ATI Q&A - Confirm 2 Variants Of 4870X2

[ten80p]

I'm assuming they'll be 1Gb and 2Gb ?

Some comments on PCI-E 8x and 16x too

http://www.techtree.com/India/News/EXCLUSIVE_--_An_Interview_with_AMD/551-90984-579.html

 

[Final8y]

HOLY-COW

AMD has already built a computer that has four 4870X2s in it. So it has eight GPUs; drivers will not be supporting eight GPUs at this point of time.

 

[slap ed]

nice read

 

[Skyboat]

HOLY-COW

Probably wanted to use it for Folding

 

[Talrinys]

Probably wanted to use it for Folding

Would be crappy compared to an Nvidia rig for folding, unfortunately

 

[Skyboat]

Only because the F@H client hasn't been updated to use all 800 shaders instead of 320. Once it is though

 

[Talrinys]

Only because the F@H client hasn't been updated to use all 800 shaders instead of 320. Once it is though

There aren't 800 shaders, there are 160, it cannot use them effectively enough And the developers have said many times that they aren't sure that they'll be able to utilize them well enough. Still hoping for it though, would buy the x2 instantly

 

[Mr Paul]

There aren't 800 shaders, there are 160, it cannot use them effectively enough And the developers have said many times that they aren't sure that they'll be able to utilize them well enough. Still hoping for it though, would buy the x2 instantly

No, theres 800 shaders.

nVidia and ATI shaders are totally different and cannot be compared.

 

[Biffa]

There aren't 800 shaders, there are 160, it cannot use them effectively enough And the developers have said many times that they aren't sure that they'll be able to utilize them well enough. Still hoping for it though, would buy the x2 instantly

Actually the developers have said that they are going to get substantially lower CPU useage and be able to at least match the NVidia core scores.

Not at all. The current kernels as written are being optimized well, but that doesn't mean they are written optimally for the newer chips or even 2XXX/3XXX. There is also some overall program structure issues and how all the functions are called and linked together that needs to be reworked to better match the newer hardware. Even the "low hanging fruit" optimizations may yield good performance gains, we shall see in the next few core updates. We will give Scott and Nvidia a run for their money, but optimization work in the face of stability takes time.

v1.04 did have improvements in CPU load, but there was concern over an instability that we believe we have tracked down and we have updated versions, with even further reductions in CPU load, in testing.

 

[Talrinys]

Actually the developers have said that they are going to get substantially lower CPU useage and be able to at least match the NVidia core scores.

Oh yeah i do remember that quote, sorry about that, was just thinking about the fact that Pande always seems to prefer Nvidia GPUs.

No, theres 800 shaders.

nVidia and ATI shaders are totally different and cannot be compared.

No there are 160 physical shaders on the card, they are awesome at games and can do 800 operations but that doesn't mean a thing, the other thing is just false marketing, something AMD has done for years now. I always hated the fact that they labelled their chips 2800+ or whatever if they weren't at that clock. All hail efficiency and i love AMD as a company but their marketing departments are great at spinning stuff. We don't call a 2ghz C2D a 2500+ just because it outruns an X2 at that speed do we?

 

[Lightnix]

There are 10 SIMD's worth of 16 units of 5 shaders. Not all shaders on R770 are born equal, though, four of them do MADD (multiply add) and one in five can do more complex operations like cos, tan, etc. The problem is there isn't always a complex operation that needs to be done. G80 (and GT200 I think) differs in that all of its shader cores are exactly alike (MADD+MUL), and can do 3 FLOPS as opposed to the 2 FLOPS per shader that RV770 can do.

I think that's right anyway...

Then you have to factor in that nVidia's shaders are running much faster than the core clock speed.

What does all this mean? Well not a lot really, just that the two shader architectures are apples and oranges, it's completely dependant on situation.

 

[Alexrose1uk]

We don't call a 2ghz C2D a 2500+ just because it outruns an X2 at that speed do we?

lol no they call it a EXXXX in a marketing system, just the same as AMD, they dont talk about core speeds either anymore. The old + ratings were based on what was available at the time, and what 'common folk' would understand to compare to, not IT bods who are a little bit more in the know than the average user!

 

[Talrinys]

There are 10 SIMD's worth of 16 units of 5 shaders. Not all shaders on R770 are born equal, though, four of them do MADD (multiply add) and one in five can do more complex operations like cos, tan, etc. The problem is there isn't always a complex operation that needs to be done. G80 (and GT200 I think) differs in that all of its shader cores are exactly alike (MADD+MUL), and can do 3 FLOPS as opposed to the 2 FLOPS per shader that RV770 can do.

I think that's right anyway...

Then you have to factor in that nVidia's shaders are running much faster than the core clock speed.

What does all this mean? Well not a lot really, just that the two shader architectures are apples and oranges, it's completely dependant on situation.

Exactly, AMD have some fantastic cards out right now, so no one should take this as a bash against AMD, that would be weird seeing as i probably have a 4870x2 on it's way. It's rather the fact that worst case scenario you're only getting 1 out of 5 of the instructions through one shader unit, because there are 160 physical units on the chip. In most games the architecture works fantastically and there's no doubt that AMD are on to a fantastic way of dealing with the graphics market right now. It is a problem in GPGPU comparisons though, that's just my point.

Sorry if it seemed a bit harsh or anything, but the steady stream of threads saying "AMD have a lot higher numbers, why isn't it the fastest by xxx percent?" are getting kind of tiring when AMD should just quote the real physical number of shaders.

 

[Talrinys]

lol no they call it a EXXXX in a marketing system, just the same as AMD, they dont talk about core speeds either anymore. The old + ratings were based on what was available at the time, and what 'common folk' would understand to compare to, not IT bods who are a little bit more in the know than the average user!

The EXXX names aren't related to clock speeds though, the AMD ones were directly related to which clock speed Intel would need to be at to beat it. Funny that they don't keep up that naming scheme now that they're at best toe to toe and Penryn usually beats Phenom clock for clock. AMD were a gamer's choice for many years for exactly the same reasons that Intel are now, this is my first Intel rig ever

 

[Alexrose1uk]

Now that the old cpus are gone, they don't need or want to use the same naming scheme, I don't blame them, like Intel have done several times, they've chosen an appropriate naming system based on the current situation. Yes, AMD were a gamers choice for year (I used AMD XP processors etc as well), however gamers and enthusiasts always go for whats best at the time, nothing new there. As I said, the E system isnt related to clock speeds, in the same way the current AMD aren't.

They've both gone model numbers, as the old systems are no longer relevant, I was simply commenting on your original statement, pointing out that its not just AMD who heavily spin, Intel are just as bad.

 

[Talrinys]

Now that the old cpus are gone, they don't need or want to use the same naming scheme, I don't blame them, like Intel have done several times, they've chosen an appropriate naming system based on the current situation. Yes, AMD were a gamers choice for year (I used AMD XP processors etc as well), however gamers and enthusiasts always go for whats best at the time, nothing new there. As I said, the E system isnt related to clock speeds, in the same way the current AMD aren't.

They've both gone model numbers, as the old systems are no longer relevant, I was simply commenting on your original statement, pointing out that its not just AMD who heavily spin, Intel are just as bad.

Completely agreed, sorry about that

 

[Deleted member 66701]

The EXXX names aren't related to clock speeds though

Yes they are - Intel base it on what a comparable P4 would be, so:-

E6600 = 6600mhz P4.
E8400 = 8400mhz P4.

Etc, etc.

Which is excatly what AMD did when they compared thier processors against the then current P4's, i.e.

XP2500 - 2500mhz P4
XP3200 - 3200mhz P4

And so on.

 

[Lightnix]

Yes they are - Intel base it on what a comparable P4 would be, so:-

E6600 = 6600mhz P4.
E8400 = 8400mhz P4.


And so on.

No they're not, an E2200 walks all over a 2.2GHz Pentium 4 in every possible way imaginable. Besides, there're overlaps, like an E6850 would beat an E8200 (at stock at least, 3GHz and 2.66GHz respectively).