Fermi needs A3 silicon? Launch in Q1 2010

[x3sphere]

As long as the Fermi launch isn't as terrible as the HD5xxx launch it's a win for NV
All these delays aren't hurting them as much as they could, given the TSMC problems.
If yield issues had been solved and AMD could flood the market, NV would be in a bit more trouble, but as it is, it looks like NV might have a chance of ramping up when TSMC claim they will have sorted their problems by.

Even if NV wanted to, I doubt they could ramp up before Q1.

Well, remember that majority of cards are sold with prebuilt computers. I think the yield issues aren't hurting ATI as bad as some may think. Places like Dell are getting priority on parts, making it harder for retail outlets to cope with demand.

 

[Genx87]

If fermi is indeed a large, complex pwnage chip that NV says it is, then this is going to be very bad. They're competing for space at TSMC, who is also having trouble with 40nm, and if Fermi is complex as I stated we're looking at low yields on top of it all.

The best prediction is to count Nvidia out until 2011, maybe longer because just because Fermi is looking like a misstep, doesn't mean NV will have great luck with it's successor either.

Lets just wait until we actually see it before declaring Nvidia out for the count shall we? The last two new arch's from Nvidia did very well.

 

[jvroig]

I agree, it's a little premature to count anybody out at this point.

Given how much of a monster card Fermi looks on paper, I almost take it as a given that, when it materializes, it will kick the butts of the 58xx cards.. But I'm not expecting more than 50% performance, certainly not 100%. I don't remember that happening before, but there was a mention of the 8800GTX. I need to be refreshed on that matter, was the 8800GTX really 100% faster than the 3xxx flagship card? (It could very well be, the 3xxx weren't exactly the best, but I just can't seem to remember 100% faster).

I'm probably one of the minority excited about Fermi for its advertised GPGPU prowess I'm a software guy, and GPGPU in particular is important to me. If it is as powerful as is currently hyped for GPGPU, then it's a winner. What's expensive for a gamer card is actually pretty cheap for a company wanting to utilize it as a GPGPU-powered machine, so Fermi can be a big win here for people and companies that are into it.

 

[Tempered81]

Was twice as fast as ATIs counterpart?

Not that close really. 8800gtx was 86% faster than a 7900gtx, and about 65% faster than a 1950xtx

 

[Genx87]

Not that close really. 8800gtx was 86% faster than a 7900gtx, and about 65% faster than a 1950xtx

The 2900 series was the counterpart to the 8800. And Nvidia has since the 7000 series card targetted SLI of the previous generation.

I'd expect Fermi to be about as fast as the 295 GTX in games.

 

[jvroig]

As fast as the GTX295? But in that case, that's not going to be 50% faster than the 5870 at all: http://www.anandtech.com/video/showdoc.aspx?i=3643&p=17

Or am I missing something? (Could be, I'm no GPU expert)

 

[coreyb]

Why would their new card equal the performance of the 295? that makes no sense.

 

[jvroig]

Presumably, that's because the new card (single configuration) should be equal to the previous generation's fastest card in SLI, and the 295 is actually SLI.

I'm not sure though if NV actually will target merely the 295. I actually have no idea what their target is, so anybody here in the know should set us all straight,

 

[nOOky]

The truth is no one really knows how fast the new cards will be, there is only speculation. It's november, and if you are in need of a new card and an nvidia fanboy you'll be waiting almost 4 months.

 

[jvroig]

Agreed. If you must absolutely wait for nVidia's Fermi, then that's still a few months away.

One could argue that getting an AMD/ATi card right now is no easy task either. That's true, but then again, when Fermi comes out, there's also no guarantee (no matter how unlikely it may be) that 40nm at TSMC will be good enough to produce enough yields for both lines of products.

 

[palladium]

Presumably, that's because the new card (single configuration) should be equal to the previous generation's fastest card in SLI, and the 295 is actually SLI.

I'm not sure though if NV actually will target merely the 295. I actually have no idea what their target is, so anybody here in the know should set us all straight,

hmm.. if the leaked specs are true ( 512 shaders) and assuming GTX380 has the same clocks as GTX 295, I'd expect it to be quite a bit faster than the 295, since 295 only has 480 shaders, plus the 380 would have significant arch improvements, and no SLI overhead.

 

[jvroig]

Thanks, palladium. I'm not a graphics card expert by any stretch of the imagination, but that does sound reasonable. The "no SLI overhead" and more shaders are probably the biggest factors. I am not sure about "arch improvements", since it seems Fermi's arch is more leaning towards GPGPU? Like I said, I'm no GPU expert. I could be terribly misinformed or could have terribly misunderstood some preview/analysis I read.

 

[Genx87]

As fast as the GTX295? But in that case, that's not going to be 50% faster than the 5870 at all: http://www.anandtech.com/video/showdoc.aspx?i=3643&p=17

Or am I missing something? (Could be, I'm no GPU expert)

I'd honestly be surprised if in games that Fermi is 50% faster than the 5870.

 

[jvroig]

Some people (you and me included, so it seems) expect Fermi to have only a modest performance advantage over 5xxx, while others seem to think it's going to be a lot more. I must have been confused and initially thought you were one of those who expect a whopping 50% or more advantage over 5870, so my reply to you, in hindsight, seems to be barking at the wrong tree. Sorry, my mistake

 

[SlowSpyder]

Why would their new card equal the performance of the 295? that makes no sense.

That would be my guess, that Fermi would be somewhere around GTX295 performance, give or take 15% or so depending on the game, settings, and resolution. But that nothing more than speculation. And we're all taking a guesses with having no idea what kind of clock speeds Nvidia will be able to pull off... will this large complex chip run at 575MHz, or will Nvidia be able to work some magic with 40nm and creat a 750MHz part? Another big question mark is MIMD... will that be a huge boost to gaming, or will that pretty much not matter for gaming at all and be a boost to GPGPU? Until we see some benches we can do little more than speculate.

Obviously multiGPU has it's pitfalls yet, but I think it's pretty much a sure thing that Fermi will be launched against existing 58x0 parts that should have (hopefully) very good availablity by then, and probably an established 59x0 part as well. Not to mention if Fermi doesn't launch unti March, it'll be a good 5-6 months after the 58x0 launch, it would be possible that AMD would have a tweaked 58x0 part out by then, something similar to what the 4890 was to the 4870. I think Nvidia is going to have to work hard to get it's entire Fermi line up out the door as quickly as possible.

 

[exar333]

with all this drama, hype, and anticipation of Fermi... if NV doesn't outright bust down the doors of the AMD 5870 series... I mean if it only bests the 5870 by 1.1x-1.2x times... then it's semi-failure. Gotta beat the 5870 by at least 1.5x, but preferably by 1.8x-2.0x, in order to get some respect here.

I definitely disagree. If the top Fermi beats the 5870 by 25% or more, that would be a big victory. The 5870 wasn't even 2x the speed of the 4780.

I see this being a rehash of the last couple years; even if Fermi bests the 5870 by 20-30%, AMD will likely have the x2 version released not too long afterwards. I really doubt a x2 Fermi will be around at launch...

 

[evolucion8]

To powerup an Dual Fermi will require a nuclear reactor hehe, but I found strange the fact that the HD 5870 is twice the hardware of the HD 4870 and yet doesn't offer the same performance as the HD 4870X2 which is also twice the power of a regular HD 4870, probably some bottleneck in the hardware or driver maturity.

 

[MrK6]

Fermi needs to be at least 20% faster than the 5870 across the board if it is going to be released in Jan./Feb. Generally speaking though, I don't think it will be. Current designs in cards seem to be following a law of diminishing returns at lower resolutions. While the 5870 was more than twice as fast as the 4870 @ 2560x1600, this wasn't the case at 1680x1050. I'd be interested to see how Fermi scales at lower levels, but I think it will follow the same trend as other cards. In either case, the entire market is a mess due to their dependence on TSMC. Hopefully this (these) incident(s) are a real eye-opener for companies to explore alternative solutions.

 

[Fox5]

Was twice as fast as ATIs counterpart?

In some situations I think it was.
A 50% lead wasn't atypical though. R600 may have been the bigger chip too. And R670 didn't improve the situation much for ATI.
HD2900 was from 0% to 50% faster than the x1950xt. The 8800GTX was 25% to 100% faster than the HD2900. X1950XT was around the speed of a 7900GTX to 50% faster.


Given how much bigger Fermi is than R800, nvidia needs to widen their lead from last gen or they have no chance.

 

[v8envy]

Fermi's design happened long before the 5 series saw the light of day. We only have previous history to guestimate from. And looking at history it was highly likely targeted to offer 5-10% less than the performance of the previous generation's highest end single card (285) in SLI.

So, 295 +10% is a reasonable expectation to have. 5850 +50% is not.

 

[palladium]

Fermi's design happened long before the 5 series saw the light of day. We only have previous history to guestimate from. And looking at history it was highly likely targeted to offer 5-10% less than the performance of the previous generation's highest end single card (285) in SLI.

So, 295 +10% is a reasonable expectation to have. 5850 +50% is not.

Agree, although if it launches at a price similar to GTX 280 (and its only 10% faster than a 295) then it'd be a great dissapointment.

 

[v8envy]

I'm sure it'd sell well enough as a halo effect card. 20% faster than a 5870 is perfect for those who insist on the best, and $600 is not unreasonable pricing for such an exclusive beast. 7800GTX 512s flew off the shelves at $750 and weren't that much better than X1800XTs of the day.

Now, if they have to re-spin again and are looking at a late Q2 product availability and competing with leaked specs of 32nm 5 series cards... That would be disappointing.

 

[palladium]

I'm sure it'd sell well enough as a halo effect card. 20% faster than a 5870 is perfect for those who insist on the best, and $600 is not unreasonable pricing for such an exclusive beast. 7800GTX 512s flew off the shelves at $750 and weren't that much better than X1800XTs of the day.

Now, if they have to re-spin again and are looking at a late Q2 product availability and competing with leaked specs of 32nm 5 series cards... That would be disappointing.

Mm, not aware that 32nm 5xxx specs have been leaked. What sort of bump in clock speed can we expect..?

 

[SlowSpyder]

http://www.theinquirer.net/inquirer/news/1561640/nvidia-fermi-delayed

I don't think that was posted here yet, TheInquirer says they don't expect any retail availability before May of next year. Ouch.

 

[Idontcare]

http://www.theinquirer.net/inquirer/news/1561640/nvidia-fermi-delayed

I don't think that was posted here yet, TheInquirer says they don't expect any retail availability before May of next year. Ouch.

Massive "ouch" if those rumors of Nvidia ordering and staging 9000 wafers with A1 FEOL had any truth to them.

The likelihood of those staged wafers still being usable for A3 steppings is quite diminished as the chances of two steppings (A2 & A3) not involving any reworking of the FEOL are quite low.

9,000 wafers x $6,000/wafer != a small number