How will ATI respond to Fermi? (poll included)

[happy medium]

I can't believe ANYONE picked nothing. Do you really think AMD is going to do nothing to counter a Fermi release?

What are the chances? Nothing?

The price and perfromance of Fermi would have to be perfectly in line with AMd's prices and performance.

If this happends, I think they are working together to keep prices high. They did do this before.

What if Fermi gtx470 is as fast as a 5870? for 349$?
Remember Nvidia has people that will pay more for there product even if its only 10% faster. AMD knows this.

There is just no way. Nothing!?

Edit: I'll tell you what they are going to do. They are going to cut prices to bleed every last sale and % of market share and get that 5890 out A.S.A.P. to match the gtx 480. Then as allways the gtx 485 will surpass the 5890 by just a tad for 50$ more.

 

[cbn]

New faster memory cost more money but why not use the new memory 5 months ago.

I think ATI decided not to use faster memory because they had no competition. Furthermore, the supply of HD58xxx was limited to the point where prices were increasing regardless of memory bandwidth. Remember HD5850 and HD5870 street price actually increased above MSRP.

With supply improving and competition increasing I would think ATI would be more inclined to consider other options.

 

[MrK6]

The GTX4x0's are barely competitive and by all reports are going to be available in very limited quantities. Meanwhile, the 5870 E6 edition is going to be coming out anyway. What else is AMD going to do besides sit there and laugh?

 

[Sylvanas]

I guess Nvidia will see no performance in the upcomming 5 driver releases?
AMd has had 5 months and at least 5 driver releases. They should be at there cards optimal perfromance wouldn't you think? Mabe they been holding back performance for the Fermi? I'll add that to the rumer pile.
Or are you saying that nvidia has mature drivers for a card with no final clocks up untill ..what 2 weeks ago?

We saw the 5870 has plenty in the tank with the release of the 10.3a Catalyst driver recently that improves pretty much everything across multiple games/engines. It's not uncommon to see releases from both companies that offer performance increases to cards 12+ months old- this demonstrates there won't be any 'ceiling' reached anytime soon.

 

[GodisanAtheist]

I wish 1&4 was an option.

If the performance between the top cards is 10%, then keeping the halo part is well within ATI's grasp (maybe reworking their tessellator for additional performance, as well). A 5890 wouldn't be an outlandish proposition.

All the same, ATI's game plan has been price/performance and the ~200 sweet spot. Introing a new Halo part would allow for ATI to bring down the cost of its 5870 & 5850 to price you just couldn't refuse.

 

[cbn]

I've seen this mentioned a number of times in different threads, but the 8800GTX days were completely different then the current situation for Nvidia.

Before the 8800 launched, Nvidia had a very competitive DX9 part out in the 7900 cards. Nvidia isn't very competitive now, and their GTX2xx cards are all EOL'd.

Back then AMD had no DX10 parts out and had nothing in sight. Right now AMD has their entire lineup of DX11 parts out. Right now Nvidia should be doing anything they can to get a potential customer to hold off on buying an AMD part and wait to see what Fermi has to offer.

Because Nvidia didn't have to worry about AMD's DX10 parts before the 8800 launch, they could be tight lipped to keep the 7900's selling. They didn't have to worry about rumors regarding AMD's DX10 parts being better. With Fermi, many rumor sites are printing a lot of negitive information, high TDP's, high power use, low numbers of launch parts, fairly unimpressive performance compared to AMD"s current parts, etc. They have all the reason in the world to shout from the rooftops how wonderful Fermi is and why you shouldn't buy that AMD card and why you should wait for Fermi to arrive in a few weeks. But they aren't. They're being completely silent... it kind of makes you wonder why.

For the very same reason you mentioned, I think ATI would be keeping any planned refresh cards like "HD5890" a complete secret.

Why tell customers about an upcoming product that is faster and better when its own cards are already selling so well?

 

[Zebo]

ATI isn't going to do anything. Charlie was dead on with every single rumor he posted. I believe his trend will continue and that Nvidia is only planning on producing 10k cards total for the worldwide launch. That means if you are one of the lucky(or stupid) ones to buy this monster you will have a collectors item. The fact is it's hot, underpowered and only 5% faster in contrived benchmarks than something ATI released 6 months ago. Nvidia knows this thing is a pig and I doubt they intend to produce any more than they need to say they actually released it.

Everything that's happened over the last few months (rebranding the 8800/9800/240 as the 340 being the biggest clue) has indicated that Nvidia is going to be absent this round. Let's just be honest, they fubar'd Fermi, the next round they will be competative with ATI (if you go by previous history last time nVidia botched it badly they will wipe the floor with ATI with the refresh part) as they fix the mistakes but right now AMD/ATI has the lead and Fermi isn't going to change that. Given the limited production run ATI has no reason whatsoever to reduce prices unless they just want to hurt Nvidia more. Northern Islands will be out before the end of the year so ATI will have their refresh part very soon anyway, given the performance, price and thermal characteristics of Fermi they don't have much reason to worry for a while.

Remember, Fermi was designed as a 512 SP chip, yields are so bad that the best they can do is 480SP (with 32 bad SP's per chip) and they are right at the limit for power (~300Watts) because the chip is so poor. Rumors out of TSMC is that Fermi yields are single digit, with the size of the chip they are getting around 102chips per 300mm wafer, single digit yields means they are getting less than 10 working Fermi's for every $5000 300mm wafer. That means each fermi is costing them close to $500. Given that the cards are selling for $400 Nvidia is taking at least a $150 loss on every sale. They aren't going to sell very many or it will put them out of business. If the rumor of 10,000 cards total is true then the are going to lose $1,500,000 selling Fermi and that's if the card only costs $50 to produce. That's a lot of money, I'd be surprised personally if you can even purchase a fermi, 10,000 cards worldwide is like 3 per city. It will be a hell of a collectors item though, one of those famous semiconductor failures in the history of computing.

Nice synopsis Rahvin. FX 5800 part deux but worse if your loss numbers are correct/

 

[cbn]

http://www.techfuels.com/latest-hardware-news/27290-hynix-announcement-7-gbps-gddr5.html

It looks like the high density (2 GB) 7 Gbps GDDR5 won't make it into production till second half of 2010.

However, I thought some lower density 7 Gbps GDDR5 was already available?

 

[Zebo]

From the sound of it all AMD has to do is a voltage bump and corresponding frequency bumps in the 5875 to beat these cards.

 

[PingviN]

I'd love to see a price drop, but with GF100 being what it seems, I don't think that's going to happen.

 

[Lonyo]

Brings me back to the "why didn't they show the benchmarks" earlier argument.
If they had good drivers, (as you say) and a faster card, release the benchies and stop AMD from selling so many direct x 11 cards back in November.

So your saying Amd drivers were not mature when they released? They managed to increase speed quite a few percent from Sept. ha?

I guess Nvidia will see no performance in the upcomming 5 driver releases?
AMd has had 5 months and at least 5 driver releases. They should be at there cards optimal perfromance wouldn't you think? Mabe they been holding back performance for the Fermi? I'll add that to the rumer pile.
Or are you saying that nvidia has mature drivers for a card with no final clocks up untill ..what 2 weeks ago?

Why would clock speed impact on driver development?
This card has been in various stages of actual production for 6 months, so they should have had cards for the last 6+ months.
If they have had cards for the last 6 months, and the primary respins have been due to yield issues (i.e. architecturally the design has remained the same), then that's 6 months in which they could and should have been developing the drivers.

If after a 6 month delay in getting the product to market their driver team hasn't been working on the drivers, and the card launches with crappy drivers, that says something about NVs overall process.
There is NO WAY that these cards should ship with crappy unoptimised drivers. Clock speeds not being confirmed shouldn't matter. They presumably had some internal target and some samples which worked at said targets to work with.

That's not to say that NV won't be able to extract more performance, but they shouldn't be able to make any significant leaps in the short term, because they've had a long time to work on these drivers.

 

[1h4x4s3x]

Nothing.
Apart from the already announced Eyefinity 6 and 2GB cards.

 

[SlowSpyder]

Brings me back to the "why didn't they show the benchmarks" earlier argument.
If they had good drivers, (as you say) and a faster card, release the benchies and stop AMD from selling so many direct x 11 cards back in November.

So your saying Amd drivers were not mature when they released? They managed to increase speed quite a few percent from Sept. ha?

I guess Nvidia will see no performance in the upcomming 5 driver releases?
AMd has had 5 months and at least 5 driver releases. They should be at there cards optimal perfromance wouldn't you think? Mabe they been holding back performance for the Fermi? I'll add that to the rumer pile.
Or are you saying that nvidia has mature drivers for a card with no final clocks up untill ..what 2 weeks ago?

I'm saying I doubt Nvidia has shown Fermi up to this point because it's a turd. If they show what it can do they're more likely to push people towards AMD then by being quiet. It's like when AMD was quiet before the 2900XT launch.

AMD had launch drivers when the 5870 came to market. I'm actually very suprised they found this much performance. I have a feeling they were holding back on purpose so that they could improve performance right before the Fermi launch, though I'm sure they'd never admit that. With 10.3 who knows, AMD's parts may actually be faster than Fermi.

I'm not saying Nvidia will have no performance improvements through driver releases, but I think Nvidia's situation is different. Nvidia wanted their driver team to have drivers ready back in November. After delay after delay in the launch they have been able to refine the drivers most likely, before the hardware was ever released. So by now, going on April, their drivers should be pretty good.

 

[Executor_]

If the rumor of 10,000 cards total is true then the are going to lose $1,500,000 selling Fermi and that's if the card only costs $50 to produce. That's a lot of money, I'd be surprised personally if you can even purchase a fermi, 10,000 cards worldwide is like 3 per city. It will be a hell of a collectors item though, one of those famous semiconductor failures in the history of computing.

I thought the 5-10k rumor was only referring to the availability of the GTX 480. Charlie reported a yield of ~2% (1.7?), ~104 candidates per wafer, and 5000 risk wafers, so that seems to make sense. Assuming nVidia managed to salvage another 10-20% of the dies as 448sp models, than that means around 50-100k GTX 470s.

 

[ronnn]

Good question.



Edit: ALso Nvidia has the game makers lined up to optimize thier games on thier hardware and historicly has great drivers.

They do have a history with their drivers. not all good

 

[Skurge]

I can't believe ANYONE picked nothing. Do you really think AMD is going to do nothing to counter a Fermi release?

What are the chances? Nothing?

The price and perfromance of Fermi would have to be perfectly in line with AMd's prices and performance.

If this happends, I think they are working together to keep prices high. They did do this before.

What if Fermi gtx470 is as fast as a 5870? for 349$?
Remember Nvidia has people that will pay more for there product even if its only 10% faster. AMD knows this.

There is just no way. Nothing!?

Edit: I'll tell you what they are going to do. They are going to cut prices to bleed every last sale and % of market share and get that 5890 out A.S.A.P. to match the gtx 480. Then as allways the gtx 485 will surpass the 5890 by just a tad for 50$ more.

I picked nothing, they don't need to do anything if Fermi isnt even available to buy.
http://www.shanebaxtor.com/2010/03/21/april-hd-5800-and-hd-5900-price-drop/

 

[happy medium]

I picked nothing, they don't need to do anything if Fermi isnt even available to buy.
http://www.shanebaxtor.com/2010/03/21/april-hd-5800-and-hd-5900-price-drop/

So you picked nothing but linked to possible price drops for the 58xx series because of Fermi?

I didn't see anything in that article about Fermi not being avalable in April?

GTx480 "massive avalability April 6"

http://www.fudzilla.com/content/view/18110/1/

Who knows?

 

[Skurge]

Is 10,000 units real availability?

 

[ronnn]

First time I read "massive" was here?

 

[Chaotic42]

Rumors out of TSMC is that Fermi yields are single digit, with the size of the chip they are getting around 102chips per 300mm wafer, single digit yields means they are getting less than 10 working Fermi's for every $5000 300mm wafer. That means each fermi is costing them close to $500. Given that the cards are selling for $400 Nvidia is taking at least a $150 loss on every sale. They aren't going to sell very many or it will put them out of business. If the rumor of 10,000 cards total is true then the are going to lose $1,500,000 selling Fermi and that's if the card only costs $50 to produce. That's a lot of money, I'd be surprised personally if you can even purchase a fermi, 10,000 cards worldwide is like 3 per city. It will be a hell of a collectors item though, one of those famous semiconductor failures in the history of computing.

I thought that companies got a break from TSMC when yields were low?

 

[Lonyo]

I picked nothing, they don't need to do anything if Fermi isnt even available to buy.
http://www.shanebaxtor.com/2010/03/21/april-hd-5800-and-hd-5900-price-drop/

I think that link (assuming it's true) says it all.

What’s interesting though is I’m hearing that the price drop will only happen if they feel threatened by NVIDIA and its new GTX 470 and GTX 480 products. There’s a few thing that need to happen with the new GTX 400 series for this price drop to happen.

First the products have to be competitive against the HD 5850, HD 5870 and HD 5970.

Secondly the stock has to be available or at least the impression that mass stock is coming.

And that's why Fermi at $350/$500 with potentially limited availability is a terrible thing.
We can only hope that either it does more than 10% performance improvement, it's MSRP at that with good availability (and retailers sell it for below that) or... we might get stuck with high prices

 

[bunnyfubbles]

The HD5800s might be ready for a price reduction regardless of competitive pressure, the cards will have been out for 6+ months and are now very easy to find in stock with their current inflated price tags.

And by price reduction all ATI needs to do is lower the prices back to the original MSRPs of $259 and $379 for the 5850 and 5870 respectively and the cards would likely sell out.

 

[Martimus]

At what point does it stop making sense for Nvidia to lower prices?

The way I look at things Fermi is primarily a HPC card that can also be sold as a gamer's card.

But why is the die so large is what I want to know? Does having a SLI bridge connecting two smaller die GPUs on a single PCB hamper HPC performance?

I don't understand this logic. nVidia probably sells 1000 gaming cards at least for every 1 HPC card they sell. Unless the HPC cards make 1000 times more profit than gaming cards, it doesn't make sense to rely on that market. They would have to cut a lot of expense if they were to go that route and still be solvent. I don't see them leaving the GPU market, or even neglecting it anytime soon.

 

[Lonyo]

I don't understand this logic. nVidia probably sells 1000 gaming cards at least for every 1 HPC card they sell. Unless the HPC cards make 1000 times more profit than gaming cards, it doesn't make sense to rely on that market. They would have to cut a lot of expense if they were to go that route and still be solvent. I don't see them leaving the GPU market, or even neglecting it anytime soon.

They aren't relying on the HPC per-se, but they are trying to grow it.
If they sell 1000 more PC GPUs than they sell in the HPC market, doesn't it make sense to try and change the balance of that?
Lets assume the profit levels are 10x what they are for GPUs, and it is 1000:1.
If they want to increase profits, then which is easier, going from current HPC to 100x HPC levels to almost double profits, or going from current GPU sales to 2x that?

The easiest way to increase profit for NV is to get the high margin HPC market buying their products in much greater numbers. Since the market is probably going to grow at a much greater rate than the desktop market, and the margins are much higher, it makes sense to invest in it while maintaining a high level of product in the desktop market.

 

[akotlar]

I think the option isn't listed. They'll release an updated product line. I think 5850 will take $299, 5870 will take $399, 58"90" (higher clocks 6DP) will take $499, 5970 $649, and 59"90" (higher clocks, 6DP) will be ~$799.

A price re-adjustment is certainly possible if Fermi outperforms 5850/5870 by a larger than expected margin. If I were in ATI's shoes and my yields were really as good as is stated (with chips costing ~$89), and I could meet my target profit margin at a lower price point I'd do that. If it meant I'd have to cut into my profit margin by a significant chunk to lower prices, I wouldn't because I'd bet on a stronger marketing push first to increase my sales enough to take the wind out of my competition. That being said I expect that a strong 470 & 480, coupled with strong reception would push ATI down to $249 - 279 for 5850, $349-$379 for 5870. If 5870 is faster than 470 and cost $29.99 more, people will buy it, based on ATI's dx11 dominance and strong power consumption profile.

I really don't see the 480 GTX's place in the market @ $449 if the 5870 is $349-379. They'd really have to make a convincing argument for PhysX, which unfortunately right now simply doesn't exist due to the super-limited game lineup and questionable immersion benefits over multi-threaded cpu physics (see Metro 2033).