Thursday, October 21st 2010
GeForce GTX 580 Expected to be 20% Faster than GeForce GTX 480
NVIDIA's next enthusiast-grade graphics processor, the GeForce GTX 580, based on the new GF110 silicon, is poised for at least a paper-launch by end of November, or early December, 2010. Sources in the video card industry told DigiTimes that the GTX 580 is expected to be 20% faster than the existing GeForce GTX 480. The new GPU is built on the existing 40 nm process, NVIDIA's 28 nm GPUs based on the Kepler architecture are expected to take shape only towards the end of 2011. Later this week, AMD is launching the Radeon HD 6800 series performance graphics cards, and will market-launch its next high-end GPU, codenamed "Cayman" in November.
Source:
DigiTimes
98 Comments on GeForce GTX 580 Expected to be 20% Faster than GeForce GTX 480
They'll probably just cut prices for the 480 this holiday and wait till next year to really come back with something.
or maybe nvdia have another sinister plan like tomshardware said :
Also I still strongly believe that Nvidia will release GF110 this year, because that must have been the plan since a long time ago, since they got the firts GF104 die on their hands. There's no news of any GF100 chip fabbed after the initial batches in Q1/early Q2. For example this is the GF100 on Wizzard's review of the MSI GTX480 Lightning card reviewed a few days ago:
As can be seen in the code 1015A3, the chip is A3 silicon fabbed in week 15 of 2010. Week 15 = April.
GF100 die as seen on Wizzard's review of the GTX480 on release day:
Week 9.
GTX460 as seen on Wizzard's review at launch day:
Week 22
Only 7-13 weeks difference. 2-3 months on production chips, GF104 test samples should have been on Nvidia's hands some weeks earlier. It should have been pretty obvious for Nvidia pretty soon that GF104 was so much better and the fact that it seems there's no newly fabbed GF100 since then (week 15) suggests that Nvidia didn't expect GF100 to last too much on the market once they saw GF104. The fact that the GTX460 was so seriously crippled in order to be able to clear GF100 inventories fast, also suggests they had something coming for which they wanted to make room for IMO. If they have no more GF100 (as can be deduced by the fact that even cards released now are remains from so long ago) they should have something to fill that segment up. I don't think they wanted to be months without anything that can even remotedly compete with Cayman, and GF100 could at least compete/be close performance wise.
Just my 2 cents.
but it think nvdia will use AMD trick, and after seeing how well the GTX460 oc i can certainly sure they will be releasing full fledged GTX 460 with significantly higher clock to minimize HD 6970 damage
Wow, the drama.
They're still awesome cards(performance wise)...
and btw i think nvdia make almost no profit for GTX 470, just think about it even the GTX 480 still sell for $400 and they have almost same PCB design and same GPU,
:rolleyes:
Now, I know what you'll say, that the GF104 in GTX 460 is not "fully-enabled", so let's examine that statment shall we; does anybody really expect a "fully-enabled" GF104 to equally perform as two of the standard GF104 used in GTX460?! :eek: That would be stupid, right? The fact is that you've made a lot of illogical speculations about the upcoming "GTX 475" without substantiating them with any benchmarks or evidence, thus sounding like a true nVidia fanboy Benetanegia. Let's be mature here and not argue over pure speculations, or it's just a waste of time.
I'm sorry to be the one breaking the news to you; but the logical conclusion is that even the "fully-enabled", "properly-clocked", and "properly-priced" GF104 you are hoping for can never touch Cayman. :)
That said; I truly hope that nVidia will come up with a pleasant surprise soon in their GTX 580 or whatever it is called, otherwise it is definitely Cayman for me. I am going to build a new gaming rig in December/January and I'll buy the best GPU that I can get for $425 at that time, regardless if it was nVidia or AMD/ATI. :rockout:
Edit: ... just one more thought came to mind: Benetanegia called Cayman an overkill, but I wonder if he said the same about the GTX 480 when it was launched for $500+ as any objective person would do ... I think not, the bias towards nVidia is clear in every word he writes ...
My (let's call it) "assumption" (although you'll see how it's not) of fully enabled GF104 performance is based on hard facts on the other hand. A GTX460 @ 820 Mhz is as fast as a GTX470 and hence also HD6870:
And with 15% more shaders/TMU/tesselators/... enabled it would be almost 15% faster because Fermi scales that way (well) based on shading performance. You don't believe me?
Let's see what the GFlops are for the Fermi lineup:
GT430 = 268.8 GFlops
GTS450 = 601.34 GFlops
GTX460 = 907.2 GFlops
GTX470 = 1088.64 GFlops
GTX480 = 1344.96 GFlops
Now let's normalize those numbers so that a GTX460 represents 85% just like in the chart above and see if there's a relation. What I'm doing is if 907 Gflops = 85% then 268.8 = 85% *268.8/907.2 = 25.18%. OK let's do it for all the cards listed above:
GT430 = 25.18% ---------> 27% on the chart
GTS450 = 56,34 %--------> 55% on the chart
GTX460 = 85% -----------> 85% on the chart obviously
GTX470 = 102% ----------> 104% on the chart
GTX480 = 126% ----------> 128% on the chart
The conclusion is no other than Fermi scales linearly with GFlops. And what would be the GFlops for the hypothetic GTX475?
384 SPs * 800/850 Mhz *2 (shader clock) *2 (FMADD) = 1228.8/1305.6 Gflops
And normalized:
GTX475 = 115/122% +/- 2%
So now that we both DO know what would be the performance of the supposed GTX475, let's explain what I meant.
If Nvidia releases that card that is within a hair of GTX480 performance for $250 anything that would sell above that price would look simply overkill/overpriced for almost anyone except enthusiasts and Cayman XT will most probably sell for more than $400. Also bear in mind that such card would cost Nvidia almost the same to make as GTX460 1GB does so selling them at $250 would be a relief rather than a curse.
Oh and BTW for most people's needs anything above a GTX460 or HD5850 is overkill. For most people, it was overkill selling at $500 and even now is still overkill (for most people) selling for $400.
1) I don't think that it is right to decide for other people what they need or don't need. How can you say that anything above the HD 5850/GTX 460 is an overkill; that means that the huge crowd who bought the highly successful 5870 are idiots, which is definitely not the case. It's a free world so just let everyone select the GPU that suits him/her best, and spare us your personal judgments.
2) If nVidia is going to release this GTX 475 or whatever it is called any time soon, and knowing that this GPU will perform as good as the GTX 480 yet selling at $250, then they would definitely inform the GPU manufacturers ahead to take certain measures to deplete their stock of GTX 480 & GTX 470 GPUs (through discounts, bundles, ... etc) since after such a GPU hits the market at that price point; they'll never be able to move the GF100 GPUs, right??? If this was the case then how come MSI has just released the N480GTX Lightning GPU and Point of View have just announced the TGT GTX 480 Beast GPU? It simply doesn't make sense. Yes, I do believe that nVidia is going to release a fully enabled GF104 chip GPU soon according to many rumors, but the performance will only compete with the 6870 and not the GTX480, thus targeting the gamers' sweet spot as they call it, but this won't be enough for enthusiasts. In fact now that I see that ZOTAC are going to soon release the GTX 460 X2, I become more and more convinced that nVidia are not going to release the long-awaited GTX 495 (dual GF104 chip GPU), otherwise ZOTAC wouldn't have taken this initiative on their own.
3) Forget about possibly fake benchmarks of the 6970 or whatever, just listen to simple reasoning. AMD already have the 5870 which is a very successful GPU performing just behind the GTX 480, and noting that it has been over a year since they released the 5870 so they have had enough time to develop the 6970, then just tell me by what logic would AMD produce this new GPU if it wouldn't beat the GTX 480? Simply not possible. Enthusiasts and Extreme Gamers (who are targeted by the Cayman XT & Pro GPUs) are always looking for what is faster and better, and that's where the 6970 fit in. Even if it was only 10-15% better than the GTX 480, but achieving that at a reasonable TDP and power efficiency, then it'll definitely be a winner, given they don't price it stupidly.
2) GTX480 performance from this GTX475 is not a certainty and neither is a necessity. It just needs to offer enough performance to play nearly all games maxed out. A GTX460/HD68xx/HD5850 already does that according to what maxed out means for the grand mayority. Anything faster is going to be welcome as long as it's priced really well, but very few are going to rush to the store to buy $400+ cards, no matter what performance it offers, because it's simply not needed, and much less if there's something for $250 that suits their needs just as well.
As to why partners are releasing those cards, because that's the way they are going to get rid of GF100. I could ask you why they released the GTX465 just a month prior to GF104 if they knew it was going to kick GTX465 in the butt. Because they had to sell them.
As for Nvidia, they've done everything on their hand to sell those GF100, including releasing the GTX460 as it is instead of releasing the full chip, in order to not steal GF100 sales. The explanation of bad yields is false. Yields work in two directions: 1) defects that render some parts unusable 2) defects that limit maximum attainable clocks on certain areas. Both always go hand in hand, they never get a lot of one type but none of the other. The situation we are seing is literally imposible: not enough chips to be able to release a 384 SP SKU from day one and every single GTX460 being able to hit 850 mhz at the same time is just not posible. If yields were bad clocks wouldn't be good either and even if that would be posible, it still does not explain why the GTX460 was not released at 750-850 Mhz to begin with. The only xplanation is that GF100 had to sell. And why did GF100 have to sell? Because something better is coming to fill that gap.
3- Yes, 15-20% faster than GTX480 is posible, and maybe some more, but those who think that Cayman is going to be twice as fast as Barts are just living in a pipe dream. I mean, I'll never say it's imposible, but it's very very very unlikely. For instance Barts is not a real improvement over Cypress/Juniper. It offers close performance while being smaller mostly because it lacks dual precision (FP64) support, just Like Juniper did. The other reason is that is very well known that Cypress had some scaling issues with more than 1440 SP. A HD5850 running at same clocks is just as fast as HD5870 and there's no way to know if the same would happen with HD5830 if it weren't for the crippled ROP count. 16 ROPs was a necessity or a trick to not expose the scaling problem? We'll never know. Barts hits a good spot and that's all. There's no way to know at this point if Cayman will scale any better than Cypress, and it needs to scale much much better if they want to make a real difference.
Another issue with 2x Barts Cayman is size. Barts while smaller than Cypress is 255 mm^2, but it lacks FP64 support. Twice that and you are at 500 mm^2, bring in some optimizations and you go lower than that, but then you have to bring FP64 support in and size goes up again. We are still on 40nm process and AMD cannot do magic as some people here think. If they could do magic they would have used it on Barts and Barts really isn't anything special except for the fact that it was designed for the best selling market and that it was designed with just enough SPs to perform well, but not too many as to meet the efficiency "wall".
EDIT: Things about Barts that make me be almost 100% sure that Cayman is not going to be massively faster than Cypress:
From Anandtech: It hints as some shader innefficiency as well as need for higher ROP power. Not really good for Cayman as it could mean that Cayman needs more than 32 ROPs in order to be much faster (like 2x) than Barts, which again makes it too big. This means again that Cayman is bigger than Barts for the performance it'd offer. Either 512 bits are used in Cayman along with slow memory, making it big, or the improved Cypress memory controler is used again (or even a bigger one to support 6 gbps memory), which makes it bigger too. Either way bigger.
and btw majority of people still with onboard graphic just take a look at intel market share, so with your logic, we won't even need discrete graphic card because majority of people just enough with onboard graphic