As the tile says amd shows its new road map at the special event in Japan http://www.fudzilla.com/news/processors/37395-amd-announces-glorious-five-year-plan
Is Fudzilla the only source for such news? An APU with 200-300watt TDP? The TDP alone would negate the benefits of an APU....
Am I reading that right, an APU with 200 TDP? Wow, I guess that is made just for a full sized desktop and no mobile variants!
Fudzilla says 200-300watt TDP for APUs. It would have to be water-cooled.....or use a custom AC Accelero Extreme......
Idk high performance computing to me means it's made for servers or something, in which case you can get kind of nontraditional with the cooling and for it to still be acceptable.
Yeah but if you're building a super computer and you're taking an X86 processor and coupling it with 5TFLOP of FP32 precision and doing crazy math workloads on it, suddenly 300w per unit doesn't look too bad.
Well when I server I'm talking more about a compute server. http://en.wikipedia.org/wiki/SuperMUC http://www8.hp.com/us/en/products/servers/high-performance-computing/apollo/apollo8000.html http://www.cray.com/products/computing/cs-series/cs400-lc They have watercooled compute systems, Nvidia is going into a lot of these new systems and I'm sure AMD wants part of that market at some point.
First AMD has to convince the devs writing the specialized software to switch from CUDA to OpenCL. AMD will have a hard time getting a foot hold in a market that's dominated by CUDA.
Hehe yes, you probably are right, but even in super computers the performance has to match up to the heat generated. I know nothing of super computers or compute servers.
For super-computers, the concern is more about performance than heat generation. With a "super-computer", exotic cooling is a bit more acceptable than it is with an HPC server. If a company or university is shelling out the cash for a "super-computer", then they already know exactly what they're buying and what the maintenance and operational costs will be. If a company is buying a "compute server"....they're really not worried about heat generation. Compute servers produce a ton of heat.
Well lets look at it like this. A CPU now a days probably on average is on the range between 70-100 TDP, add overclocking in and that raises even higher. A video card takes about what, 200W at its peak, maybe more? Kind of seems like AMD is placing a high end gpu on the same dye as a cpu, for compute computing this is pretty awesome. Cooling wise, it is going to need a huge tower cooler. Now the advantage is the thermal control with the gpu and cpu cores will be one in the same, and can throttle and lower at the same time and rate. Memory is going to be something else too, looks like AMD is planning stacked HPC in the APU now. Are they going to integrate some memory into the dye as well? Hopefully maybe move the high end APU to DDR4? Only time will tell! If they are planning on making a higher powered x86 SoC I am very excited for this and cannot wait to see something come from this!
I'm waiting for AMD's "Zen" architecture personally. That will be my next upgrade, depending on overall cost of parts.
http://www.fudzilla.com/news/processors/37395-amd-announces-glorious-five-year-plan Sounds a bit like VISC, and if so, whether it also applies to their x86 offerings.
APUs already reached Full SoC level. Now, you can have APU on card similar to graphics card and it can be full system or part of large system. Or you can have full high end system at size of bit larger external drive, who has that? Considering Carrizo fits 4CPU+512SP into 40W TDP, that is a lot of CU/SP in 200-300W. And as for memory... HSA tells all, why not? HBM all the way, but 2nd generation will be required for capacity.