AMD: R8xx Speculation
- Thread starter Shtal
- Start date
And what if Nvidias card can battle the X2, but there's nothing below it? Just GT300 for two or three quarters? Welcome back healthy margins!
@flopper:
25 watts seem really good - GTX28x does just above 30, hd4870/90 around 60 and some partner 4850s 20 watts right now and i think that's (20-30) about a fair amount for high performance cards. Cooling should possibly be almost silent but of course depending on cooling need of other components than the GPU also.
http://www.fudzilla.com/content/view/15436/1/
Fudo is on a roll at the moment and so far the picture looks outstanding IMHO, apart from the rather high estimated (?) load power consumption.
And yes I know there has been another thread about the dreaded past 1.6x from Fudo, but here's the explanation what he originally meant:
And no it's a bad average persentage at all; au contraire. The "up to" rates will obviously be a lot higher than that.
Fudo is on a roll at the moment and so far the picture looks outstanding IMHO, apart from the rather high estimated (?) load power consumption.
And yes I know there has been another thread about the dreaded past 1.6x from Fudo, but here's the explanation what he originally meant:
And no it's a bad average persentage at all; au contraire. The "up to" rates will obviously be a lot higher than that.
It's more than twice the texel and pixel fillrates also; there must be a damn good reason why they've opted for 32 pixels/clock and the only other thing I can think of as a layman is a healthy efficiency increase in the rasterizing part of the chip.
That silly arithmetic/FLOP craze everyone is in should come slowly to an end; I can see other parts of a GPU scaling at equivalent ratios and there's no such thing as good enough fillrate or bandwidth. The latter seems to get increasingly critical. Those that thought that arithmetic is be all end all part of the GPU future equasion might want to re-estimate.
DegustatoR
Legend
The fan is silent in idle even on 4870X2.
190W however is more than 4890 and leads to ~360W 5870X2 (assuming it'll have the same frequenies) which is 100W more than 4870X2.
So it's fan noise under load what you should be cautious of.
And what if Nvidias card can battle the X2, but there's nothing below it? Just GT300 for two or three quarters? Welcome back healthy margins!
If they don't radically change the architecture, it's almost impossible. ATi's current architecture is way more efficient in die-size terms than Nvidia's. To be able to beat an HD5870X2, Nvidia has to produce a chip able to deliver over 3+ Tflops (let's suppose a 50% scaling from HD5870 to HD5870X2). That's almost 3 times the spec of GT200 with the current architecture.
And still, as you said, even if they manage to beat the HD5870X2 with a enormous, 600+ mm^2 chip, they are not going to have anything to beat the 5850, that will cause lot of pain... i mean, with those specs, a 5850 should perform greatly even at FullHD resolutions, 50% than a HD4890 for 299$/250 euros
Nvidia must change strategy with GT300 or they will keep losing market share.
Nice
If they don't radically change the architecture, it's almost impossible. ATi's current architecture is way more efficient in die-size terms than Nvidia's. To be able to beat an HD5870X2, Nvidia has to produce a chip able to deliver over 3+ Tflops (let's suppose a 50% scaling from HD5870 to HD5870X2). That's almost 3 times the spec of GT200 with the current architecture.
And still, as you said, even if they manage to beat the HD5870X2 with a enormous, 600+ mm^2 chip, they are not going to have anything to beat the 5850, that will cause lot of pain... i mean, with those specs, a 5850 should perform greatly even at FullHD resolutions, 50% than a HD4890 for 299$/250 euros
Nvidia must change strategy with GT300 or they will keep losing market share.
I agree
Because FLOPs by themselves define any GPUs performance? Without knowing what NV's X11 architecture exactly looks like as a whole that kind of speculative math is so useless that it's not even funny anymore.
And still, as you said, even if they manage to beat the HD5870X2 with a enormous, 600+ mm^2 chip, they are not going to have anything to beat the 5850, that will cause lot of pain... i mean, with those specs, a 5850 should perform greatly even at FullHD resolutions, 50% than a HD4890 for 299$/250 euros
Nvidia must change strategy with GT300 or they will keep losing market share.
D12U sizes now over 600 square millimeters? Do you know what the hell you are talking about?
AMD is releasing a fine bunch of GPUs with the 5850 and the 5870. What will come after that let's just care about that whenever it arrives. Whoever feels needs an upgrade today will have two new very good GPUs to pick from.
I remember long, long time ago I was playing with guess numbers about R800. here http://forum.beyond3d.com/showthread.php?p=1246364#post1246364
But the truth turns out here: http://www.fudzilla.com/content/view/15436/1/
I was correct on the 32 ROP's and 1600 shaders, and almost the clock frequency. I did said 512bit memory controller, but I was thinking back then for dual chip, but for single chip I was correct as well 256bit, but I did not post for that spec for single chip.
From Shtal
But the truth turns out here: http://www.fudzilla.com/content/view/15436/1/
I was correct on the 32 ROP's and 1600 shaders, and almost the clock frequency. I did said 512bit memory controller, but I was thinking back then for dual chip, but for single chip I was correct as well 256bit, but I did not post for that spec for single chip.
From Shtal
Yeah, nice
How does a 4890 perform on a similar system?
20fps?
http://hartware.de/review_892_11.html
Crysis (very high) – FSAA 4x / AF 8x
Intel Core 2 Duo 3.33 GHz, 2 GByte, Intel X38, Windows Vista
last table:
GeForce GTX 295 - 34fps
Radeon HD 4870 X2 - 31fps
GeForce GTX 285 - 23fps
GeForce GTX 280 - 21fps
GeForce GTX 260 - 17fps
Radeon HD 4870 - 17fps
Come on, don't get mad!
What i meant was that Nvidia has to change strategy: they cannot use anymore the strategies they used for G80 and GT200 (high-end parts months before the mass market parts). But that's my opinion, you can have a different one.
If they don't radically change the architecture, it's almost impossible. ATi's current architecture is way more efficient in die-size terms than Nvidia's. To be able to beat an HD5870X2, Nvidia has to produce a chip able to deliver over 3+ Tflops (let's suppose a 50% scaling from HD5870 to HD5870X2). That's almost 3 times the spec of GT200 with the current architecture.
And still, as you said, even if they manage to beat the HD5870X2 with a enormous, 600+ mm^2 chip, they are not going to have anything to beat the 5850, that will cause lot of pain... i mean, with those specs, a 5850 should perform greatly even at FullHD resolutions, 50% than a HD4890 for 299$/250 euros
Nvidia must change strategy with GT300 or they will keep losing market share.
Minor nitpick.
5870 delivers, 2.64Tflops. So 5870X2 will give
scaling | flops
50% | 3.96T
70% | 4.49T
80% | 4.75T
They need a 4.5Tflop gt300 to stay alive.
Assuming 512alu * 2(fma) *1.5Ghz = 1.5Tflop, which is less than what 5850 (assunimg 825 Mhz clock for it too) delivers.
To get there, I can see the following scenarios (in the attached image, flops are in gflops). Considering that gtx 285 runs at 1.5Ghz, I cant see (ie most probable event) gt300 beating 5850 in peak flops, let alone 5870 or it's x2 cousins.
Flops arent everything, I agree, but amd will certainly balanced their architecture as much as they could. Considering the workloads are so math heavy these days, I am not sure if amd will want to have done evergreen differently after they see gt300.
Their best bet is to deliver a rv770 like increase in efficiency. It's possible certainly.
Or they can sell gtx295's at 300 a pop.
Attachments
my 2cents:
-loved the idle operations....that is where most spend their time at...AMD have learned much to keep a GPU cool and quiet.
-seems a long card...for 40nm, 1GB framebuffer...the exhaust area is cut by half to placed the 3 display out...will it affect heat dissipation..?
-Crysis fullHD @ 4xAA @ 30fps...nice! and it is only on a PII 955...interestingly AMD is in a catch22 situation..i am sure Core i7 will give higher frames on newer game engines...do AMD advertise for Intel or bite the bullet and lose some frames with PII...lolz
-what is SFR? This is the new dual card renderer is it...from chiphell comments, this SFR is more optimized for dual cards..
-so 5850 is a 5870 with disabled parts...ACC anyone?
-loved the idle operations....that is where most spend their time at...AMD have learned much to keep a GPU cool and quiet.
-seems a long card...for 40nm, 1GB framebuffer...the exhaust area is cut by half to placed the 3 display out...will it affect heat dissipation..?
-Crysis fullHD @ 4xAA @ 30fps...nice! and it is only on a PII 955...interestingly AMD is in a catch22 situation..i am sure Core i7 will give higher frames on newer game engines...do AMD advertise for Intel or bite the bullet and lose some frames with PII...lolz
-what is SFR? This is the new dual card renderer is it...from chiphell comments, this SFR is more optimized for dual cards..
-so 5850 is a 5870 with disabled parts...ACC anyone?
Ok now go back and compare the FLOP rates of a GTX295 vs a 4870X2 and do the speculative math as many times until you finally realize that FLOP != FLOP between those two architectures.
Come on, don't get mad!
What i meant was that Nvidia has to change strategy: they cannot use anymore the strategies they used for G80 and GT200 (high-end parts months before the mass market parts). But that's my opinion, you can have a different one.
And you know that they are still using that strategy? I doubt that GT212 got cancelled without deciding to bring the D12 Performance part earlier then planed.
Well, I don't know why you say this and I have no idea what you mean by this, so until you can make your point more precisely/verbosely, I will have to stick to my notions (even if they are ill formed) for now.
Similar threads
