Anand's review of the 9800GTX+ is up. Maybe the first post should be changed to: "Faster than 4850 in some games and settings, slower in others."
Anand's conclusion:
"The overclocked, 55nm 9800 GTX+ manages to rarely outperform the 4850 but for the most part isn't competitive enough to justify the extra $30. AMD was quick to point out that by the time the 9800 GTX+ ships that it will also have factory overclocked 4850s, which should make things even more interesting. Because, honestly, a factory overclocked Radeon HD 4850 is far more attractive to us than an overclocked GTX+. "
http://www.anandtech.com/video/showdoc.aspx?i=3340&p=1
Anand's conclusion:
"The overclocked, 55nm 9800 GTX+ manages to rarely outperform the 4850 but for the most part isn't competitive enough to justify the extra $30. AMD was quick to point out that by the time the 9800 GTX+ ships that it will also have factory overclocked 4850s, which should make things even more interesting. Because, honestly, a factory overclocked Radeon HD 4850 is far more attractive to us than an overclocked GTX+. "
http://www.anandtech.com/video/showdoc.aspx?i=3340&p=1
Maybe I missed it....but does anyone know when these things are actually supposed to be available?
JPB
Diamond Member
- Jul 4, 2005
- 4,064
- 89
- 91
The only part I find any truth in...is the highlighted part. And I do understand they have hard feelings against NV for cutting them off.
Its the Inquirer
Nvidia price protects channel partners
NVIDIA IS DOING a mighty strained dance right now over finances, one might almost call it twitching. In its usual way, the company is trashing partners to make its bottom lines look a little better.
The big problem came when Marv Burkett, Nvidia CFO, mouthed off on last quarter's conference call a couple of months ago, promising the financial world to raise margins by 100 basis points (one per cent for the non-finance people out there). The problem? He had to have known it couldn't be done. If we knew about massive problems like GT200 yields just days later, he had to have known when he was on the CC. That is between him and the SEC though.
All this is fine and dandy, but compounded with the 9800GTX+ fiasco, it brings things to a whole new level. A week ago, the 9800GTX parts were selling for $270 or so. Nvidia then realised it was going to get its teeth kicked in by the 4850, and dropped the price to $199.
Nvidia will look you in the eye and tell you that it is because of a shrink of the G92 from 65nm to 55nm, aka the G92b. That explanation doesn't hold water, though. The difference between the BoM (Bill of Materials) for the two boards is almost entirely down to the silicon. For the G92, the die was 17x18mm, 306mm^2 for the math impaired. The shrink is 15.5x15.5mm or 240.25mm^2.
Pulling out the handy dandy INQUIRER TSMC Silicon Costing Calculator (R)(C)(TM) (ITSMCSCC for short, rhymes with 'FDIV bug'), we get cost of about $49 for the G92 and $39 for the G92b. This means that to meet 50+ per cent margins, they need to sell the kits for minimum $100 and $80 respectively, likely more when you take packaging into account.
If you recall back to that famous conference call, NV blamed the margin drop on crappy G92 yields. Fair enough. Now it is doing a shrink which increases defectivity going from 65 to 55, but gains them more die candidates, and gives a little better electrical characteristics. On top of that, it is upping the bins from 675MHz to 738, roughly 10 per cent higher clocks. There goes any cost savings from the shrink, but the fab gets lots of drop out 675MHz parts.
In the end, the cost savings from the shrink are little if any, but we will go with the $10 figure until we get the real numbers from our NV engineering moles in a few days. Let's also be generous and assume that the kit cost drops for both parts by $20, and that the boards go down by $5, something that is very unlikely, but not totally impossible.
With the 9800GTX priced at $200, the GTX+ priced at $230, and last week the GTX used to cost $270, we can deduce that $70 and $40 went to the great margin pool in the sky. Now, cost saving will eat $25 or so of this, leaving $45 and $15 respectively unaccounted for. Guess who eats that?
Well, immediately Nvidia does. In a generous move, the firm is price protecting the entire channel, top to bottom. People inside the green nuthouse tell us this is costing "tens of millions", but would not specify if it is high or low tens. Whoops, there go margins for the next CC... unless...
The 'unless' basically erases any goodwill the company might have earned with price protection, it is nuking MDF funding, so no more kickbacks for you. To make matters much happier, the top brass are simply reneging on already promised kickbacks, or so several partners are howling. The only reason to do this is to artificially pump up the margins next quarter on the backs of their partners. Wall Street, pay close attention to the details in the next Q's CC.
The real problem here is that while NV may make 50+ per cent margins, the partners are barely scraping by. 8800 class boards can't be made at a profit anymore, even on top of the $20 NV kickbacks, so the MDF funds were the only thing keeping some of them in the black... barely.
There is one huge NV partner which is essentially bankrupt, being kept alive only because it owes NV far too much money, and writing that off would raise far too many questions. Two or three others are looking longingly at greener pastures, or in this case, less green, and almost none is making any net money on NV products. The MDF removal will likely tip a few over unless NV acts quickly.
If these companies act quickly though, it will lessen their ability to cover up the margin loss, increasing their exposure to real questions during the next CC. If you know the egos involved, you probably realise how adverse they are to concepts like having to answer questions directly with consequences attached. Rock and a hard place.
To sum things up, if NV turns one way, it is screwed, if it turns the other, the partners are screwed. It's covering its own ass by taking partner funding and propping up an increasingly shaky bottom line. This is temporary though, one has to wonder about the partner count by year end, and what Nvidia will cover those losses up with.
It's going to be a fun thing to watch. µ
Its the Inquirer
Nvidia price protects channel partners
NVIDIA IS DOING a mighty strained dance right now over finances, one might almost call it twitching. In its usual way, the company is trashing partners to make its bottom lines look a little better.
The big problem came when Marv Burkett, Nvidia CFO, mouthed off on last quarter's conference call a couple of months ago, promising the financial world to raise margins by 100 basis points (one per cent for the non-finance people out there). The problem? He had to have known it couldn't be done. If we knew about massive problems like GT200 yields just days later, he had to have known when he was on the CC. That is between him and the SEC though.
All this is fine and dandy, but compounded with the 9800GTX+ fiasco, it brings things to a whole new level. A week ago, the 9800GTX parts were selling for $270 or so. Nvidia then realised it was going to get its teeth kicked in by the 4850, and dropped the price to $199.
Nvidia will look you in the eye and tell you that it is because of a shrink of the G92 from 65nm to 55nm, aka the G92b. That explanation doesn't hold water, though. The difference between the BoM (Bill of Materials) for the two boards is almost entirely down to the silicon. For the G92, the die was 17x18mm, 306mm^2 for the math impaired. The shrink is 15.5x15.5mm or 240.25mm^2.
Pulling out the handy dandy INQUIRER TSMC Silicon Costing Calculator (R)(C)(TM) (ITSMCSCC for short, rhymes with 'FDIV bug'), we get cost of about $49 for the G92 and $39 for the G92b. This means that to meet 50+ per cent margins, they need to sell the kits for minimum $100 and $80 respectively, likely more when you take packaging into account.
If you recall back to that famous conference call, NV blamed the margin drop on crappy G92 yields. Fair enough. Now it is doing a shrink which increases defectivity going from 65 to 55, but gains them more die candidates, and gives a little better electrical characteristics. On top of that, it is upping the bins from 675MHz to 738, roughly 10 per cent higher clocks. There goes any cost savings from the shrink, but the fab gets lots of drop out 675MHz parts.
In the end, the cost savings from the shrink are little if any, but we will go with the $10 figure until we get the real numbers from our NV engineering moles in a few days. Let's also be generous and assume that the kit cost drops for both parts by $20, and that the boards go down by $5, something that is very unlikely, but not totally impossible.
With the 9800GTX priced at $200, the GTX+ priced at $230, and last week the GTX used to cost $270, we can deduce that $70 and $40 went to the great margin pool in the sky. Now, cost saving will eat $25 or so of this, leaving $45 and $15 respectively unaccounted for. Guess who eats that?
Well, immediately Nvidia does. In a generous move, the firm is price protecting the entire channel, top to bottom. People inside the green nuthouse tell us this is costing "tens of millions", but would not specify if it is high or low tens. Whoops, there go margins for the next CC... unless...
The 'unless' basically erases any goodwill the company might have earned with price protection, it is nuking MDF funding, so no more kickbacks for you. To make matters much happier, the top brass are simply reneging on already promised kickbacks, or so several partners are howling. The only reason to do this is to artificially pump up the margins next quarter on the backs of their partners. Wall Street, pay close attention to the details in the next Q's CC.
The real problem here is that while NV may make 50+ per cent margins, the partners are barely scraping by. 8800 class boards can't be made at a profit anymore, even on top of the $20 NV kickbacks, so the MDF funds were the only thing keeping some of them in the black... barely.
There is one huge NV partner which is essentially bankrupt, being kept alive only because it owes NV far too much money, and writing that off would raise far too many questions. Two or three others are looking longingly at greener pastures, or in this case, less green, and almost none is making any net money on NV products. The MDF removal will likely tip a few over unless NV acts quickly.
If these companies act quickly though, it will lessen their ability to cover up the margin loss, increasing their exposure to real questions during the next CC. If you know the egos involved, you probably realise how adverse they are to concepts like having to answer questions directly with consequences attached. Rock and a hard place.
To sum things up, if NV turns one way, it is screwed, if it turns the other, the partners are screwed. It's covering its own ass by taking partner funding and propping up an increasingly shaky bottom line. This is temporary though, one has to wonder about the partner count by year end, and what Nvidia will cover those losses up with.
It's going to be a fun thing to watch. µ
Denithor
Diamond Member
- Apr 11, 2004
- 6,298
- 23
- 81
How come no one is looking at SLI 9800GTX+ yet? Afraid of encroaching on the GTX 280 for $200 less?
Regarding the whole PhysX/CUDA issue, is the GTX 280 expected to have simply 240/128 more computing power (versus 9800GTX stream processors) or is there more to it than that? And how then does this compare to the 800 stream processors on the AMD cards?
Regarding the whole PhysX/CUDA issue, is the GTX 280 expected to have simply 240/128 more computing power (versus 9800GTX stream processors) or is there more to it than that? And how then does this compare to the 800 stream processors on the AMD cards?
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
I haven't looked to see who wrote that article yet, but it actually looks almost well thought-out. If the inq wasn't so obviously biased then they might be taken seriously some times. It's almost like in men in black where tommy lee jones grabs the natl enquirer to find out where the aliens are...
and regarding the 9800gtx black that's great, but it's got to be creeping too close to $299...hmmm, which would I rather own, an overclocked 9800gtx or a 4870...tough choice there...I dunno, what do you guys think?
and regarding the 9800gtx black that's great, but it's got to be creeping too close to $299...hmmm, which would I rather own, an overclocked 9800gtx or a 4870...tough choice there...I dunno, what do you guys think?
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
Originally posted by: keysplayr2003
Bryan, isn't going from 65nm to 55nm just an optical shrink? Like 90nm to 80nm? Meaning, no reworking of the transistors are necessary? in turn, meaning that TSMC is more than familiar with G92b, because it is the same core as G92 which TSMC has been fabbing for Nvidia for quite a while now. Now they are just shrinking it?
TSMC 65 to 55nm linear shrink.
"TSMC?s 55nm process technology is a 90% linear-shrink process from 65nm including I/O and analog circuits. The process delivers reportedly significant die cost savings from 65nm, while offering the same speed and 10 to 20% lower power consumption. Because the 55nm process is a direct shrink, chip designers can leverage existing libraries and port their 65nm designs with minimal risk and effort. The 55nm logic family includes general purpose (GP) and consumer (GC) platforms. Initial production of the 55GP begins this quarter, followed later in the year by 55GC."
So your comment I bolded above. Did you just throw a guess out there on that one? Because G92b can't be anything other than a knee jerk? I have to admit, when you guys get your mind set on something, all kinds of cool stuff comes out.
G92b has been a topic for discussion for a while now. The timing of it's release makes sense. They had it ready to go, and were just waiting for the proper time to announce it.
It's always good to have an answer to your competitors products, both in performance and price point. TSMC was probably waiting for Nvidia to give word to ramp up production, hence the July availability reports. Why make something if you don't need it? While the current 9800GTX is fairly close in performance to the 4850, the 9800GTX+ should be a bit quicker. The price point of 4850 is 199.00, but if they implement a better cooler on it, that price may rise a few bucks. All in all, these cards will line up well against each other.
I'd like to see a quick fix for the fan speeds on the 4850, similar to the 8800GT fan speed fix. Either that, or a better cooler.
Sorry for the long postage.
keys
I realize that it's just a linear shrink, but I think that anand hit the nail on the head with this one:
"A very smart man at Intel once told me that when designing a microprocessor you can either build a new architecture, or move to a smaller manufacturing process, but you don't do both at the same time. The reason you don't do both is because it significantly complicates the design, validation and manufacturing processes - you want to instead limit the number of variables you're changing in order to guarantee a quick ramp up and good yields of your silicon.
NVIDIA followed this rule of thumb with the GT200, building its "brand new" (or at least significantly evolved) architecture on a tried-and-true 65nm process instead of starting at 55nm. Despite AMD building both RV670 and the new RV770 GPU on TSMC's 55nm process, NVIDIA hadn't built anything on a smaller than 65nm process, including the 1.4 billion transistor GT200."
JPB
Diamond Member
- Jul 4, 2005
- 4,064
- 89
- 91
XFX 9800GTX Black Edition
I have considered this card myself. Now if only it will go down to $199.99 :thumbsup:
Well this just shows Anand is in fact human and is capable of making mistakes. The RV670 wasn't a new architecture on a new process, it was an R600 refresh on a smaller process. RV770 is the new architecture built on a proven process which is no different than what NV did with G80 and GT200 at 90nm and 65nm, respectively. The only difference this time around is NV will go with an optical shrink half step to 55nm instead of a full process node (45nm or whatever is next).
His conclusion doesn't jive with his results either, with the 9800GTX+ winning 4 of the 7 games tested. Personally I think they half-assed the preview with mostly archived results but were obligated to get something out on the 9800GTX+ so they could justify posting 4870 benches tomorrow. Fair coverage to both on their upcoming paper launches (which is something AT said they would no longer do). Just shows review sites are under pressure to keep up as well.
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
$257 shipped - $20 mir isn't good enough, I agree. the problem is that even if the black edition were $20 you'd still be better off with an oc edition 4850 with a dual slot cooler. That's the scary part here for nvidia: actual retail prices on 4850 are probably going to stabilize in the $180 range once newegg etc run out of "free" witcher and other extras, so the oc/dual slot editions should be available for ~ $199. What does nvidia do then, offer 9800gtx for $150 and 9800gtx+ for $175??? Nvidia still has gpu market dominance and will be able to sponge off of g80/g92's past success for a few months, but they need to get gt 200 down to the midrange NOW if they intend to get their margins back to acceptable levels for the rest of this round.
Now, ben has brought up a valid point that tesla is kicking the crap out of intel right now on the ultra high-end performance market. This bears intense scrutiny as this could very well enable nvidia to greatly reduce gt 200 and g92(b) prices but still keep overall margins acceptable. Funny thing is, tesla still needs cpus to run with it. Intel cpus? probably not... hmm, what about opteron cpus? isn't that new supercomputer that just broke the petaflop barrier a tesla/opteron combo? Once again, amd still wins. Intel hates nvidia so they have crossfire but not sli on their mobos. Nvidia and intel hate each other so tesla gets combined with opterons...maybe we'll see amd continuing for a while longer yet by leveraging intel vs nvidia to their advantage.
Now, ben has brought up a valid point that tesla is kicking the crap out of intel right now on the ultra high-end performance market. This bears intense scrutiny as this could very well enable nvidia to greatly reduce gt 200 and g92(b) prices but still keep overall margins acceptable. Funny thing is, tesla still needs cpus to run with it. Intel cpus? probably not... hmm, what about opteron cpus? isn't that new supercomputer that just broke the petaflop barrier a tesla/opteron combo? Once again, amd still wins. Intel hates nvidia so they have crossfire but not sli on their mobos. Nvidia and intel hate each other so tesla gets combined with opterons...maybe we'll see amd continuing for a while longer yet by leveraging intel vs nvidia to their advantage.
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
go read that expreview article if you haven't yet. they go into quite a bit more detail, but the 2 articles do generally agree with each other. this is very bad news for nvidia right now.
re the paper launches, the only reason we haven't seen any from either camp for the past couple of years is that nvidia has been too dominant. Every time they released something (gtx, ulta, 8800gt, etc) amd simply had no answer. now that the competitive game is going again they are both going to be doing a lot of paper launches in the future imho.
The expreview article is in Chinese, how am I supposed to read it? And no Babelfish isn't the answer, I don't want to read about analogies of "two red dragon shoot over green dragon flying" translated literally. With the actual review numbers, the 9800GTX actually performs closer to the 4850 than other reviews showed. The 4870 does look strong though and impresses where the 4850 does not.
Not really sure what the point of your whole process argument is. There is no advantage as long as both go to the same fab unless one or the other has exclusive rights on a particular process. Anand is wrong, both are using a similar strategy to what that "wise man at Intel" advised.
Well I don't see how one is better than another really. 4870 is paper launching because of GDDR5 supply, 9800GTX+ is clearly a competitive response and short of supply due to ramp up time. But like I said, AT pledged a few generations ago that they would not review paper launched products and here they are, reviewing paper launch parts.
bryanW1995
Lifer
- May 22, 2007
- 11,144
- 32
- 91
yeah, I didn't read very much of the red dragon eats the green dragon's milkshake crap, either, but I DID look at the charts. 4870 was ~ 2 % faster across all tests with and without AA (roughly equal in both) and ~ 5% faster with lost coast taken out (they mentioned something about how it was like call of juarez is for amd but I got lost in the translation). Either way, 5 % is nothing to brag about unless your card has a $100 price advantage. I still very strongly believe that nvidia WILL make a significant number of gtx 260's and will thus be forced to price it competitively with 4870. I think that under normal circumstances amd would up the 4870 to $349 or even $399, but the 4850 would pwn it at that price so they're really stuck at $299 imho. This whole release has been weird to say the least as the midrange is actually dictating high end transaction prices. This could very well be the wave of the future as sli and crossfire become more and more seamless over time.
re the 4850 vs 9800gtx, wtf are you talking about? they showed that across all games/resolutions tested, the 4850 was 4% faster without AA and 20% faster with AA. a 9% core and shader bump isn't going to fix that.
re the 4850 vs 9800gtx, wtf are you talking about? they showed that across all games/resolutions tested, the 4850 was 4% faster without AA and 20% faster with AA. a 9% core and shader bump isn't going to fix that.
That's the point though, previous benches had the 9800 losing worst to the 4850, particularly with AA. The 9800GTX+ should actually mimic the results found in the various previews floating around, faster with no AA, usually faster with 4xAA, slower with 8xAA.
I've already stated why numerous times, its just a marginal improvement over G92 which has been available for longer at cheaper prices too. Much of this is based on personal experience with a G80 GTS, G92 GT and G80 GTX. On paper they showed quite a bit of difference in performance but in reality, there was very little difference. Maybe an extra setting here, 2x AA there, but nothing you could actually detect outside of a benchmark. Many others have echoed the same sentiment in this thread and others. My GTX 280 on the other hand is an entirely different story and the HD4870 from the benches looks to be a similar order of magnitude, building on the improvements in AA that the 4850 showed but raising performance to a genuinely playable level.
Except not really.
Cards like the 9800GTX are only competitive now BECAUSE the 4850 was so impressive.
When comparing to $150 cards like the 8800GT, it's still impressive since it pulls 10+FPS in many of the games benched by Anandtech. At least 5FPS in almost all. And in some places, it completely demolishes the 8800's (like Bioshock). We're seeing further drops now, but only because of the pressure from the 4850 (and now 9800GTX).
When the GTX280 and GTX260 was released, there was no such pressure because of the really high prices.
Really? Because the G92 GTS which by all reviews performs within 2-5% of a 9800 GTX wasn't at $200 or less for months now and as low as $160 MIR recently? Same for the G92 GT which was 90% of a G80 GTX selling for $180 or less for months, and as little as $120-130 recently? And how much improvement was G80/G92 over previous generations? 2-3x? 10 FPS is barely noticeable outside of an Excel spreadsheet unless your FPS are so low you're already staring at an unplayable mess (Crysis).
From Anandtech. Ignoring Crysis since you said it's a mess.
CoD4
4850: 66FPS
88GT: 52FPS
26%
ET: QW
4850: 75FPS
88GT: 48FPS
56%
AC
4850: 46FPS
88GT: 37PS
24%
Oblivion
4850: 35FPS
88GT: 31FPS
13%
The Witcher
4850: 34FPS
88GT: 29FPS
17%
Bioshock
4850: 87FPS
88GT: 57FPS
52%
Add in better AA, DX10.1 (which allows more better AA) and a price that forces the previous equivalent card to drop $100, I'd say the 4850 is a very important card.
As for the GTS, again, it's the pressure that has yielded the price drops to the point they are today (supposed price anyway since Newegg still lists them for $200).
CoD4
4850: 66FPS
88GT: 52FPS
26%
ET: QW
4850: 75FPS
88GT: 48FPS
56%
AC
4850: 46FPS
88GT: 37PS
24%
Oblivion
4850: 35FPS
88GT: 31FPS
13%
The Witcher
4850: 34FPS
88GT: 29FPS
17%
Bioshock
4850: 87FPS
88GT: 57FPS
52%
Add in better AA, DX10.1 (which allows more better AA) and a price that forces the previous equivalent card to drop $100, I'd say the 4850 is a very important card.
As for the GTS, again, it's the pressure that has yielded the price drops to the point they are today (supposed price anyway since Newegg still lists them for $200).
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes Discussion Threads
- Started by Tigerick
- Replies: 22K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 23K
-
Facebook
Twitter