GF 104 is 366mm2

[Cookie Monster]

GF104 has only 15.5million cards to go to catch up to Cypress.
Let me know how that works out for you and the rest back at Headquarters.^_^

That figure includes the entire evergreen family.

I think SI will be fine, it's more of a question of how aggressive AMD wants to be with their GPU size than what Nvidia will have at that time, I think. Remember, AMD has 200mm2 of die size they can tack on before they reach Fermi's GPU size. So if AMD does make a 430mm2 part (just pulling a number out of my ass) there shouldn't be any reason they can't 'tack on more stuff' to reach Fermi's performance and still have the smaller GPU. Just like the 5870 vs. the GTX460, when we are talking about GPU's of a near equal size AMD just has an advantage right now. With a slightly revised and tweaked SI part and more silicon I'm sure they can match Fermi.

With the 4890 they made a different chip, but it was really the same architecture and they kept the part very close to the same size. Unless AMD can unlock a lot of performance through architecture enhancments, I see them adding more silicon to make sure they are able to out perform Fermi.

But then, with 40nm still being the only process for a while yet that makes sense for high end GPU's, where does this leave Nvidia to turn to when SI is out? An even bigger GPU? A 4890 type of part of their own?

Interesting times coming up.

The thing is that the architecture AMD has been using slowly reaching its near end (although its had a pretty good run). They are showing too many bottlenecks as seen by the lack performance increases when adding extra logic units, bandwidth or even increasing the core clock frequency (the performance increases are not linear). Sure they could just "tack on" stuff as you say but as said before, the returns are slowly being diminished. Not only that but its more of a short term solution which isn't preferable especially in this industry. However that being said, im beginning to wonder if those transistors in the original cypress that didnt make the cut for the one being shipped now is going to be retacked on.

The real threat lies in Fermi 2 and 28nm. Because of the rather huge plunge nVIDIA took when they revamped their architecture, they can start to optimize for perf/mm^2 since all the hard work has already been covered. All they need to do is die shrink, add extra logic units with the appropriate amount of clock speeds to reach their desired performance goals since the performance returns are quite linear atm.

 

[Wreckage]

Despite costing around $200 a GTX460 can compete with a 5870 in newer DX11 games. While also supporting 3D and physics something that smaller gpu has been lacking for a while now.

Not that anyone really cares (or even knows about) the mm2 size of their GPU.

I would be more concerned about my video card supporting simple features like AA.

 

[HurleyBird]

The thing is that the architecture AMD has been using slowly reaching its near end (although its had a pretty good run). They are showing too many bottlenecks as seen by the lack performance increases when adding extra logic units, bandwidth or even increasing the core clock frequency (the performance increases are not linear). Sure they could just "tack on" stuff as you say but as said before, the returns are slowly being diminished. Not only that but its more of a short term solution which isn't preferable especially in this industry. However that being said, im beginning to wonder if those transistors in the original cypress that didnt make the cut for the one being shipped now is going to be retacked on.

Well, to be fair Fermi has quite a bit less perf/shader in most games than G200b also. Yes, Fermi does scale a bit better with added clock speed, but that could all be down to a single bus inside the chip being a bottleneck, it doesn't mean that the entire architecture is broken.

No idea why people seem to be falling into the fallacy that newer has to equal more advanced. AMD's architecture has better perf/watt, perf/transistor, and transistors/mm2 than Nvidia's. In almost every benchmark that I can think of, AMD is ahead of Nvidia architecturally by about a generation (being on the same node plus the fact that Nvidia makes giant chips is what evens the score). The exception is of course tessellation, but that may have more to do with the fact that AMD decided to trim down Cypress mid-way through the project. SI will show whether or not this is the case.

 

[ronnn]

So it looks like the ati idea of a small efficient upper mid range gpu has caught on. Nvidia pr and fans are now promoting their cheap upper mid range bargain basement gpu as the new champ.

Price war works for me, but as said repeatedly last year - probably doesn't work for amd or nvidia.

 

[evolucion8]

Despite costing around $200 a GTX460 can compete with a 5870 in newer DX11 games. While also supporting 3D and physics something that smaller gpu has been lacking for a while now.

Not that anyone really cares (or even knows about) the mm2 size of their GPU.

I would be more concerned about my video card supporting simple features like AA.

MWHAHAHAHAHA, funny. Anandtech's review of the GTX 460 states the opposite.

Source: http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king/19

"They still have the upper-hand at performance-per-watt, and with just how similar the GTX 460 and the Radeon 5850 are in terms of die size and power consumption there’s clearly some flexibility on their part to change things. The Radeon 5830 must come down in price or go away entirely, it’s what happens to the 5850 that’s the question. We’ve seen the GTX 460 lock horns with the 5850, and while the 5850 is undoubtedly the faster gaming card the $300 price point no longer makes as much sense as it once did with a $230 1GB GTX 460 below it. AMD either needs a 5840, or a price drop on the 5850 to bring its price more in line with its performance."

3D and PhysX that's all I can say about it. PhysX will die due to the lack of quality tittles supporting it (Batman AA is awesome, Cryostasis sucks, Darkvoid, Oh my God with those cheesy phisics effects, D: ) and the terrible implementation of PhysX in most other games including Mirrors Edge (Crappy window shattering), GRAW, GRAW 2 D: patthetic particle effect, Cryostasis again, the water looks like it was mixed with milk and lots of poltergeist like things would happen in many rooms, creeeeppy and buggy as hell. Understand that developers takes nVidia's money and implements cheap physX effects for the money, not for a favor or for the love of it, and 3D, well, will be the future, but not yet, I will not buy a TN based LCD for the sake of 3D.

Lack of AA support, you are right, but your baiting is too childish to fall onto. Singularity doesn't support AA currently, but tools like RadeonPRO works like a champ, I'm able to use AA and near 100percent CF scaling with such little tool. AMD might be a little behind in terms of driver quality and support, but not even in your best dreams, a GTX 460 can match the faster HD 5870. Keep dreaming....

 

[happy medium]

The performance difference between the 5850 and gtx 460 1gb @ 1900x1080 was 8 percent favoring the 5850. As per the Anandtech review.

Even though Nvidia's drivers are steadily improving the gtx line, I don't see the gtx 460 competing with a 5870 unless overclocked over 900 core.

On the other hand, with more driver improvements I do see it catching the 5850.

I've been reading the driver comparisons over at Alienbabletech and there really hasn't been much real improvment since the 10.3's for single ATI cards. Nvidia on the other hand seems to be showing real improvements every month.

When I say "REAL improvements", I mean not what ATI/Nvidia says ,but what the driver reviews are showing.
That explains why the gtx 470 is so close in performance to the 5870 now.

 

[Scali]

But then, with 40nm still being the only process for a while yet that makes sense for high end GPU's, where does this leave Nvidia to turn to when SI is out? An even bigger GPU?

If anything, GF104 shows that nVidia knows how to get more performance out of the Fermi architecture without having to pile on extra transistors.
If all else fails, they could do a refresh of the architecture, bringing some of the new GF104 improvements to the 'full size' Fermi. But under normal circumstances, I would expect nVidia to wait with that until 28 nm is available.

 

[Scali]

Nvidia on the other hand seems to be showing real improvements every month.

I think especially in Vista/Windows 7 there are some improvements left on the table for GF100/GF104.
Some of my D3D/OpenGL code can run at insane framerates in Windows XP on my GTX460 (think 12500 fps)... but much slower in Vista/Windows 7 (more like 5000ish).

I know from earlier times with a 9800GTX card, I got around 8000-9000ish fps in Vista/Windows 7 aswell, just like it did in XP (although initially those cards also suffered from quite a performance penalty in Vista/Win7 compared to XP).

So I think there is some CPU overhead there that nVidia has yet to cure.

 

[Cookie Monster]

If anything, GF104 shows that nVidia knows how to get more performance out of the Fermi architecture without having to pile on extra transistors.
If all else fails, they could do a refresh of the architecture, bringing some of the new GF104 improvements to the 'full size' Fermi. But under normal circumstances, I would expect nVidia to wait with that until 28 nm is available.

I agree. There was a rumour of a possible Bx revision for GF100 but that really depends on what SI/NI will turn out to be.

But for nVIDIA it will probably be better for them to work on a 28nm chip which could potentially be a 768SP/128TMU chip (similiar spec to a dual full fledged GF104).

 

[AtenRa]

28nm graphics cards will not be released sooner than Q3 2011 so I don’t think NV will wait so long and with SI around the corner in Q4 2010, NV probably will release a B stepping GF100 or equivalent in late Q4 2010.

 

[GaiaHunter]

But that's not the point, is it?
My point is that the GTX465 has pretty much the same price, performance and features as the GTX460, yet the GTX460 is a better choice.
The GTX465 was NOT good enough to make me buy one and get rid of that flaky Radeon, but the GTX460 takes power consumption out of the equation, so it's the perfect card for me at this point.

But I'm talking about people that buy GPUs to play games.

You bought that GTX460 because you wanted a card that has cuda and other stuff and the GTX 460 is the first DX11 card with cuda that isn't a power consumption monster.

So in fact even if a 5870 was the same price of a GTX 460 you would have bought a GTX 460 while a gamer would buy a 5870 without thinking twice.

So you are talking of a very niche situation while I was talking generally.

 

[Keysplayr]

But I'm talking about people that buy GPUs to play games.

You bought that GTX460 because you wanted a card that has cuda and other stuff and the GTX 460 is the first DX11 card with cuda that isn't a power consumption monster.

So in fact even if a 5870 was the same price of a GTX 460 you would have bought a GTX 460 while a gamer would buy a 5870 without thinking twice.

So you are talking of a very niche situation while I was talking generally.

If that were the case, a 5870 and GTX460 being the same price and from a gamers perspective, I'd agree with you.

 

[Scali]

You bought that GTX460 because you wanted a card that has cuda and other stuff and the GTX 460 is the first DX11 card with cuda that isn't a power consumption monster.

Actually, I wanted a card that has OpenCL.
But even for games I would recommend the GTX460 to anyone.

 

[SlowSpyder]

That figure includes the entire evergreen family.



The thing is that the architecture AMD has been using slowly reaching its near end (although its had a pretty good run). They are showing too many bottlenecks as seen by the lack performance increases when adding extra logic units, bandwidth or even increasing the core clock frequency (the performance increases are not linear). Sure they could just "tack on" stuff as you say but as said before, the returns are slowly being diminished. Not only that but its more of a short term solution which isn't preferable especially in this industry. However that being said, im beginning to wonder if those transistors in the original cypress that didnt make the cut for the one being shipped now is going to be retacked on.

The real threat lies in Fermi 2 and 28nm. Because of the rather huge plunge nVIDIA took when they revamped their architecture, they can start to optimize for perf/mm^2 since all the hard work has already been covered. All they need to do is die shrink, add extra logic units with the appropriate amount of clock speeds to reach their desired performance goals since the performance returns are quite linear atm.

Right, AMD's current architecture as it sits appears to be running out of steam, just adding more 'stuff' won't get them the performance they are looking for. But SI is said to be, at least some parts, a rework. If AMD does make the larger GPU, and keep clock speeds up, they will likely use more power, but they have a big cushion between a 5870's power use and a GTX480's power use. We don't know what SI will bring to the table, so it's hard to say how efficient it will be or what parts will be reworked. But, as it sits Nvidia is pretty much out of room to add more power use or more logic, save for the 32 unused shaders that are already built in.

28nm will be a whole different story. Fermi might be a shrink, it might be a shrink with more 'stuff' added, we don't know. And AMD might bring SI2 or NI, and we have no idea what to expect. But my comments were aimed at the current 40nm process and future parts that will come from that process.

 

[evolucion8]

I think especially in Vista/Windows 7 there are some improvements left on the table for GF100/GF104.
Some of my D3D/OpenGL code can run at insane framerates in Windows XP on my GTX460 (think 12500 fps)... but much slower in Vista/Windows 7 (more like 5000ish).

I know from earlier times with a 9800GTX card, I got around 8000-9000ish fps in Vista/Windows 7 aswell, just like it did in XP (although initially those cards also suffered from quite a performance penalty in Vista/Win7 compared to XP).

So I think there is some CPU overhead there that nVidia has yet to cure.

That might explain as well the great CPU dependence that the GTX 4 series has which has been showed in reviews.

 

[Wreckage]

but not even in your best dreams, a GTX 460 can match the faster HD 5870. Keep dreaming....

http://www.xbitlabs.com/articles/video/display/hardware-tesselation_6.html#sect0

Dreams do come true!

 

[evolucion8]

http://www.xbitlabs.com/articles/video/display/hardware-tesselation_6.html#sect0

Dreams do come true!

Please, cherry picking benchmarks, I can do the same!!

http://www.xbitlabs.com/articles/video/display/hardware-tesselation_8.html

Look at the beating that the HD 5870 is giving to the GTX 460, in a far more demanding game than Allien vs Predator.

http://www.hardwarecanucks.com/foru...orce-gtx-460-1gb-gtx-460-768mb-review-19.html

The GTX 460 1GB meanwhile is actually able to go toe to toe with the $299 HD 5850 on a number of occasions and simply blows the HD 5830 out of the water when image quality settings are increased. The only stumbling block seems to be 2560 x 1600 but the GTX 460 really isn’t meant for the crowd that can afford 30” monitors anyways.

http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king/19

We’ve seen the GTX 460 lock horns with the 5850, and while the 5850 is undoubtedly the faster gaming card the $300 price point no longer makes as much sense as it once did with a $230 1GB GTX 460 below it. AMD either needs a 5840, or a price drop on the 5850 to bring its price more in line with its performance.

http://www.hardocp.com/article/2010/07/12/nvidia_geforce_gtx_460_review/10

This is a truly fabulous video card. Its performance either matches or outpaces the competition from AMD. It consumes relatively little power, especially under full load. Its fan keeps it cool and quiet. (They were comparing it against the HD 5830!!!, and it outperformed it by a slim margin)

http://www.techpowerup.com/reviews/Zotac/GeForce_GTX_460_1_GB/31.html

http://www.xbitlabs.com/articles/video/display/nvidia-geforce-gtx-460_16.html#sect0

Resolution: 1600x900 - As for the competition against the Radeon HD 5850, the new card loses eight out of the 18 tests but wins another eight while being much cheaper.

Resolution: 1920x1080 - If compared to the Radeon HD 5850, the GeForce GTX 460 has 6 wins, 11 losses and 1 draw, which is good considering the difference in pricing.

Resolution: The resolution of 2560x1600 is not really meant for graphics cards priced at $200, yet the GeForce GTX 460 1GB has good results, outperforming the GeForce GTX 465 as well as the Radeon HD 5830 although the average advantage over these cards lowers to 4 and 19%, respectively. It is harder to compete with the Radeon HD 5850 at that resolution. The latter has better texture-mapping resources, so the new card is an average 9% slower.

Wreckage: You are welcome to prove me wrong, but since you came back from your ban I expected from you a change and being more neutral or if you favor a GPU vendor, at least admit their rights and wrongs, but you're hopeless....

 

[SlowSpyder]

evolucion8, don't fall into the trap. He knows what he's donig, it's being done on purpose. ABT has an article where a Radeon 4850 over takes a GTX260 216 in some benches, I was was going to post that, but thought better of it. He's just trolling, no point in feeding him.

The 460 is a fantastic card at it's price, but any of us (even fanboys who can display any sort of logic) know it's a fantastic card that fits between the 5830 and 5850. It's no more a 5870 slayer than a 5850 is a GTX480 slayer. Is it a 'good' card for Nvidia's bottom line when AMD's smaller and more mature GPU is faster and selling for significantly more? I wouldn't think so, but who knows.

 

[Wreckage]

Please, cherry picking benchmarks, I can do the same!!

Read the whole review. I just picked one as a clear illustration.

 

[evolucion8]

evolucion8, don't fall into the trap. He knows what he's donig, it's being done on purpose. ABT has an article where a Radeon 4850 over takes a GTX260 216 in some benches, I was was going to post that, but thought better of it. He's just trolling, no point in feeding him.

The 460 is a fantastic card at it's price, but any of us (even fanboys who can display any sort of logic) know it's a fantastic card that fits between the 5830 and 5850. It's no more a 5870 slayer than a 5850 is a GTX480 slayer. Is it a 'good' card for Nvidia's bottom line when AMD's smaller and more mature GPU is faster and selling for significantly more? I wouldn't think so, but who knows.

You are right, is just that I hate trolls getting away and nobody do nothing about it.

 

[Wreckage]

You are right, is just that I hate trolls getting away and nobody do nothing about it.

So I guess you did not read the whole review? Just call me names and hope that will change the facts. Out of 6 benchmarks the 5870 wins 2 breaks even on 1 and loses 3.

So unless you are calling Xbit a troll site, maybe something should be done about your insults.

 

[Grooveriding]

Comprehensive review of the 460

http://www.anandtech.com/show/3809/nvidias-geforce-gtx-460-the-200-king/1

It comes nowhere near a 5870, stop feeding this troll

 

[Skurge]

You are right, is just that I hate trolls getting away and nobody do nothing about it.

We've all tried doing something, but he keeps posting his nonsense without being banned or even warned. I've just given up.

 

[Wreckage]

So no comment on the Xbit review then? Just going to ignore it and call me names?

As I stated "in newer DX11 games". This is what we are discussing.

 

[happy medium]

a GTX460 can compete with a 5870 in newer DX11 games.

Post # 57

And he backed it up with a link ,after evolucion8 said "

AMD might be a little behind in terms of driver quality and support, but not even in your best dreams, a GTX 460 can match the faster HD 5870. Keep dreaming....

Overall, I think we all know a gtx 460 cannot match a 5870. But NEVER match it, yes it can.