Sign in with
Sign up | Sign in
Your question

ATI HD 2900 Pro?

Last response: in Graphics & Displays
Share
September 22, 2007 7:13:43 PM

From the sources I have read the upcoming ATI HD 2900 Pro is basically an underclocked XT, which means that it could be an excellent bang for buck card.

My questions
1) Are there any leaked benchmarks?
2) Does anyone know when these cards are due for release because I have noticed some sites taking pre-order?

Thanks

More about : ati 2900 pro

September 22, 2007 7:21:42 PM

cards will probably launch monday or tuesday since his and gecube have them ready for shipping.

there's a lot of talk about 256-bit or 512-bit. seems that HIS and Gecube have 512-bit versions (r600) cards just slower clocked than XT. There also seem to be r670 cards (256-bit) with the same name : HD2900PRO

within a couple of days it will all be clear wheter or not there will be different versions.

if there is a 512-bit version for a price that is that low they can count me in!
September 23, 2007 12:19:20 PM

if all it required was an OC I would be in as well, but we'll have to wait and see
Related resources
a c 169 U Graphics card
September 23, 2007 12:24:16 PM

if it will be something like the difference between X1900XT and X1900XTX , then many will buy it although u have too wait and be sure
September 23, 2007 2:43:57 PM

only time will tell... hope they come out soon.
a c 130 U Graphics card
September 23, 2007 4:18:36 PM

I dont think the overclocking option would be a gimmee even if they are basicaly underclocked XTs as my feeling is that these will surley be the chips that dont quite cut it as XTs same as we had with the X800 series ie failed chips flogged off as lower spec cards some did oc some didnt.
Mactronix
September 23, 2007 9:44:15 PM

The 256bit versions have even lower clocks and 256mb and are called 2900GT's.

Even if you cant overclock em at all its still an awesome card! 512bit interface and 1gb of memory all for £180!!
September 24, 2007 3:26:26 PM

1gb memory doesn't mean anything if its a slow card otherwise though, not that I think it is a slow card
September 24, 2007 3:39:51 PM

There have not been any benchmarks yet. In fact, there has been very little information about this card. I'm still a little confused about which card is which between this and the 2950 pro (if they are both real).
September 24, 2007 4:04:22 PM

actually, there is a benchmark, I found this:
http://pclab.pl/art28806-3.html
in polish, but the numbers are numbers.

The author claims he got an unit for tests
a c 169 U Graphics card
September 24, 2007 4:12:22 PM

wait for more trusted sites like Tomshardware , xbitlabs ,anandtech ,...
September 24, 2007 4:45:34 PM

as a polish speaking person...

tested card is HIS Radeon HD2900Pro

On the first page they state that HD2900Pro chips are R600s that could not work at XTs clocks and they would try to see if the card could be OCed to XT level
2nd and 3rd pages are fairly straightforward - hardware + benchmarks. additionally they mention the price - no more than 950 PLN which is less than 8800 GTS 320MB. regular XT is around 1300 PLN, thus Pro is around 25% cheaper than it's bigger brother
4th... overclocking... They state that they first tried 650 core (from 600 @ stock) - worked fine. than 675, 700... it OC'ed all the way to 750/900 which is above stock speeds of the XT !
5th page is conclusion. many pluses, only two minuses - it's loud and consumes a lot of power (surprised?)

oh, btw - pclab is one of the best polish sites in terms of hardware testing and benchmarking, I think it's a legit review

my conclusion - if I were not moving to Japan in a couple of days for the next year or so - this would be a perfect card for me. pity one cannot buy a laptop with some enthusiast card...
September 24, 2007 5:40:54 PM

wow, so when it comes out, it could be awesome?
September 24, 2007 5:52:22 PM

It will depend heavily on price. Here in the U.S, it needs to be priced below the 320MB 8800GTS. Which means less than $280. If it is, the HD2900Pro will be an excellent deal. The bad part for ATI (but good for us) is that nVidia is in a much stronger position and can afford to lower the prices on the 320MB 8800GTS. ATI cannot. This coupled with the possibility that nVidia may come out with a much more competitive mid-range card than their crap 8600GTS doesn't help ATI much.

It's also likely that these HD 2900Pro won't be around long due to limited stock.
September 24, 2007 6:03:09 PM

well, thats good in the short term, but I don't want AMD dieing personally
a b U Graphics card
September 24, 2007 6:15:14 PM

Anoobis said:
The bad part for ATI (but good for us) is that nVidia is in a much stronger position and can afford to lower the prices on the 320MB 8800GTS. ATI cannot.


First of all how would that be good for us, nVidia's profitability obviously isn't coming because of lower prices? As for ATi/AMD , if these are parts that would otherwise go unused, then selling them at $200 makes more sense than just stockpiling them. Their replacements will be cheaper so it doesn't make sense long term either.

I agree that ATi can't try to make sales by losing money in order to buy customers, but if there parts were going to go unsold anyways, then it's adding to their margins for the overall R600 line.

Quote:
This coupled with the possibility that nVidia may come out with a much more competitive mid-range card than their crap 8600GTS doesn't help ATI much.


But we don't know what nV has to offer, alot will depend on whether it too beats the GTS, because unlike these crippled XT parts, the 8700 will need to finance itself.

Quote:
It's also likely that these HD 2900Pro won't be around long due to limited stock.


I agree with that, I suspect they'll be replaced by a dedicated part eventually, in the same fashion the R9500 and plain GF6800 were in the past. Selling cripples only works until the demand for those cripples outweighs the number of natural cast-offs you have.
September 24, 2007 10:24:24 PM

Quote:
First of all how would that be good for us, nVidia's profitability obviously isn't coming because of lower prices? As for ATi/AMD , if these are parts that would otherwise go unused, then selling them at $200 makes more sense than just stockpiling them. Their replacements will be cheaper so it doesn't make sense long term either.

You're telling me that a lower priced 320MB 8800GTS is a bad thing? I could care less who makes what card. The 320MB 8800GTS is a great card. If this HD2900Pro competes well with it across the board Nvidia could lower their prices on the 320MB 8800GTS to offset the HD2900Pro. A lower priced 320MB 8800GTS is a plus to us.

Quote:
I agree that ATi can't try to make sales by losing money in order to buy customers, but if there parts were going to go unsold anyways, then it's adding to their margins for the overall R600 line.
[/quotemsg]
I'm not disagreeing with you on this. nVidia is just in a better position to deal with a price war than AMD/ATI. It makes good sense for ATI to use up this inventory of lower-bin R600s and use up the inventory. But even if they're are lower-bin, they still costed ATI money to make. Money that ATI doesn't have a whole lot of to throw around. Ditto for nVidia except that they haven't been having as many money issues as AMD/ATI has.

Quote:
But we don't know what nV has to offer, alot will depend on whether it too beats the GTS, because unlike these crippled XT parts, the 8700 will need to finance itself.

Which is why I said "the possibility" of a more competitive mid-range card. The performance of the 8700 (or whatever the call it) remains to be seen. Talk about another sad disappointment if it doesn't cut the muster in regards to the sorry 8600GTS.
a b U Graphics card
September 24, 2007 11:14:18 PM

Anoobis said:

You're telling me that a lower priced 320MB 8800GTS is a bad thing? I could care less who makes what card. The 320MB 8800GTS is a great card. If this HD2900Pro competes well with it across the board Nvidia could lower their prices on the 320MB 8800GTS to offset the HD2900Pro. A lower priced 320MB 8800GTS is a plus to us.


Yeah, but is it as low as it would go if there were more competition, not likely. Before the HD2900 launched it was cheaper than it is now, it went up, how is that better for us? And my comment is that nV is making record profits because they are selling at margins they couldn't get away with when there was tighter competition, so yeah I disagree with your statment that nV's position is good for us, I agree easily that it's bad for AMD, but that was mor about their own product & actions than nV's. It's not good for us IMO because greater competition is better for us, and leads to better prices which we don't currently have. nV may be able to lower prices whihc will THEN be better for us, but their current situation is what your sentence refers to, and really if ATi were in a better position then these prices cuts would've come a long time ago.

Quote:
nVidia is just in a better position to deal with a price war than AMD/ATI.


I don't disagree with that, like intel they are making 'exess profits' (ie larger than equilibrium) and thus they can easily rob from product A to finiance product B, which is pretty much what they both did in the past and gave us cheap middle-range products very soon after launch.

Quote:
It makes good sense for ATI to use up this inventory of lower-bin R600s and use up the inventory. But even if they're are lower-bin, they still costed ATI money to make. Money that ATI doesn't have a whole lot of to throw around.


But you miss the point that this is a sunken cost. It's money already spent and that can't be used to sell as an XT, so another SKU/model had to be created to finally get money for those parts, so it's not that they are losing $50 per chip, it's that they are geting X amount of dollars that they weren't before for those chips that can't be sold as XTs. Only if they sell more than their cast offs and need to make fully functional XTs into PRO/GT boards do they start to actually have a negative impact on margins and profits. Think of it like selling your waste product, sure it's not junk, but it's something you otherwise get nothing for, so unless your 'junk' business becomes so succesful that you have to create that 'junk' to satisfy those customers, then you should be more profitable than before.

Quote:
Ditto for nVidia except that they haven't been having as many money issues as AMD/ATI has.


Yes, but it's a different situation, the GF8700 is not a cast-off, the GTS is. So for the comparison, youu replace the GTS-320 with the GF8700 if you don't have enough cripples for the same price as it costs you to make new chip and it's offspring. So nV's still not going to introduce a GF8700 unless the benefits outweight the costs of dropping the GTS-320's MSRP. And it's not surprising that a fast 65nm part might offer enough benefit to replace the GTS-320, and likely have other spin-off beneifts for a GF8700GS and mobile parts.

Quote:

Which is why I said "the possibility" of a more competitive mid-range card. The performance of the 8700 (or whatever the call it) remains to be seen. Talk about another sad disappointment if it doesn't cut the muster in regards to the sorry 8600GTS.


And that's the thing, at this point in time if the GF8700 doesn't beat the GF8800GTS-320 then you create a weird pressure where if ATi can shoot that gap, then what do you do, put the GF8700 against the HD2900Pro because it's cheaper to make per die (if you don't already have enough crippled G80s naturally), or lower the price of the GTS-320 even more?

And of course that answer depends on the HD2900Pro's performance, as well as the GF8700's performance. If the GF8700 can outperform in DX10 titles and at the 16x10 and under sweetspot of the GTS-320 then it's got 2 compelling reasons to pick it over what it's replacing. But if it falls short, and costs close to as much (say $20-40 from the GTS-320) then you may have the same situation the GF8600GTS encountered where at launch it's fighting against it's bigger brother for value.

No matter what the HD2900Pro needs to have price/performance viability against the GTS-320 under its current inflated (IMO) pricing, the thing is which ends up being the best money policy for nV dropping the price of the GTS-320 or else high-bining a GF8700 SKU (with 'Ultra Extreme' models) which may still win the price/performance crown, but for nV will offer better margins, as well as push more new parts to spread out their cost.

IMO, I suspect the later, and mainly because the longterm attraction of a GF8700 style part offers more benefit than continuing to cripple the G80. They have alot more wiggle room to work with, but just because you're flush wit money doesn't mean you want to waste it if you don't have to.

The two biggest issues IMO are the level of rejects from both the G80 and the R600, if they're very high, then the GTS-320 against HD2900Promakes alot of sense from both sides, but if their rejection rate in that speed range is low (either more good parts or more complete failures) then a cheaper per unit GF8700 and RV6xx part makes much more sense for both. And if ATi/AMD is selling full speed XTs as Pros, then they need to move quickly on a replacement, otherwise they aren't losing money so much as recouping money that would otherwise be lost. The only advantage in that case for a new chip would be to reduce the high power requirements, which really isn't attractive for this price/performance range, even the GTS-320 keeps some away (because while 125W is better than 150W it's still no GF7900GS for power consumption).
a b U Graphics card
September 25, 2007 12:26:47 AM

Anoobis said:
...This coupled with the possibility that nVidia may come out with a much more competitive mid-range card than their crap 8600GTS doesn't help ATI much.

It's also likely that these HD 2900Pro won't be around long due to limited stock.


Just before heading home I checked Fudzilla, and this makes that last statement interesting....

http://www.fudzilla.com/index.php?option=com_content&task=view&id=3217&Itemid=1

http://www.fudzilla.com/index.php?option=com_content&task=view&id=3218&Itemid=1

http://www.fudzilla.com/index.php?option=com_content&task=view&id=3228&Itemid=1

If that stuff from FUAD is to at all be believed and the Q4 shipments are to consumers, not just AIBs, then the HD2900Pro will likely be simply a sell of of bum parts, and quickly replaced by the RV670, which would give AMD just what nV needs, cheap dedicated product to fit in that market segment. That's going on the assumption that neither AMD nor nV can satisfy demand of a $200 GTS-320/Pro with crippled higher end parts alone. And if they are competative, then we should have another price-war regardless of financial health (remember ATi is a profitable segment of AMD).

Half the TDP of the HD2900 also helps fulfil the 'less power consumpition' aspect too.

Could be a very interesting pre-Xmas launch season. :sol: 

Of course these are really simply the cards we should've had for the summer. Like the X1600->X1650 it'd be nice to pretend the GF8600/HD2600 never existed.
a c 130 U Graphics card
September 25, 2007 7:43:39 AM

I think the pro version just comes down to a quick buck for AMD/ATI due to a high level of failures of XT chips(no i dont have any proof just my opinion).I dont think Nvidea will react at all to these PRO cards even if they do go out at a good price for the consumer they can afford to ignore them if the supplies are low but if the market suddenly gets flush with them then they can react with a short term price drop.
Either way if you were in the market for this kind of card at the time they are released it would be good unless as i say the Nvidea cards stay where they are and you are a die hard fan.
Mactronix
September 25, 2007 8:46:39 AM

according to that review it still uses 200W and a loud cooler. I thought it was supposed to be ~150W
September 25, 2007 2:43:48 PM

http://www.tweaktown.com/reviews/1189/his_radeon_hd_290...

Wow...This card is going to be quite a bargain....It's pretty much specc'ed exactly like a 2900XT...just with a lower clock. For this price range this card is going to be VERY popular...VERY..popular..more sense to buy then then the 2900XT imo...lol...I wonder how much they are all gonna be priced at. This should drive down the price of the x1950xt's that are still on the net as well. Great move from AMD...But now we wait to see what nvidia reveals..either a midrange...or new high end..(8800 refresh? 8900 GTX?...or 9800 GTX...or a new mid range to battle the 2900 pro...time will tell..)
a c 169 U Graphics card
September 25, 2007 2:57:55 PM

or 8700GTS ?
a b U Graphics card
September 25, 2007 4:31:15 PM

abigsmurf said:
according to that review it still uses 200W and a loud cooler. I thought it was supposed to be ~150W


How would would a lower clocked card 'still use 200W' when the HD2900XTs don't use 200W?

September 25, 2007 5:07:04 PM

It takes less power..Still more then a 8800 GTS but well within contending range...
September 25, 2007 5:13:58 PM

Quote:
Yeah, but is it as low as it would go if there were more competition, not likely. Before the HD2900 launched it was cheaper than it is now, it went up, how is that better for us? And my comment is that nV is making record profits because they are selling at margins they couldn't get away with when there was tighter competition, so yeah I disagree with your statment that nV's position is good for us, I agree easily that it's bad for AMD, but that was mor about their own product & actions than nV's. It's not good for us IMO because greater competition is better for us, and leads to better prices which we don't currently have. nV may be able to lower prices whihc will THEN be better for us, but their current situation is what your sentence refers to, and really if ATi were in a better position then these prices cuts would've come a long time ago.

I don't think you're hearing me TGGA. These are all ifs but if the HD2900Pro turns out to be a solid contender against the 320MB 8800GTS, ATI will be competitive again. Not only will they be competitive, but they will be more competitive in the mid-range market where they're already strong. The mid-range market which brings in more money than the uber high-end range. Unless nVidia has something new to fire back with that's better and cheaper to make than the 320MB 8800GTS, the only way they can compete with ATI at the moment is to lower prices which is good for us. My argument is that nVidia is in a strong position and can afford to do this and that this HD2900Pro could be a godsend for ATI because it's exactly what they need. A solid performing yet well-priced mid-range DX10 (regardless of how trivial that really is) card. I hope it sells like hot-cakes. If it turns out good I'm going to start wondering how well it stacks up against my X1950XT.

Quote:
But you miss the point that this is a sunken cost. It's money already spent and that can't be used to sell as an XT, so another SKU/model had to be created to finally get money for those parts, so it's not that they are losing $50 per chip, it's that they are geting X amount of dollars that they weren't before for those chips that can't be sold as XTs. Only if they sell more than their cast offs and need to make fully functional XTs into PRO/GT boards do they start to actually have a negative impact on margins and profits. Think of it like selling your waste product, sure it's not junk, but it's something you otherwise get nothing for, so unless your 'junk' business becomes so succesful that you have to create that 'junk' to satisfy those customers, then you should be more profitable than before.

There was no point to miss. I was simply stating pretty much what you just said. They have inventory on hand that they need to get rid of. Inventory that cost them money to make. Inventory that they can now sell with these HD2900Pros.

Quote:
Yes, but it's a different situation, the GF8700 is not a cast-off, the GTS is. So for the comparison, youu replace the GTS-320 with the GF8700 if you don't have enough cripples for the same price as it costs you to make new chip and it's offspring. So nV's still not going to introduce a GF8700 unless the benefits outweight the costs of dropping the GTS-320's MSRP. And it's not surprising that a fast 65nm part might offer enough benefit to replace the GTS-320, and likely have other spin-off beneifts for a GF8700GS and mobile parts.

Honestly man, I don't know where you keep getting this, but I haven't once compared the upcoming 8700GTS (or whatever they call it) to the the 320MB 8800GTS. I never said, nor do I expect the 8700GTS to compete or beat the 320MB 8800GTS or this new HD2900Pro. I hope for it to be better than the dog that is the 8600GTS. If it is, it's just yet another bit of bad news for ATI.

Quote:
And that's the thing, at this point in time if the GF8700 doesn't beat the GF8800GTS-320 then you create a weird pressure where if ATi can shoot that gap, then what do you do, put the GF8700 against the HD2900Pro because it's cheaper to make per die (if you don't already have enough crippled G80s naturally), or lower the price of the GTS-320 even more?

Like my note said above, I don't expect the 8700GTS to compete at all with the 320MB 8800GTS. I expect it to replace the horrible 8600GTS and deal with nVidia's lack of competition against the stellar X1950Pros and X1950XTs. Again, that would be a spot of bad news for ATI.

Quote:
And of course that answer depends on the HD2900Pro's performance, as well as the GF8700's performance. If the GF8700 can outperform in DX10 titles and at the 16x10 and under sweetspot of the GTS-320 then it's got 2 compelling reasons to pick it over what it's replacing. But if it falls short, and costs close to as much (say $20-40 from the GTS-320) then you may have the same situation the GF8600GTS encountered where at launch it's fighting against it's bigger brother for value.

No matter what the HD2900Pro needs to have price/performance viability against the GTS-320 under its current inflated (IMO) pricing, the thing is which ends up being the best money policy for nV dropping the price of the GTS-320 or else high-bining a GF8700 SKU (with 'Ultra Extreme' models) which may still win the price/performance crown, but for nV will offer better margins, as well as push more new parts to spread out their cost.

IMO, I suspect the later, and mainly because the longterm attraction of a GF8700 style part offers more benefit than continuing to cripple the G80. They have alot more wiggle room to work with, but just because you're flush wit money doesn't mean you want to waste it if you don't have to.

The two biggest issues IMO are the level of rejects from both the G80 and the R600, if they're very high, then the GTS-320 against HD2900Promakes alot of sense from both sides, but if their rejection rate in that speed range is low (either more good parts or more complete failures) then a cheaper per unit GF8700 and RV6xx part makes much more sense for both. And if ATi/AMD is selling full speed XTs as Pros, then they need to move quickly on a replacement, otherwise they aren't losing money so much as recouping money that would otherwise be lost. The only advantage in that case for a new chip would be to reduce the high power requirements, which really isn't attractive for this price/performance range, even the GTS-320 keeps some away (because while 125W is better than 150W it's still no GF7900GS for power consumption).

Aside from your comparison of the 8700GTS against the HD2900Pro/320MB 8800GTS, I agree with that.
September 25, 2007 5:31:32 PM

TheGreatGrapeApe said:
How would would a lower clocked card 'still use 200W' when the HD2900XTs don't use 200W?


The review shows total system power... not GPU power. I think he may have been confused.
a c 130 U Graphics card
September 25, 2007 5:44:30 PM

Just a query but wouldnt they (ATI )have already allowed for a margin of waste which would already be written off against in the budget before a die was cast.
And also wouldnt they have known and therefore already accounted for the fact that they could have fails to resell as a different spec of the range.

Or do you think that having a fail is a main pain financially as i would have thought that the business model of the process wouldnt allow for more than about 3%.
If the margins are this tight then the Pro would represent loss reduction rather than free cards as it were.

But on the other hand if its all accounted for fails and all then ATI are essentially printing free money with the Pro...


Note: accidentally edited Mactronix's post. - TGGA

** Sorry man, I hit edit instead of reply and didn't notice what I did until it went up.
I forgot what you had as the last line it was a question. you might want to fix. **
September 25, 2007 6:48:22 PM

^^Makes sense to me. As with every series, disabled high end cards fill the mid-range. ATi would have accounted for this in their original plans. I highly doubt this card is selling at a loss anyway.
a b U Graphics card
September 25, 2007 7:18:14 PM

SORRY for the mix-up was trying to post from work. Anywhoo, hopefully kept the integrity of what you had (can't go back and undo unfrotunately). Please re-edit it if anything major is missing.

mactronix said:
Just a query but wouldnt they (ATI )have already allowed for a margin of waste which would already be written off against in the budget before a die was cast.


Could do, but then you would just restate your earning/expenses. There's a few ways of writting it down, but until they have a SKU to use those parts on then their effective value is 0, but just like finding a new use for X or Y widget, it's a loss or a 0 until that use is empyed.

Quote:
And also wouldnt they have known and therefore already accounted for the fact that they could have fails to resell as a different spec of the range.


Once again that depends on other factors, like do you have another cheaper to produce part that will better fit that range that comes out before you have sufficient quantities to make another SKU? If the RV670 was ready for launch at the same time the RV630 did, would they have any market for such a part? They wouldn't throw the pats out, but until you can find a market opportunity to them, they are still a 0 on the balance sheet.

Quote:
Or do you think that having a fail is a main pain financially as i would have thought that the business model of the process wouldnt allow for more than about 3%.


The failure rates of the G80 and R600 are much higher than 3%, you're talking about double digits, and early on significantly large double digits. As time goes on the process improves, but you bin what you can if you can and keep them for the potential of future sales. However if your top of the line competes against the #2 or #3 part from the competition, your opportunities to sell those failed parts is greatly reduced. I think if the HD2900XT had blown the Ultra out of the water, you'd see crippled Pro much sooner, because the revenue from the full R600 would be higher, and the revenue from the cripples would also be higher because they could get a higher price equilibrium. But without that opprotunity from the start, likely AMD's plans for the cast-offs was greatly diminished.

Quote:
If the margins are this tight then the Pro would represent loss reduction rather than free cards as it were.


And it depends on how they accounted for them originally, loss reduction is the same things. And remember the 'cards' aren't free, it's the chips that are already a sunken cost, you don't have a choice to make them, they are a by-product of making the marquee product. So selling off the also-rans depends on the market you can generate for them.

Quote:
But on the other hand if its all accounted for fails and all then ATI are essentially printing free money with the Pro


Once again it depends. Accounting and Economics aren't the same thing. If ATi knows from the start that they will get something for the parts inthe future then they can account for it by amortizing the loss at a rate that gives them the best return on that expense; but they can also considering it as a cost of production of their primary product where they don't build in a value to those bin'ed parts, which is helpful if you are trying to do something for taxes. This is more effecive when you don't have a good idea of what that future value will be, but it's also not great for the bottom line and the stock prices unless you have a longer term strategy to make it appear like positive growth in the future. If they had the HD2900Pro from the start like they have in the past with the R9700/9700Pro, X800Pro/X800XT , X1800XL/XT then it's easier to account for it, but seeing as it's been more than a quarter since the launch of the HD2900, then it makes it diffcult to have it be a current-cost/loss or even to properly amortize it, but the later is the best way to do it from an accounting perspective, but I'm talking about it from an economic perspective that looks at it from both the very short and very long term simultaneously.

Anywhoo, from the economic perspective this is essentially a sunken cost, but from an accounting perspective you can report it 8 ways from Sunday, of course it used to be 50 before SOX.
September 25, 2007 7:26:47 PM

Ordered a HIS 2900Pro :)  My 7800GT was starting to lag, so I'm hoping the 2900pro will be good for upcoming games. If I can overclock to XT speeds that would be awesome, but I have a feeling it wont.

Thanks for the advice given
Regards
Speedbird
a c 130 U Graphics card
September 25, 2007 8:29:53 PM

To TheGreatGrapeApe
No worries about the impromptu edit the last bit was basically what do you guys make of my logic on the subject, and you more than answered that.
I'm quite surprised that a company of the size ATI are would run waste/failures at such a level i have worked in three different production environments(that's where the 3% came from) from packing to management level and would have been skinned alive for waste figures of 5% never mind double figures.
Mactronix :) 
a b U Graphics card
September 25, 2007 8:51:13 PM

Anoobis said:
I don't think you're hearing me TGGA. These are all ifs but if the HD2900Pro turns out to be a solid contender against the 320MB 8800GTS, ATI will be competitive again. Not only will they be competitive, but they will be more competitive in the mid-range market where they're already strong. The mid-range market which brings in more money than the uber high-end range. Unless nVidia has something new to fire back with that's better and cheaper to make than the 320MB 8800GTS, the only way they can compete with ATI at the moment is to lower prices which is good for us.


I agree, but the GF8700 is almost a given, I'm still not convinced of the RV670 reaching market before '08. We've heard positive news right before learning about delays before. And like I mentioned the HD2900Pro will likely sell well, but AMD definitely doesn't want it to be their long term solution.

Quote:
My argument is that nVidia is in a strong position and can afford to do this and that this HD2900Pro could be a godsend for ATI because it's exactly what they need. A solid performing yet well-priced mid-range DX10 (regardless of how trivial that really is) card. I hope it sells like hot-cakes. If it turns out good I'm going to start wondering how well it stacks up against my X1950XT.


I agree with the overall sentiment, but I disagree with this being what they need. It is short term (clear out the old bined parts for sure), but their long term benefit/health/desire IMO is the rV670 which if anywhere near the rumours, will be exactly what they want in the segment, and IMO once it's ready then wherever it performs will displace the R600 variant there, so if it performs as well as the Pro then the left over go to the GT.

Quote:
Inventory that cost them money to make. Inventory that they can now sell with these HD2900Pros.


No, but that's not how the cost is associated, since there was no Pro from the start. It didn't cost them the money to make the Pro as it was not what they were making, it's the XT they were making and the costs are associated with that production. Saying it's inventory that cost them money to make associates the cost with making the Pros and not what it's really the cost of making which is the XT.

Quote:
Honestly man, I don't know where you keep getting this, but I haven't once compared the upcoming 8700GTS (or whatever they call it) to the the 320MB 8800GTS. I never said, nor do I expect the 8700GTS to compete or beat the 320MB 8800GTS or this new HD2900Pro. I hope for it to be better than the dog that is the 8600GTS. If it is, it's just yet another bit of bad news for ATI.


Well I got that idea from this statement "...This coupled with the possibility that nVidia may come out with a much more competitive mid-range card than their crap 8600GTS doesn't help ATI much." that refers to this series of cards unless you're talking about some other parts. Obvious the 'much more competative mid-range is the GF8700 and then in relation to the rest of the discussion then the two in that equation would be the GTS-320 it's replacing and the HD2900Pro it's up against. Did you mean some other coupling of cards/events?

Quote:
Like my note said above, I don't expect the 8700GTS to compete at all with the 320MB 8800GTS. I expect it to replace the horrible 8600GTS and deal with nVidia's lack of competition against the stellar X1950Pros and X1950XTs. Again, that would be a spot of bad news for ATI.


Well based on early rumours (which is still this segment's bread and butter) the GF8700 model is suposed to perform close to the GTS-320 due to it's higher clockes, but somewhat less memory bandwidth. So it depends on your look of the market, IMO the GF8700's lesser parts will end up being the GF8600GTS' replacement, where the GF8700GTX or whatever replaces the GF8800GTS-320 and the GF8700GS replaces the GF8600GTS.

Quote:
No matter what the HD2900Pro needs to have price/performance viability against the GTS-320 under its current inflated (IMO) pricing


No argument there, but it is the longer term strategy that nV has to deal with, whereas this HD2900Pro is little more than dumping chips, whatever the GF8700's strategy it has to be more concrete, just like the RV670.

If you don't see the GF8700 in this market segment, then it doesn't make as much sense, but considering that all the early statements say that it is going to replace the GTS-320, then that's why I treat it as such and as the HD2900Pro's primary competition once it reaches market (likely after the HD2900Pro) at which time you're right, then the pressure is on AMD to bring that RV670 to market as well. Which is the whole reason why I think the HD2900Pro is a very short term solution, for two reasons, A) sell the cheap HD2900Pro before nV can get their GF8700 to market, which cuts into nV's profits on the GTS if they lower the GTS' price, and also gives AMD something in that segment, and sell off as many of the crippled R600 chips while you still can, because the moment the RV670 arrives, if erformance is the same, then you couldn't sell one for anywhere near the same money because the RV670 will have the power consumption and UVD advantage over the Pro.

Either way we won't know until they finally do arrive.
a b U Graphics card
September 25, 2007 9:42:28 PM

mactronix said:
To TheGreatGrapeApe
No worries about the impromptu edit the last bit was basically what do you guys make of my logic on the subject, and you more than answered that.


Cool, glad I didn't damage it too much.

Quote:
I'm quite surprised that a company of the size ATI are would run waste/failures at such a level i have worked in three different production environments(that's where the 3% came from) from packing to management level and would have been skinned alive for waste figures of 5% never mind double figures.


Yeah it's really amazing, at first they (both AMD and nV) usually run much less than 50% succes for the first few 300mm wafers then get up to a greater succes rate from there. But it also of course depends on the chosen speed, if they lowball the speed because they don't need much faster then their yield is much better. If you're struggling to make the fastest part possible, then your yield suffers greatly because only so many chps can go so fast, and then others ust plain don't have all part working (some defective shaders or ROPs or what not) and then others that just don't plain work at all. It's really not that uncommon, and it was suposedly one of the beneficial aspects of IBM's agreement with nV for the NV30 chips made at the Fishkill fab, nV only paid for working parts. However places like TSMC basically charge you per wafer regardless of the good/bad. As time goes on the yields improve, and you also have lower binned SKUs sometimes (SEs, GSs, LXs, GTOs, etc).
If they were making general purpose parts (like the Rage chips still used in servers) then likely they would have perfected the process and have low enough target to get to the low single digit wastage, but with this vboutique marquee, "better , stronger , faster" mentality wanting to beat the competition, they pay for it in lower yields and higher 'failure' rates. Overall they likely still sell somewhere close to 80-90+% of the chips on each wafer, but the number that make the top notch part which is the best case scenario goal is usually well below that number. I doubt that nV is getting 30% yield for Ultra eligible parts on the G80, but due to the strength of their other models it doesn't matter. ATi on the other hand has to make all of their chips their very best chips right now. That's definitely hurting teir original strategy which likely relied on us seeing some castoffs sold against the GTS-640 initially, rather than now so much later and against the GTS-320.
a c 130 U Graphics card
September 25, 2007 10:05:51 PM

That certainly makes for some interesting and insightful reading ape do you recon that the new architecture is whats causing the ATI boys the problems?
Just to clarify when you say
"ATi on the other hand has to make all of their chips their very best chips right now. That's definitely hurting teir original strategy which likely relied on us seeing some castoffs sold against the GTS-640 initially, rather than now so much later and against the GTS-320."
Do you mean that they are concentrating on the good chips in the wafers and setting the failures aside until now or that they are some how oh i dont really know how to put it say slowing the process down to get only good chips or some such?
Mactronix
September 25, 2007 10:56:55 PM

Quote:
Anoobis wrote: My argument is that nVidia is in a strong position and can afford to do this and that this HD2900Pro could be a godsend for ATI because it's exactly what they need. A solid performing yet well-priced mid-range DX10 (regardless of how trivial that really is) card. I hope it sells like hot-cakes. If it turns out good I'm going to start wondering how well it stacks up against my X1950XT.


Quote:
TGGA wrote: I agree with the overall sentiment, but I disagree with this being what they need. It is short term (clear out the old bined parts for sure), but their long term benefit/health/desire IMO is the rV670 which if anywhere near the rumours, will be exactly what they want in the segment, and IMO once it's ready then wherever it performs will displace the R600 variant there, so if it performs as well as the Pro then the left over go to the GT.

Yes I see what you mean and I guess I should have worded that a bit different. I didn't see this HD2900Pro as something for ATI to pin their hopes on. I meant it was a source of quick income for them.

Quote:
Anoobis wrote: Inventory that cost them money to make. Inventory that they can now sell with these HD2900Pros.


Quote:
TGGA wrote: No, but that's not how the cost is associated, since there was no Pro from the start. It didn't cost them the money to make the Pro as it was not what they were making, it's the XT they were making and the costs are associated with that production. Saying it's inventory that cost them money to make associates the cost with making the Pros and not what it's really the cost of making which is the XT.

Yes, I see what you mean. I should have been more specific. I don't see the HD2900Pros as new inventory they had to make. I see them as HD2900XTs that couldn't perform to the HD2900XT spec but do function perfectly at lower clock speeds. This is a common practice with CPU manufacturers. However, regardless of what a card starts out to be and ends up to be isn't where I was headed. The cards still cost money to make, bottom line. That's the point I'm making. If ATI doesn't use them, it's inventory that they have to drag along like an anchor.

Quote:
TGGA wrote: Well I got that idea from this statement "...This coupled with the possibility that nVidia may come out with a much more competitive mid-range card than their crap 8600GTS doesn't help ATI much." that refers to this series of cards unless you're talking about some other parts. Obvious the 'much more competative mid-range is the GF8700 and then in relation to the rest of the discussion then the two in that equation would be the GTS-320 it's replacing and the HD2900Pro it's up against. Did you mean some other coupling of cards/events?

I can see where you're coming from on this and I suppose I should have worded that better. I see the 8700GTS in a different light. I see it as a replacement to the 8600GTS and the direct competitor to the X1950Pro and X1950XT. Those cards pose a major threat to nVidia because they're such strong performers. The 8700GTS to me is what the 8600GTS should have been. Apparently I'm incorrect though. I didn't see or know the 8700GTS was supposed to replace the 320MB 8800GTS.

Quote:
TGGA wrote: No matter what the HD2900Pro needs to have price/performance viability against the GTS-320 under its current inflated (IMO) pricing

No argument there, but it is the longer term strategy that nV has to deal with, whereas this HD2900Pro is little more than dumping chips, whatever the GF8700's strategy it has to be more concrete, just like the RV670.

Umm, that first quote was your's not mine???? Perhaps if we had a decent forum interface that wasn't complete crap like this new one and allowed us to properly quote messages, there wouldn't have been any confusion.
a b U Graphics card
September 25, 2007 11:35:45 PM

For mactronix....

No it's a little different, I'm saying that for whatever reasons (poor design, bad 80nmHS speed limits, whatever) that AMD's fastest part is not fast enough to compete with nV's 2nd fastest part so it competes with their 3rd fastest part sometimes even their 4th fastest G80 part. So they don't get the luxury of selling a much slower part, because with the XT sometimes losing to a GTS-320, anything significantly lower (that they'd have a noticeable benefit in yield %) would likely fall below the GTS-320 in performance. So like Anoobis mentioned they need to sell it for below the price of the GTS-320. This is doable, but likely not a long term strategy, since once again that level of even 75% as fast as the R600XT (with all the functioing SPUs) is probably still a small number compared to dies that can't even go 500mhz, or with damaged SPUs, ROPs, etc. which if they could sell (and might yet as GTs) would give them more options. However right this minute and for all this time after launch they've only had one option, and that's to sell whatever they can get off that wafer as an XT and nothing else. They can add 512MB and make 2 different cards, but the core chip is the same, and that's where AMD's money is made/lost essentially.

Now they are suposedly adding potentially 2 outlets for their castoffs, the PRO and the GT, but if the PRO is anywhere near the specs we've heard about it likely doesn't give them anywhere near as much castoff potential as a GT.

So if you look at it, if nV has a part that can't go Ultra speeds, then it becomes a GTX, if it either can't go GTX speeds or has some damaged SPUs then it becomes a GTS, and then of those they can further differentiate their market presence by chosing 320MB instead of 640MB, then beyond that, they become shelf products until another SKU comes out or they end of life them and dust-bin them. All of which means they are likely using more of each wafer than ATi is where basically they have 1 option (with 2 memory sizes) and where the parts that don't meet that spec currently are doing nothing, until the PRO and GT there was no outlet for the parts that weren't up to snuff. However if the R600XT at launch were say 10% faster/better than the Ultra, then the slower parts could compete with the GTX, and the crippled parts could compete with the GTS. But since ATi didn't end up with a part that could do that, their very best sell for much less than they hoped, and against the lower side of the G80, which means any cast-offs have to sell for even less and compete outside of that class entirely. And by the amount of time it took to make it to HD2900Pro launch it seems that the yield of those chips isn't much different from the XTs, and so it took a heck of a long time to reach volumes large enough to be able to sell HD2900Pros without basically selling functional XTs as Pros, which makes less sense than selling the XTs for a few bucks less.

Now slowing the process down doesn't help since increasing the number of wafers produced is really the only way to get better at making the wafers, and thus making more viable dies per wafer.

Another thing though is that from an accounting perspective the revenue from the limited number of Pros and GTs that will be sold can be rolled into the cost of the R600, either in recouping R&D cost, or else financing price cuts on the XT, which up until now has had to finance everything on it's own. I don't suspect much change in the XT's price, but I wouldn't be surprised if after the Pro arrives, that the XT doesn't drop a little bit closer to the GTS-640's price. Overall that would seem to hurt it, but if let's say from every wafer AMD gets 50-XTs (which would be a very low number since the G80 can theoretically max out at ~140-150 per 300mm wafer and is about the same size as the R600 +/- mm) and then also 10 Pros and 15 GTs then while they are still a long way from high yield only being able to sell about 50% of the possible chips per wafer, they'd be getting much more money than when that was 33% even if the cost per chip were dropped for the XT.

So think of it like this, AMD makes their money by selling the chips to the AIBs, who then sell them to us. And view it that for every $1 charged to the AIB usually translates to about $5-10 to the consumer based on past statements of this relationship. Now, If AMD was selling the XT for $50 per chip before the new products getting $2,500 per wafer, well now that they have a 3 part lineup they can now do some strategic pricing and sell the above yields at $45 for the XT , $30 for the Pro, and $20 for the GT, then they can now get from that same 300mm wafer $2,250 for the XTs, $300 for the Pros, and $300 for the GTs, which means that they can now put products into each of the three markets, at lower prices than before and still make a profit of $350 more per brand new wafer than they did before, even with only the Pros or GTs that would be $50 more per wafer. AND they would also start making additional money from the chips they already binned which either were written off already or amortized long term. But it shows that even current production would still benefit even while dropping the price, and all the while the yield is still terrible and the yield of the lower end parts has only adding half the number of chips you had before (not much if you think about it). this is why those additional SKUs are so important. Now the biggest problem is somewhat illustrated in my above example. What do you think the demand of a much cheaper HD2900Pro and GT are going to be compared to the HD2900XT? do you think combined they'd be equal to the number of people buing XTs or do you think each of them is likely to outstrip XT demand based on damand at a given price? So if you want to sell this part at an attractive price, then you need to have much larger volumes than the XT, and with your current yield formula if you simply pluck them from the current wafers, then you could ruin the balance of returns, so unless your yield numbers are more along the lines of 15-XTs 25-Pros and 50-GTs, then you want to stockpile alot of Pros to make up for the demand without compromising the XTs which can still make a large profit at a lower price (to better compete against it's rival, which in turn increases the demand for entire wafers).

I'm not sure how many times I've overlapped in the above, but I hope it shows how even with the know low yields, and likely minimal increased yields of only slightly lower clocked Pros (which in my example add only 20% chips) could have a great impact. If you look at it like an accountant and try to make up for lost dollars that were already part of last quarter, you're more screwed because you're looking to make up for money lost in previous production while the most important thing right now is to increase positive profits. It would've been much easier if the R600 came out beating the Ultra because then that yield would mean that not only could ATi have maybe charged $75 for each of those 50 XTs above, but they would've had a much more attractive market for cast-offs, and then getting $50 for each of those PROs and $40 for each of those GTs might have meant from the start they'd have much more reason to be making those parts, even if it meant that they lost some XTs to get enough PROs and GTs, because overall they'd be making much more money, especially much more than the wound up with with essentially a part that fell short of their task.

Anywhoo, understand that nV has already gone a long way to maximize their profit per 300m wafer with the Ultra, GTX, and two GTSs. I've heard they currently get around 90-110 useable chips per wafer (and would be about 60-80% yield if the 140-150 max were accurate), which means they're able to get as much money as they can for each one of those wafers TSMC sells them. Up until now all ATi's been able to get is 1 single chip, the XT, whose yield rate sounds much much lower.

Of course tis is only 1 part of the equation as the cost of boards, memory, etc sets a limit on how much price changes they can influence, but as the chip business is really what AMD and nV sell, not the boards (AMD's small BBA and FireGL market doesn't count IMO), this is how they have to plan and sell their parts.

Alright gonna stop now beause I'm tired and just finishing up work and I think I'm kinda going on and on :pt1cable:  , but hopefully it made sense overall.
a b U Graphics card
September 25, 2007 11:43:13 PM

Anoobis said:
Umm, that first quote was your's not mine???? Perhaps if we had a decent forum interface that wasn't complete crap like this new one and allowed us to properly quote messages, there wouldn't have been any confusion.


Yeah, essentially I know we agree in principle, I think it's the finer details that are in question. and yes the forum interface doesn't help, especially when I have extra buttons for the mod stuff, and the posts get so long I trip over my own statements.

I understand what you're saying about carrying over the HD2900Pro in inventory but like I mention in my reply to Mactronix, I think that's accountant thinking, which should be used to decide on new SKus and such. It's good to help move around final costs and such, but it's terrible when trying to decide on the products on a go forward basis as it relies even more on guessing the value of the inventory and how much you should charge for costs that have already occured instead of trying to maximize current profits. I know everyone does it, heck we do it here at work too, but it's a very bad practice IMO.
a c 130 U Graphics card
September 26, 2007 4:32:03 AM

To TheGreatGrapeApe
Thank you for Your time and effort.you have explained it perfectly i now have a lot better understanding of how it all works.
Mactronix :) 
!