Amd/ATI Dual Core GPU (maybe next)

[Heinz68]

Nice performance increase plus maybe dual core, way to go AMD
"AMD next generation RV870 key specs revealed"
Date: 2008-7-18 Author: Rocky at Hardspell (http://en.hardspell.com/doc/showcont.asp?news_id=3768)
It is said that AMD next generation graphic core possible be named RV870 and according to TSMC technology, it will use 40nm or 45nm technology. The core area of RV870 will be about 140m㎡ which is much smaller than RV770 260m㎡. As we know by now, it will have 192 ALU. RV770 each ALU matched 5SP and then RV870 will have 960SP. In order to control the core area, it is still 256bit. We believe RV870 will be 1.2 times than RV770 in performances, but this will be decided by the clock of RV870.

It is also said that AMD next generation R800 will use new design. We know Radeon HD3870X2 and the coming Radeon HD4870X2 both used single PCB+dual graphic core design while R800 will possible use real dual core design. If so, AMD next generation flagship R800 will be the first dual core GPU. The specs of R800 will double RV870.

Advanced 45nm (40nm?) will bring RV870 smaller core area. The current RV770 did well in performances but the temperature is really terrible. If RV870 can settle this problem and further improve performances, it will be really excited for us and it will be the first real dual core GPU possibly.

[CaptNKILL]

Hmm... that'd be interesting.

Though these guys have been pumping out rumors like crazy lately. I wouldn't take anything they say too seriously.

[Redeemed]

By combining both GPUs into the same die you'd just make one big GPU essentially. GPUs are already broken up into smaller units (you could technically call them "cores") hence why they're massive parallel. I'm calling BS as this just doesn't seem pheasible.

[CaptNKILL]

By combining both GPUs into the same die you'd just make one big GPU essentially. GPUs are already broken up into smaller units (you could technically call them "cores") hence why they're massive parallel. I'm calling BS as this just doesn't seem pheasible.

If there is any truth in it, its probably related to making multiple GPUs work as one rather than requiring Crossfire, bridge chips, driver profiles or doubled frame buffers.

ATI is focusing on making one high end chip and then sticking more of them together to fill in the ultra high end segment (like the X2 cards), rather than developing a specific chip for each price segment.

If they could stick a couple chips on a board and have them act as one GPU it'd save them a lot of time in driver development not to mention all the money that wouldn't be spent on bridge chips and having twice as much memory on board.

Still, this rumor is probably a bunch of BS. It'd be nice, for sure, but I doubt its real.

[bacon12]

AMD's dual core GPUs will out perform their CPUs if they aren't careful.:p

[hemmy]

Technically, GPUs already have multiple cores.

Dual core GPUs is incorrectly used here IMO.

[CaptNKILL]

Technically, GPUs already have multiple cores.

Dual core GPUs is incorrectly used here IMO.

People keep saying this, but I think its pretty clear what is meant every time the term "core" is used when talking about GPUs.

No one says their GPU has 128 cores. They say it has 128 shader processors. The shader processors may technically be "cores" but if a GPU is "dual core" it obviously means it has two GPUs.

[mullet]

I concur with Sir Capytan's analysis of the situation.

[Heinz68]

By combining both GPUs into the same die you'd just make one big GPU essentially. GPUs are already broken up into smaller units (you could technically call them "cores") hence why they're massive parallel. I'm calling BS as this just doesn't seem pheasible.
It's not going to be the first time when people talking about future technology as not pheasible and BS, were proven wrong.

Now I'm not an expert so can only provide my very simplistic view, yours did not go in much more details anyway.

First about the one big GPU
HD 4870 die size is 256 mm2, very small compared to GTX 280 huge 576 mm2 size.
Look at AnandTech (http://www.anandtech.com/video/showdoc.aspx?i=3341&p=2) at visual comparision
http://img524.imageshack.us/img524/2130/diesizect5.jpg
The rumors are that the RV870 is going to use 40nm technology, which is much smaller, some predict 140mm2 die size. So if it needs to be at least doubled for R800 it shouldn't be problem.

Second about the shader processors.
The RV870 is rumored to have 1000 shader processors, the R800 (Dual Core) can double it to 2000.
No idea why it would work only on 2x GPU.

Third point.
AMD with so much experience on multi cores should be able to deliver it.

Now all this is going to bring the high end PC gaming closer for the masses, no more $2000 tri-SLI.

For some people worrying about not being able to show the size of the e-penis anymore, I say don't worry there is always going to be some way to show it. :)

[Redeemed]

It's not going to be the first time when people talking about future technology as not pheasible and BS, were proven wrong.

Now I'm not an expert so can only provide my very simplistic view, yours did not go in much more details anyway.

First about the one big GPU
HD 4870 die size is 256 mm2, very small compared to GTX 280 huge 576 mm2 size.
Look at AnandTech (http://www.anandtech.com/video/showdoc.aspx?i=3341&p=2) at visual comparision
http://img524.imageshack.us/img524/2130/diesizect5.jpg
The rumors are that the RV870 is going to use 40nm technology, which is much smaller, some predict 140mm2 die size. So if it needs to be at least doubled for R800 it shouldn't be problem.

Second about the shader processors.
The RV870 is rumored to have 1000 shader processors, the R800 (Dual Core) can double it to 2000.
No idea why it would work only on 2x GPU.

Third point.
AMD with so much experience on multi cores should be able to deliver it.

Now all this is going to bring the high end PC gaming closer for the masses, no more $2000 tri-SLI.

For some people worrying about not being able to show the size of the e-penis anymore, I say don't worry there is always going to be some way to show it. :)


To better clarify, the term "dual-core" cannot be accurately applied. If the GPU were literally "dual-core"... then that'd mean it'd only have 2 shader processors essentially which is a massive step backwards.

So, if you have to seperate GPUs, each GPU having 100 stream processors... and you then combine them onto one die it'd be no different than just developing a single GPU that had 2000 stream processors. The two would literally be one in the same. I hope that makes more sense.

[Xion X2]

Actually, it doesn't. The GPU and its internals can be compared to a CPU and its internals. They each have a group of parts that make up the whole. So, when saying "core" of a graphics card, you are basically talking about the GPU just as you would be talking about the CPU if you were to say "core" in reference to it. The only real difference is that a graphics card has its own circuitry on the board (vram, regulators, etc.) and a CPU relies on the motherboard circuitry.

Anyway, this is nothing but semantics and irrelevant.

It shouldn't be that surprising that ATI is mimicking a CPU design since they merged with AMD. Only makes sense they may go this route. Wouldn't surprise me if this rumor is true.

[Heinz68]

To better clarify, the term "dual-core" cannot be accurately applied. If the GPU were literally "dual-core"... then that'd mean it'd only have 2 shader processors essentially which is a massive step backwards.

So, if you have to seperate GPUs, each GPU having 100 stream processors... and you then combine them onto one die it'd be no different than just developing a single GPU that had 2000 stream processors. The two would literally be one in the same. I hope that makes more sense
Not really. I don't remember anybody referring to the stream processors on the GPU as 100 Cores etc. It almost looks to me like playing with words. I'm sure you know what Dual Core GPU means in this case.
End off argument at least for me, I know your points you know mine. Lets see what will happen.

PS I also remember when Intel first came with dual core CPU everybody saying "that's not really dual core it's 2 CPU glued together", the problem for AMD was, Intel must have been using some great glue.

[Redeemed]

Not really. I don't remember anybody referring to the stream processors on the GPU as 100 Cores etc. It almost looks to me like playing with words. I'm sure you know what Dual Core GPU means in this case.
End off argument at least for me, I know your points you know mine. Lets see what will happen.

PS I also remember when Intel first came with dual core CPU everybody saying "that's not really dual core it's 2 CPU glued together", the problem for AMD was, Intel must have been using some great glue.

The A64 X2s obliterated the dual core P4s (Intel's first dual-core chips). Intel never actually got ahead until the release of the Core2 Duos.

As for your first part of your post... fair enough.

[Runningman]

For the love of god, not this rumor again!!!

[Ninja Prime]

The whole idea in this is to make GPUs modular. If you can get multi-GPUs to work well enough together, you can make your whole product line based off of one chip. Meaning you only have to make one chip. This would be huge for money/manufacturing savings if they could pull it off.

Like this:

Top of the line card: ATI 5870: 4 GPUs on a die, 1024 bit memory interface(4x256).
Middle of the road card: ATI 5850: 2 GPUs on a die, 512 bit memory interface(2x256).
Low end card: ATI 5830: 1 GPU on a die, 256 bit memory interface.

All this with one manufacturing line, one actual die, all you need to do is slap more dies on and more memory to make a faster card. This way, it could be far cheaper to make video cards, you could keep dies small where yields are very high and you could pump out a huge volume of one die and just sort out how many you need for each market afterwards.

[Viral]

I don't like the fact that the author of the article is unsure whether they'll use 40nm or 45nm technology considering it's a fact that TSMC will be using their 40nm process for GPU's.

I think the idea behind this is to properly share memory. Using a fast link, the two 256-bit controllers can act as one 512-bit controller. This will be needed if they want their next X2 card to stand a chance against the GT300 with its rumored 512-bit GDDR-5. Also, it will mean they don't need to use twice the amount of memory and as sch can be more easily price competitive.

[Xion X2]

The whole idea in this is to make GPUs modular. If you can get multi-GPUs to work well enough together, you can make your whole product line based off of one chip. Meaning you only have to make one chip. This would be huge for money/manufacturing savings if they could pull it off.

Exactly, and it looks as if they're already heading in that direction with the 4850/4870 series. Same chip, different clockspeeds & RAM configurations.

[CaptNKILL]

The whole idea in this is to make GPUs modular. If you can get multi-GPUs to work well enough together, you can make your whole product line based off of one chip. Meaning you only have to make one chip. This would be huge for money/manufacturing savings if they could pull it off.

Like this:

Top of the line card: ATI 5870: 4 GPUs on a die, 1024 bit memory interface(4x256).
Middle of the road card: ATI 5850: 2 GPUs on a die, 512 bit memory interface(2x256).
Low end card: ATI 5830: 1 GPU on a die, 256 bit memory interface.

All this with one manufacturing line, one actual die, all you need to do is slap more dies on and more memory to make a faster card. This way, it could be far cheaper to make video cards, you could keep dies small where yields are very high and you could pump out a huge volume of one die and just sort out how many you need for each market afterwards.
Just like 3dfx with the VSA-100 chips back in the Voodoo 4 and 5 days. The Voodoo 4 cards had 1 chip and 32Mb of memory, the Voodoo 5 5500 had 2 with 64Mb and the Voodoo 5 6000 had 4 with 128Mb (though it was never mass produced). Though I think they only really had 32Mb available because of SLI.

[bkswaney]

Just like 3dfx with the VSA-100 chips back in the Voodoo 4 and 5 days. The Voodoo 4 cards had 1 chip and 32Mb of memory, the Voodoo 5 5500 had 2 with 64Mb and the Voodoo 5 6000 had 4 with 128Mb (though it was never mass produced). Though I think they only really had 32Mb available because of SLI.

"LOL" I was thinking what U wrote. :p

[Amuro]

Intel's Larrabee GPU will feature multiple x86 cores.