea that because it shows up a few months later it has to be larger % faster than otherwise would be if they showed up on the same day? Lets remember right now AMD has supply issues with their chip. And that most likely wont be fixed until Fermi shows up anyways.

I'd guess when Cypress came out Nvidia breathed a sigh of relief because it allowed them to bin lower on a process that is currently having problems. Which should allow them greater yields.

 

[MrK6]

Why? Because you have some idea that because it shows up a few months later it has to be larger % faster than otherwise would be if they showed up on the same day?

Is this a serious question or sarcasm? There's no smilies so I can't tell. If that's a serious question, may (a) god have mercy on your soul.

Lets remember right now AMD has supply issues with their chip. And that most likely wont be fixed until Fermi shows up anyways.

So because they have supply issues, they can't sell any chips at all? Also, if TSMC is having trouble simply banging out AMD's finalized design, how much time do you think they're going to waste monkeying around with NVIDIA's R & D?

I'd guess when Cypress came out Nvidia breathed a sigh of relief because it allowed them to bin lower on a process that is currently having problems. Which should allow them greater yields.

Right, it must be a great relief to have no finalized product and to get manhandled for six months by the competition. I'm sure that was all in their plan. Do yourself a favor and NEVER operate your own business.

 

[SlowSpyder]

Why? Because you have some idea that because it shows up a few months later it has to be larger % faster than otherwise would be if they showed up on the same day? Lets remember right now AMD has supply issues with their chip. And that most likely wont be fixed until Fermi shows up anyways.

I'd guess when Cypress came out Nvidia breathed a sigh of relief because it allowed them to bin lower on a process that is currently having problems. Which should allow them greater yields.

IDC explained it to me very well, but Mrk6 brought up a good point... but I think Mrk6's point depends on when Fermi hits the market, which is something we just don't yet. Right now estimates are Q110, with many estimates I've seen stating March-ish launch and some estimates as late as May before there is any real retail availability. I have a feeling by March AMD will have their supply issues worked out, probably before than, at least I hope so for their sake. But if Fermi slips, or even really isn't available until May, is ~14% better DP performance a big win if your part launches some 8 months after your competition? I guess a lot depends on how well Fermi games, how big the GPGPU market is by then, how much each companies part costs, etc. Though it makes sense from a business stand point (as IDC explained), I can't help but think that if Fermi launches that much later than AMD's part and is only marginally faster then it is somewhat disappointing. I guess we'll have to wait and see how it all plays out.

 

[Genx87]

Is this a serious question or sarcasm? There's no smilies so I can't tell. If that's a serious question, may (a) god have mercy on your soul.

Try answering it without being an ass.

So because they have supply issues, they can't sell any chips at all? Also, if TSMC is having trouble simply banging out AMD's finalized design, how much time do you think they're going to waste monkeying around with NVIDIA's R & D?

Who said anything about not selling at all?!?!?!?!? I dont even know what your second question means.

Right, it must be a great relief to have no finalized product and to get manhandled for six months by the competition. I'm sure that was all in their plan. Do yourself a favor and NEVER operate your own business.

Nobody is getting manhandled right now because there is a major supply issue. That was my previous point. Nvidia could be relieved about Cyrpress performance because they can have a lower bin product that will have better yields on a troubled process.

And what is with the insults? Cant carry on a conversation without being a child?

 

[Genx87]

IDC explained it to me very well, but Mrk6 brought up a good point... but I think Mrk6's point depends on when Fermi hits the market, which is something we just don't yet. Right now estimates are Q110, with many estimates I've seen stating March-ish launch and some estimates as late as May before there is any real retail availability. I have a feeling by March AMD will have their supply issues worked out, probably before than, at least I hope so for their sake. But if Fermi slips, or even really isn't available until May, is ~14% better DP performance a big win if your part launches some 8 months after your competition? I guess a lot depends on how well Fermi games, how big the GPGPU market is by then, how much each companies part costs, etc. Though it makes sense from a business stand point (as IDC explained), I can't help but think that if Fermi launches that much later than AMD's part and is only marginally faster then it is somewhat disappointing. I guess we'll have to wait and see how it all plays out.

I would honestly be surprised if Fermi doesnt show up until May in supply. I could see a debut in Jan with tight supply through March though. If it shows up in May then, yeah, I think that would be a bit disappointing. Because ATI should be about ready over the Summer for a respin or new arch. But I am ASSuming Fermi shows up in the late Jan\Feb timeframe. Which imo isnt that bad.

I guess the performance situation is all in the eye of the beholder. I expect Fermi in games to be about GTX 295 performance levels. Others seem to expect it to outpace that by a greater margin.

 

[MrK6]

IDC explained it to me very well, but Mrk6 brought up a good point... but I think Mrk6's point depends on when Fermi hits the market, which is something we just don't yet. Right now estimates are Q110, with many estimates I've seen stating March-ish launch and some estimates as late as May before there is any real retail availability. I have a feeling by March AMD will have their supply issues worked out, probably before than, at least I hope so for their sake. But if Fermi slips, or even really isn't available until May, is ~14% better DP performance a big win if your part launches some 8 months after your competition? I guess a lot depends on how well Fermi games, how big the GPGPU market is by then, how much each companies part costs, etc. Though it makes sense from a business stand point (as IDC explained), I can't help but think that if Fermi launches that much later than AMD's part and is only marginally faster then it is somewhat disappointing. I guess we'll have to wait and see how it all plays out.

It's true, we will have to wait and see. The TSMC supply issues are somewhat stagnating the entire market, which has a synergistic relation to the stagnating gaming market (where's our next Crysis?), all which gives way to a larger time frame to play around with than we usually expect. However, I'm still saying NVIDIA really dropped the ball on this one. I mean, GPGPU is Fermi's big deal, right? But it only has 14% faster DP performance? The market is slow, but not that slow.

 

[Idontcare]

However, I'm still saying NVIDIA really dropped the ball on this one. I mean, GPGPU is Fermi's big deal, right? But it only has 14% faster DP performance? The market is slow, but not that slow.

The Tesla C2050 and C2070 products will retail for $2,499 and $3,999 and the Tesla S2050 and S2070 will retail for $12,995 and $18,995. Products will be available in Q2 2010.

http://www.nvidia.com/object/io_1258360868914.html

At those pricepoints it would appear their customers are more than willing to deal with the lacklustre timeline and the meager performance improvements relative to the competition's offerings in both categories.

If that is "dropping the ball" to you then I can only imagine what you think the pricepoints should have been were the performance specs even higher or the timeline even tighter.

 

[MrK6]

At those pricepoints it would appear their customers are more than willing to deal with the lacklustre timeline and the meager performance improvements relative to the competition's offerings in both categories.

If that is "dropping the ball" to you then I can only imagine what you think the pricepoints should have been were the performance specs even higher or the timeline even tighter.

Really? Did you forget to include the figures for pre-ordered units as well as a list of companies ready and willing to support Tesla? Or are you trying to make a point using projected MSRP's on a product that won't be released for another four months? Don't be ridiculous.

 

[Idontcare]

Not at all, just stating the facts.

Someone within Nvidia appears to consider a 6 month too late and $2.5k pricetag product as being just fine for a product that you state represents Nvidia "dropping the ball".

Just makes me wonder what the MSRP ought to have been in your mind if the performance were all the higher or had the timeline for release been all the sooner as you seem to feel is necessary for Nvidia to have avoided dropping any balls.

Would an MSRP of $4k been warranted then? What exactly is the pricepoint penalty you see Nvidia incurring for their too slow and too late product in your expert opinion?
Do you see the current MSRP as being over-stated by an order of magnitude and Nvidia will be trying to give away their slow-as-molasses Tesla's for $250 or less when and if they ever show up?

 

[GodisanAtheist]

Keep in mind the new Tesla cards have a number of server/HPC specific components that add greatly to the price tag and have NOTHING to do with its actual performance (ECC memory, for example). These things were added SPECIFICALLY at the request of NVIDIA's customers and did not exist in prior Tesla models.

 

[Janooo]

What makes Tesla so special that it's going to be that much expensive?
DDR3 ECC memory? Better build PCB?

 

[MrK6]

Try answering it without being an ass.

Then think before you post and ask a decent question. Do you seriously think they will be able to come to the party as late as they are with a part that is only marginally faster and make the same impact as if they had released at the same time as their competitor's launch (i.e. 5800 series in Sept.)? Technology moves quickly, even in this slowed market, and companies need to be on the cutting edge if they want to make a splash.

Who said anything about not selling at all?!?!?!?!? I dont even know what your second question means.

Then what did your comment allude to? I'm saying that AMD is still going to sell a lot of parts in between now and when Fermi is finally released. My questions asked if TSMC is the weakest link in all of this, AMD is only sending in orders to be produced, while NVIDIA (I imagine) must monopolize a lot of time trying to tweak and re-run it's revisions. Basically, it seems it's much simpler for TSMC to meet AMD's demands than NVIDIA's (this might be too simple of a conclusion though).

Nobody is getting manhandled right now because there is a major supply issue. That was my previous point. Nvidia could be relieved about Cyrpress performance because they can have a lower bin product that will have better yields on a troubled process.

And there's still ~3-4 months before Fermi's projected release, right? Continuing from above, if the supply issue hurts/slows AMD's 5xxx series sales, it also hurts/slows NVIDIA's Fermi development. How are they relieved by Cypress's performance? That they can lower their projected specs and achieve higher yields? Sure, I can agree with that, but it doesn't change the fact that even at these lower expectations they still seem to be screwed.

And what is with the insults? Cant carry on a conversation without being a child?

There's no insults there; I'm simply asking you to think before you post and present arguments/opinions in full. If you're that sensitive to text on a screen, I will try to be "warmer."

Not at all, just stating the facts.

Someone within Nvidia appears to consider a 6 month too late and $2.5k pricetag product as being just fine for a product that you state represents Nvidia "dropping the ball".

Just makes me wonder what the MSRP ought to have been in your mind if the performance were all the higher or had the timeline for release been all the sooner as you seem to feel is necessary for Nvidia to have avoided dropping any balls.

Would an MSRP of $4k been warranted then? What exactly is the pricepoint penalty you see Nvidia incurring for their too slow and too late product in your expert opinion?
Do you see the current MSRP as being over-stated by an order of magnitude and Nvidia will be trying to give away their slow-as-molasses Tesla's for $250 or less when and if they ever show up?

Straw man arguments are a stupid way to present yourself as a partially competent participant in this debate.

Moving on, my point is that you quoted MSRPs as a justification that Fermi will be a raging success, which is a ridiculous assumption. How many times has NVIDIA overpriced itself in the past? They "dropped the ball" by planning to release a product six months later that, by all reports, is only marginally faster than the competition. Could they still be successful? Sure, the market could be down enough that this is still adequate. All I'm saying is that compared to their previous performance, this is laughable.

 

[v8envy]

You could also view the higher than normal MSRPs as an indicator of poor performance. Their market analysts took a gander at the pricing vs demand curves based on what engineering told them and decided they can move X units at any price (P) regardless of performance. You can always count on those people all the way to the right of the curve. Another maximum for X * (P - production cost) may not even exist, or may be impractical for other reasons.

Vertical markets are not something you can analyze pricing for and draw any valid conclusion. It's all about relationships and perceived value. Not price vs performance. That is only applicable to mass market commodity items with little differentiation.

 

[scooterlibby]

T
There's no insults there

Straw man arguments are a stupid way to present yourself as a partially competent participant in this debate.

Seems like insults to me, can't we all just get along?!?

 

[SlowSpyder]

Mrk6, the discussion was going pretty nicely, even with some differing opinions. When you disagree with someone's opinion, no need to suggest their arguement are stupid or use sarcastic comments.

I'll let IDC speak for himself, but what I get is that he's saying that it very well is likely to still be a successful part from a business stand point. Would they be able to charge more if they got the 768GF of DP performance? Not likely, it's still much faster than the previous generations of Tesla parts, and now they get to harvest more usable GPU's to sell, saving money.

On the flip side, given prior estimates, will people spending that much money on a part be willing to spend the money seeing as it's not as fast as once reported it would be? Will the part being late compared to their competitors part by a good amount of time possibly push those users and companies to look to their competitor? I don't think so seeing as Nvidia has a lot of software that works on their GPGPU and not AMD's Stream, but you never know.

 

[Idontcare]

Straw man arguments are a stupid way to present yourself as a partially competent participant in this debate.

Moving on, my point is that you quoted MSRPs as a justification that Fermi will be a raging success, which is a ridiculous assumption. How many times has NVIDIA overpriced itself in the past? They "dropped the ball" by planning to release a product six months later that, by all reports, is only marginally faster than the competition. Could they still be successful? Sure, the market could be down enough that this is still adequate. All I'm saying is that compared to their previous performance, this is laughable.

I understand that I am no match for your caliber of cerebral capacity, but if I may trouble you further with a request to assist my diminished competency in regards to the part in bold...where did I say that in any of my posts?

If you are (falsely) attributing that statement as being authored by myself than I can only refer you here for further edification.

And you still haven't answered my question...you stated Nvidia dropped the ball, Nvidia has stated their MSRP's for a Tesla product that will be 6 months behind cypress and only 14% higher peak performance.

So what in your opinion is Nvidia's penalty for having dropped the ball? If there is no penalty then ascribing their actions as having dropped the ball is without relevance.

You are the one who is intent on establishing your opinion that Nvidia "dropped the ball" so I am assuming it is relevant to you to also establish the merits of this assertion. If you have no desire to establish the merits of your opinions then why take the time to express them in the first place?

Granted I am barely a partially competent participant in this debate, so if you have time and desire to respond then please don't be afraid to use really small words to increase the odds of my comprehending them.

 

[Martimus]

Moving on, my point is that you quoted MSRPs as a justification that Fermi will be a raging success, which is a ridiculous assumption.

IDC never said that the product would be a raging success. In fact after reading this thread, I don't think he believes it will considering his comments on Llano and other APU's (maybe that was in a different thread). I'm not sure where you got that idea from.

That said, I do agree that the MSRP of the products that IDC has brought up isn't much of a counter that they are late to the party. The MSRP 4-6 months before the product is even released really doesn't mean a whole lot. It is just as likely to be an early estimate of what such a product might sell for, as it is to be a truly exhausted study of what the market will truly bare at the time of launch.

Yes, it is unlikely to even get into a business before having an idea of the profit behind the activity, but things change so quickly in the time frame between when they expected to have the product released, and its actual release date. ATI may have GPGPU units shipping by then, which would take away the advantage that nVidia counted on when they developed those figures.

 

[Idontcare]

IDC never said that the product would be a raging success. In fact after reading this thread, I don't think he believes it will considering his comments on Llano and other APU's (maybe that was in a different thread).

Yep, AMD definitely has a huge opportunity here to take the legs right out from under Tesla and what I will call the discrete GPGPU if their APU's with x86 ISA support for compatibility gain even the slimmest of traction IMO.

That said, I do agree that the MSRP of the products that IDC has brought up isn't much of a counter that they are late to the party.

And I hope you aren't thinking I brought up the MSRP as a counter to anyone's assertions it is late to the party.

Project management of IC's is so complicated that first of all I wouldn't even want to waste my time attempting to rationalize any opinion I might have regarding whether or not Fermi is "late". Project management involves the maximization and minimization of so many competing priorities and risk criterion that I personally just find it preposterous to even presume that I know enough of the decision loops on the specific project to pass judgement regarding when it was planned to debut versus any business-related reasons why it might have been intentionally targeted at a later release date than what our expectations drive us to want to believe. I've managed enough projects to know what I'd need to know at a bare minimum before making any feeble attempts at doing this, and that info is rarely in the public domain so why commit to an exercise in futility and second-guessing?

That said, my point in bringing up Nvidia's stated MSRPs for Tesla is simply that given that MrK6 felt the performance and timeline merits Fermi being considered as being a "dropped ball" I am curious what he thinks the financial incentives were for Nvidia to not have dropped the ball.

Presumably MSRPs do incorporate some degree of "pricing in" the penalty of being late and slow, it might not be fully priced in but regardless what did Nvidia stand to gain from a Tesla MSRP standpoint (and from there, gross margins and profits) had the performance been higher by any arbitrary amount and had it been released to the market sooner by any arbitrary amount.

 

[SolMiester]

They "dropped the ball" by planning to release a product six months later that, by all reports, is only marginally faster than the competition.

I'm keeping this for a sig when nV release Fermi.....LOL, how anyone can make such sweeping assessments on an as yet un-released product with no concrete spec or performance numbers is laughable.....

AFAIK, nV has yet to release any product after ATi, that hasn't had a better performance %...