AMD: R8xx Speculation

[DegustatoR]

Now that we know that juniper is 1040 million transistors and 170mm2 we can say that probably Cypress (2150 million transistors and 334mm2) has a lot of redundancy transistors. (24SIMDS?)

170x2=340, 1040x2=2080. 70M is a lot?

 

[Kaotik]

170x2=340, 1040x2=2080. 70M is a lot?

You know, Juniper isn't exactly Cypress cut in half, they both have a lot of functionality in there of which there's same amount of units in both chips (PCI-E interfaces, tesselator, UVD, display engines and so on)

 

[Blazkowicz]

I saw the reviews, 5770 scores around -10% than what i expected.
In some games like Far Cry 2 or like Company of Heroes the 5770 is faster than a 4890.
But overall it seems like -10% of the 4870 1GB perf. (1920X1200 4AA 16AF)

To me that's an absurdly fast 128bit card, roughly matching the 256bit 4870 and the 448bit GTX260. I didn't quite expect that . (I could have, given the beast that RV740 is already)

 

[neliz]

When I said, there is no "X2" .. I ment: This

 

[seahawk]

Fatanstic, do you know if the +30-50% GPU clock are true?

 

[DegustatoR]

Fatanstic, do you know if the +30-50% GPU clock are true?

It's probably closer to -30-50% GPU clocks thus it's not x2 anymore since it has lower than 5870 clocks and lower than 5870 x2 performance. So it'll be called 59x0 instead of 58x0 X2.
Am I doing it right, neliz?

 

[neliz]

It's probably closer to -30-50% GPU clocks thus it's not x2 anymore since it has lower than 5870 clocks and lower than 5870 x2 performance. So it'll be called 59x0 instead of 58x0 X2.
Am I doing it right, neliz?

That's more like it! attaboy!

 

[Sampsa]

NH has updated their news post since it was mostly based on speculation and widely referred in other media:

http://www.nordichardware.com/news,10043.html

 

[trinibwoy]

Why though? According to Xbit a single 5870 draws less power than a 4870X2. I would expect higher clocks if anything.

 

[Sampsa]

Why though? According to Xbit a single 5870 draws less power than a 4870X2. I would expect higher clocks if anything.

Because 5900 series is something else than 5870 refresh like 4890 was to 4870

 

[trinibwoy]

Maybe but what does that have to do with clocks? Or are we back to the RV870 has more than 20 SIMDs rumour?

 

[fellix]

Or are we back to the RV870 has more than 20 SIMDs rumour?

Meh, not enuf rasterizers then...

 

[DegustatoR]

Maybe but what does that have to do with clocks?

5870 clocks are too high to make a 5870X2 card without lowering these clocks (TDP and power consumption won't allow that).
One might say that 5870 in the EG line is like 4890 in 700 line already.
Thus they'll need to make a "new 4870" which will be used in the "new 4870X2". Such "new 4870" will probably have the same difference to the 5870 as 4870 had to 4890. Lower clocks. Thus a card based on two such chips can't be called 5870X2 since it's not 5870 x2, it'll be slower, closer to 5850X2.
Well, that's how one might think about it.

 

[Kaotik]

5870 clocks are too high to make a 5870X2 card without lowering these clocks (TDP and power consumption won't allow that).
One might say that 5870 in the EG line is like 4890 in 700 line already.
Thus they'll need to make a "new 4870" which will be used in the "new 4870X2". Such "new 4870" will probably have the same difference to the 5870 as 4870 had to 4890. Lower clocks. Thus a card based on two such chips can't be called 5870X2 since it's not 5870 x2, it'll be slower, closer to 5850X2.
Well, that's how one might think about it.

Is it? It's 26W higher than single HD4870 TDP of 160W.
When you look at 4870 vs 4870X2 TDP wise, they could add 2nd RV770, 1.5GB mem, PLX bridge and required power circuitry with mere 126W increase in TDP over single HD4870.
I don't see fitting full blown HD5870 "X2" within 300W being mission impossible, or even impractible, by any means.
Ignoring TDP, typical load values are even more close, heck, in many cases even lower than HD4870 on the HD5870

 

[no-X]

There are 2 versions of RV870 GPU. The difference is leakage and probably power consumption, too. Maybe ATi will use the less power hungry version for dual-configuration. Other possibility is new GPU revision (just like nVidia created 3rd revision of GT200b for the GX2).

 

[DegustatoR]

Is it? It's 26W higher than single HD4870 TDP of 160W.
When you look at 4870 vs 4870X2 TDP wise, they could add 2nd RV770, 1.5GB mem, PLX bridge and required power circuitry with mere 126W increase in TDP over single HD4870.
I don't see fitting full blown HD5870 "X2" within 300W being mission impossible, or even impractible, by any means.
Ignoring TDP, typical load values are even more close, heck, in many cases even lower than HD4870 on the HD5870

RV790 had even less TDP difference with RV770 but AMD was unable to make 4890X2.

There are 2 versions of RV870 GPU.

Really? Anything to base that on?

 

[Iron Tiger]

Meh, not enuf rasterizers then...

Not even enough bandwidth on the 5870.

 

[neliz]

Really? Anything to base that on?

Knowledge and AMD's word.

 

[Kaotik]

RV790 had even less TDP difference with RV770 but AMD was unable to make 4890X2.

Unable or unwilling due some business decision?

 

[no-X]

Really? Anything to base that on?

http://forum.beyond3d.com/showpost.php?p=1344008&postcount=29
http://forum.beyond3d.com/showpost.php?p=1344083&postcount=33