Confirmed? fx will shift to fm2/+

[bunnyfubbles]

Both our positions shouldn't be mutually exclusive from a platform standpoint.
IE: edram/l4 could benefit both camps
Or having more than just a cookie cutter single die would open up more posibilities

so potentially offering:

2 modules + gpu
4 modules without gpu
or
3 or 4 modules + gpu and edram/hbm/whatever

The first two could potentially have similar die area and cost to manufacture, while the 3rd option a potential higher end "extreme" offering.

no doubt, I'm just thinking in terms of what might be realistically feasible. AMD abandoning the traditional FX line in favor of a focus on APUs tells me they wouldn't scale up die size enough for 4 modules without sacrificing GPU, let lone a 3-4 module extreme edition with an even bigger GPU. That leads me to curb my fantasy to an AMD APU that has midrange dGPU potential rather than entry level that we're currently limited to.

Sure, I'd love a 3-4 module part with an iGPU that was 2-3x more powerful, and I also understand people wanting to see a proper successor to the FX8300s, I just don't see even my 'fantasy' of a 2 module part with 2+ x the GCN cores as likely even though I find it the most reasonably possible, I just don't think AMD is in a position to cater to enthusiasts at all.

 

[ElFenix]

This chip is Richland which has Piledriver cores. Same cores as Vishera.
Take a 4 module Vishera, remove 2 modules, remove L3, add the GPU portion and you have this "new" APU. Nothing to get excited about.

good point. i can't keep all the different core/product code names straight anymore.

you forgot the final step, which is fuse off the GPU 😀

so potentially offering:

2 modules + gpu
4 modules without gpu
or
3 or 4 modules + gpu and edram/hbm/whatever

AMD is all in for APU/SoC being the future, so you're not going to get a die without gpu on it.

 

[Soulkeeper]

good point. i can't keep all the different core/product code names straight anymore.

you forgot the final step, which is fuse off the GPU 😀



AMD is all in for APU/SoC being the future, so you're not going to get a die without gpu on it.

That's a pretty definitive statement. Currently their products reflect this, but assuming that they'll never release a gpu-less part again don't seem a bit much to you ?

 

[parvadomus]

I dont know why AMD opted for 2 modules + 512 shaders when they could have gone 3M + 256 or 384 shaders..

 

[ElFenix]

That's a pretty definitive statement. Currently their products reflect this, but assuming that they'll never release a gpu-less part again don't seem a bit much to you ?

some ARM server part won't have gpu. but i highly doubt they ever again make a consumer die without a gpu on it. soon it'd be like making a part without an FPU. remember when those were actually an optional feature?



fun historical note: the i487 "math coprocessor" part that you could install into a 486SX system actually had an entire 486DX chip in it.

 

[Abwx]

I dont know why AMD opted for 2 modules + 512 shaders when they could have gone 3M + 256 or 384 shaders..

GPU perfs wise a Richland would trounce an hypothetical 3M + 256 shaders Kaveri
and be about on par with a such 384 shaders variant so it would make no sense, the only relevant solution would be with 3M + 512 shaders but it would had increased Kaveri s size up to 290mm2 or so , an economicaly risky option.

 

[Enigmoid]

GPU perfs wise a Richland would trounce an hypothetical 3M + 256 shaders Kaveri
and be about on par with a such 384 shaders variant so it would make no sense, the only relevant solution would be with 3M + 512 shaders but it would had increased Kaveri s size up to 290mm2 or so , an economicaly risky option.

Only with a better memory controller. Performance in AMD's APUs seem to be limited by bandwidth.

That said the FX line seems to have better bandwidth than the APU line (l3 maybe?). With BW limitations 3M + 384 would perform nearly as good (the 384 and 512 kaveri chips perform within ~10%) as 2M +512. Shame.