- Joined
- Sep 29, 2004
- Thread Starter
- #81
^^^^ Not at all, you're forgetting just how horrible the cache performance is in Bulldozer. I suspect leaving out the L3 will barely effect performance. AMD architectures have traditionally not benefitted much by using more cache.
I think you're actually missing the big picture here, and that's the nature of the GPU built in. No longer a VLIW design but rather GCN which means its effectively vector based, making them far more suited to general compute tasks than anything nVidia currently have. Basically more types of tasks will get a boost form using the GPU due to its higher flexabillity. I'd still expect nVidia to be faster in some specific tasks though. GP Compute tasks will thrive on this architecture and run orders of magnitude above ANYTHING else out there. It's actually the straight Piledriver chips that are hobbled due to no integrated GCN based GPU.
Of course, its actually supercomputers where I expect these will get the greatest use. It'll be a few years before we'll see desktop apps that make use of this.
I think you're actually missing the big picture here, and that's the nature of the GPU built in. No longer a VLIW design but rather GCN which means its effectively vector based, making them far more suited to general compute tasks than anything nVidia currently have. Basically more types of tasks will get a boost form using the GPU due to its higher flexabillity. I'd still expect nVidia to be faster in some specific tasks though. GP Compute tasks will thrive on this architecture and run orders of magnitude above ANYTHING else out there. It's actually the straight Piledriver chips that are hobbled due to no integrated GCN based GPU.
Of course, its actually supercomputers where I expect these will get the greatest use. It'll be a few years before we'll see desktop apps that make use of this.