Strongest AMD Cpu

[harrynowl]

I meant the 1100T and the FX-6300.

Umm.. no they're very different

[Majestic]

Umm.. no they're very different

 

It doesn't show

http://images.hardwarecanucks.com/image//skymtl/CPU/FX-9590/FX-9590-62.jpg

[Deletive]

7850k sorta is if you think about it in a funny way.

[patrickjp93]

Yeah, I believe most of the improvements are masked by old testing methods. Which is understandable if they want to compare it to previous models, but annoying nontheless. 

There's an advanced AVX 2-enabled FLOPs measuring tool out there but the name isn't coming back to me. If someone can hunt it down I can show a 4790k at 4770k clock speeds and someone just has to get the 9590 results for me. Being able to do 3 128-bit or 6 64-bit integer manipulations per cycle per core on Intel chips vs. the 2/4 per AMD core per 2 cycles is just not a contest. In floating point mathematics the disparity is even greater.

[Majestic]

There's an advanced AVX 2-enabled FLOPs measuring tool out there but the name isn't coming back to me. If someone can hunt it down I can show a 4790k at 4770k clock speeds and someone just has to get the 9590 results for me. Being able to do 3 128-bit or 6 64-bit integer manipulations per cycle per core on Intel chips vs. the 2/4 per AMD core per 2 cycles is just not a contest. In floating point mathematics the disparity is even greater.

 

You mean linpack perhaps?

[patrickjp93]

7850k sorta is if you think about it in a funny way.

That's entirely true, but of course developers are taking forever to develop for heterogeneous acceleration. LibreOffice Calc and Microsoft Excel handily give the advantage to the 7850k over the 3770k, and that's just the 1st release of HSA-enabled solvers. I imagine the 7850k can beat the 4770k (CPU only) in raw calculations if the software is tailored correctly. Now, with heterogeneous software taking advantage of 20 Intel GPU cores * 4 ALUs/Core and 4FLOPs/ALU * 1350 MHz, well, the 7850k just loses.

 

With Broadwell we will see a 96-core Intel iGPU solution at the highest end. It was projected that this performance will bring up to 2 TeraFLOPs of performance in just the shaders. If you add in the CPU cores it's roughly 2.6 TFlops. AMD's nowhere close to that on an iGPU.

[patrickjp93]

You mean linpack perhaps?

I think there's a different one, but can you provide a link to the documentation for it?

[Majestic]

I think there's a different one, but can you provide a link to the documentation for it?

 

I don't know, I just tried to google it.

https://software.intel.com/en-us/articles/intel-math-kernel-library-linpack-download

[harrynowl]

I'd like to see a minimum of 16 GCN CUs on the next APU from AMD otherwise Intel are going to take over in integrated performance. On the other hand iris pro chips are not exactly easy to get hold of. On the other hands AMD has gaming optimized drivers help and amd seems to be ahead in the development of a heterogeneous platform.

[patrickjp93]

I'd like to see a minimum of 16 GCN CUs on the next APU from AMD otherwise Intel are going to take over in integrated performance. On the other hand iris pro chips are not exactly easy to get hold of. On the other hands AMD has gaming optimized drivers help and amd seems to be ahead in the development of a heterogeneous platform.

From 28 to 20nm I can see 2-4 more GPU cores, but the CPU cores need an overhaul too so I'd be happy with 10 GPU cores honestly. We still have no clue how many streaming processors are actually on Intel GPUs. I really wish they'd reveal how many per EU.

 

Edit: http://www.realworldtech.com/sandy-bridge-gpu/3/   <--This can't be right that 1 EU is 1 streaming processor. Supposedly a 96 EU solution has 2 TFlops of performance according to Intel. That's more than AMD's 7850k with 512 streaming processors. There's no way Intel put so many ALUs on an EU to make that work. I think RWT needs to do more research because that math doesn't add up.

 

http://www.hitechlegion.com/reviews/processors/38897-amd-a10-7850k-kaveri-quad-core-apu-review?showall=1&limitstart=

With this I become even more skeptical.

[harrynowl]

From 28 to 20nm I can see 2-4 more GPU cores, but the CPU cores need an overhaul too so I'd be happy with 10 GPU cores honestly. We still have no clue how many streaming processors are actually on Intel GPUs. I really wish they'd reveal how many per EU.

 

Edit: http://www.realworldtech.com/sandy-bridge-gpu/3/   <--This can't be right that 1 EU is 1 streaming processor. Supposedly a 96 EU solution has 2 TFlops of performance according to Intel. That's more than AMD's 7850k with 512 streaming processors. There's no way Intel put so many ALUs on an EU to make that work. I think RWT needs to do more research because that math doesn't add up.

 

http://www.hitechlegion.com/reviews/processors/38897-amd-a10-7850k-kaveri-quad-core-apu-review?showall=1&limitstart=

With this I become even more skeptical.

http://www.anandtech.com/show/6993/intel-iris-pro-5200-graphics-review-core-i74950hq-tested/2

 

May be an interesting read, the smallest implementation of HD graphics on i7 CPUs is the HD4600 I believe, which is 20 EUs, for a total of 80 ALUs and 10 SIMDs.

 

Well 12 CUs is what the Xbox one has (it does have 14 but as 2 are disabled for yields). What I'd really like to see is motherboards that sacrifice the expansion near the bottom and replace it with GDDR5 chips

[GlassBomb]

I want AMD to improve on their FX line of processors. The competition is so far ahead in terms of performance to watt.

On Topic. The ridiculous 9590 from AMD is probably the highest performing one that they have on the consumer side, currently.