Possible AMD Fusion details

[iMacmatician]

So on this thread there's some info (of unknown reliability, although it fits with past reports) on AMD's upcoming Fusion CPUs.

"AMD Family 12h" appears to be Llano, while "AMD Family 14h" appears to be Bobcat/Ontario.

My estimate is FS1 = regular notebook, FP1 = low-power notebook, FM1 is desktop, and FT1 is Ontario.

Llano appears to be ~205 mm^2 for quad-core variant, with the GPU speculated to be between 240 and 480 SPs. Ontario seems to be ~75 mm^2, and its GPU is loosely speculated to be 80 SPs.

This info may be useful given the AppleInsider rumor that Apple may be using AMD CPUs in the future.

 

[Eidorian]

That's a lot of sockets. 80 HD 5000 class shaders is pitiful.

 

[Hellhammer]

30W for quad core, sounds cool! Any info about clock speeds yet? Turbo? Maybe AMD can finally provide decent mobile CPUs to fight against Intel

 

[Eidorian]

30W for quad core, sounds cool! Any info about clock speeds yet? Turbo? Maybe AMD can finally provide decent mobile CPUs to fight against Intel

Hyperthreading and Turbo is supposedly on the feature sheet.

Otherwise the HD 5000 style IGP is what I'm interested in. DirectCompute 5.0 compatibility should be easy for this IGP.

 

[iMacmatician]

Any info about clock speeds yet? Turbo?

Not much I could find except for the "3 GHz+" AMD mentioned last year, 1MB L2 cache/core on Llano, and Evergreen derivative GPU. As Eidorian said, there appears to be Turbo CORE. There's a good deal of speculation on the unknown features though, some of which I noted above.

 

[Eidorian]

iMacmatician, the GF104/6/8 have shown up on SemiAccurate (Guru3D -> HardOCP -> HardOCP Forums?) again but once more no one has any concrete information.

240 shaders doesn't make any sense for GF104 (GTX 460) given that Fermi is built on blocks of 32.

 

[Hellhammer]

Hyperthreading and Turbo is supposedly on the feature sheet.

Otherwise the HD 5000 style IGP is what I'm interested in. DirectCompute 5.0 compatibility should be easy for this IGP.

Sweet. I have feeling that AMD might actually win this round if their pricing is reasonable. The IPG sounds very nice though, should be much better than Intel?

Not much I could find except for the "3 GHz+" AMD mentioned last year, 1MB L2 cache/core on Llano, and Evergreen derivative GPU. As Eidorian said, there appears to be Turbo CORE. There's a good deal of speculation on the unknown features though, some of which I noted above.

+3GHz with Turbo sounds possible

iMacmatician, the GF104/6/8 have shown up on SemiAccurate (Guru3D -> HardOCP -> HardOCP Forums?) again but once more no one has any concrete information.

240 shaders doesn't make any sense for GF104 (GTX 460) given that Fermi is built on blocks of 32.

Maybe they are using 7 and half blocks then They should be out pretty soon shouldn't they?

Anyway, I'm off to bed for this night! Nice to talk with you guys as you actually know something (a lot actually), not just quote Wikipedia and other lazy sources

 

[iMacmatician]

There's speculation on Beyond3D (not about 240 SPs specifically) that SMs could be cut in half or SMs could be modified to have 48 SPs instead of 32.

 

[Eidorian]

Sweet. I have feeling that AMD might actually win this round if their pricing is reasonable. The IPG sounds very nice though, should be much better than Intel?

Both Intel and AMD presented their "next generation" IGP technologies at Computex last week. Intel had a relatively idle Mass Effect 2 one while AMD had a Aliens vs. Predator with action in DirectX 11. No word on settings or resolutions but both companies claimed the usage of all the fancy, shiny effects you'd see from a dedicated GPU.

I'm still going to bet on AMD being able to provide a much more robust solution then Intel. ATI brings a lot to the table but Intel has also invested a lot with Larrabee.

Maybe they are using 7 and half blocks then They should be out pretty soon shouldn't they?

No one knows but many people claim that they do. At least outside of nVidia itself. I'll keep waiting.

There's speculation on Beyond3D (not about 240 SPs specifically) that SMs could be cut in half or SMs could be modified to have 48 SPs instead of 32.

I remember a few mentions that ASUS had some GF106/8 mobile parts at Computex. It would be something along the lines of a GTX 460M.

In a previous generation, the structure from the G80 to the later G92 didn't change that much (128:64:x) but the G92 did have fewer ROPs. (16 vs. 24)

 

[iMacmatician]

Anyway, I'm off to bed for this night! Nice to talk with you guys as you actually know something (a lot actually), not just quote Wikipedia and other lazy sources

Good night!

Both Intel and AMD presented their "next generation" IGP technologies at Computex last week. Intel had a relatively idle Mass Effect 2 one while AMD had a Aliens vs. Predator with action in DirectX 11. No word on settings or resolutions but both companies claimed the usage of all the fancy, shiny effects you'd see from a dedicated GPU.

Interestingly enough the AMD chip was Ontario, not Llano.

 

[Hellhammer]

I'm still going to bet on AMD being able to provide a much more robust solution then Intel. ATI brings a lot to the table but Intel has also invested a lot with Larrabee.

AFAIK, Larrabee wasn't as successful as they hoped. AMD should be able to bring a lot better IGP because they are the same with ATI now and ATI is the king of GPUs atm. If AMD can deliver quad core with IGP and TDP of 30W, it would be brilliant. AMD's pricing is also a lot more reasonable than Intel's so let's keep waiting, 2011 sounds to be great year for computing

 

[Eidorian]

Interestingly enough the AMD chip was Ontario, not Llano.

Quite true, you're going to need the ~480 SPs for that game as well.

 

[DoFoT9]

AFAIK, Larrabee wasn't as successful as they hoped. AMD should be able to bring a lot better IGP because they are the same with ATI now and ATI is the king of GPUs atm. If AMD can deliver quad core with IGP and TDP of 30W, it would be brilliant. AMD's pricing is also a lot more reasonable than Intel's so let's keep waiting, 2011 sounds to be great year for computing

30w would be very nice. what im most interested in is the clock frequency! software doesnt seem to be moving along terribly fast, and its nice that quad/octo cores are coming along but i feel that increases in clock speed are needed as well as increases in parallelism.

 

[Hellhammer]

30w would be very nice. what im most interested in is the clock frequency! software doesnt seem to be moving along terribly fast, and its nice that quad/octo cores are coming along but i feel that increases in clock speed are needed as well as increases in parallelism.

That's why there is Turbo 1.5GHz is fine as base clock if it can overclock near 3GHz with Turbo. With 32nm process, I would expect +2GHz as base clock with reasonable TDP plus hopefully insane Turbo speeds (~3.5GHz?)

 

[DoFoT9]

That's why there is Turbo 1.5GHz is fine as base clock if it can overclock near 3GHz with Turbo. With 32nm process, I would expect +2GHz as base clock with reasonable TDP plus hopefully insane Turbo speeds (~3.5GHz?)

the idea of boosting up the CPU is a nice concept for the current situation, but we do not know if it will last, or be needed. currently we need it. a 2.8GHz i7 860 for example can TB up to 3.46Ghz for single threaded processes only. dual threaded processes are scaled down more, and tri-threaded even more so.. etcetc. with the introduction (down the track of course ) of "proper" multi threaded apps - there will be less need for TB and personally id rather constant clocks so i get constant performance (e.g. the best mix of parallelism and frequency).

thats intel related stuff, is the AMD based Turbo concept similar to that? i wanna know lol

 

[Hellhammer]

the idea of boosting up the CPU is a nice concept for the current situation, but we do not know if it will last, or be needed. currently we need it. a 2.8GHz i7 860 for example can TB up to 3.46Ghz for single threaded processes only. dual threaded processes are scaled down more, and tri-threaded even more so.. etcetc. with the introduction (down the track of course ) of "proper" multi threaded apps - there will be less need for TB and personally id rather constant clocks so i get constant performance (e.g. the best mix of parallelism and frequency).

thats intel related stuff, is the AMD based Turbo concept similar to that? i wanna know lol

After quickly reading about AMD's Turbo CORE it seems like it always overclocks three cores (in six core CPUs), so the gain isn't that big (could go much more with only one core). Of course, AFAIK, Thuban was the first chip with it so we can expect some improvement in Fusion, possibly same as Intel's (so it would configure it depending on the load i.e. 1, 2 or 3 cores shut down)

Feel free to correct if I'm wrong, as I said, that was after quick googling

 

[DoFoT9]

After quickly reading about AMD's Turbo CORE it seems like it always overclocks three cores (in six core CPUs), so the gain isn't that big (could go much more with only one core). Of course, AFAIK, Thuban was the first chip with it so we can expect some improvement in Fusion, possibly same as Intel's (so it would configure it depending on the load i.e. 1, 2 or 3 cores shut down)

Feel free to correct if I'm wrong, as I said, that was after quick googling

3 cores hey? yea if they allowed it to be variable then they could OC 1 core to >4GHz i guess? maybe.. depending on the thermals of course.

 

[iMacmatician]

Looking at Intel CPU turbo modes, the lower the GHz (or TDP) of the CPU, the higher the relative Turbo Boost. From that I'd expect Llano and Ontario to Turbo quite well.

 

[DoFoT9]

Looking at Intel CPU turbo modes, the lower the GHz (or TDP) of the CPU, the higher the relative Turbo Boost. From that I'd expect Llano and Ontario to Turbo quite well.

which is fair enough, there would physically be a limit of how high the CPUs can be clocked - even on 1 core, due to the thermals and whatnot. given the much lower TDP of these AMD chips i presume the ability to Turbo would be much more expansive.