By using this site, you agree to our Privacy Policy and our Terms of Use. Close
fatslob-:O on 08 November 2013
Edit Thread
curl-6 said:
fatslob-:O said:
curl-6 said:
fatslob-:O said:

The IPC is actually worse.  

5 gflops per core at 1.2ghz is pretty bad but you shouldn't be depending too much on the CPU to help with rendering because all of the newer consoles are GPU accentric and CPUs rarely do anything for rendering if any at all. 

Marcan, the hacker who stripped down Wii U and first revealed it's clock speeds and architecture, stated that Espresso, Wii U's CPU, should "win big on IPC on most code" compared to the less efficient CPUs of PS3/360.

Espresso may be lower clocked than its rivals but it has far more cache, a shorter pipeline less affected by performance robbing bubbles, out of order execution, and a DSP for offloading the audio workload. 

Oh I had meant IPC in terms of Flops not IPC in terms of other things such as integer etc etc but yeah it's out of order design really helps out to geet the most of it's power and the cache will reduce performance hits but I'd rather push aside the ibm espresso and let the latte graphics chip do the talking. 

Fair enough.

I guess it's just that with Espresso I at least know what I'm talking about, while Latte is still so nebulous, haha.

Well it's not your fault. Most people couldn't even begin to understand how a GPU works other than the mantra that the GPU is known to be massively parallel. After all they are the ones with the shit ton of transistors so it's only fair that they would have the most complexity involved in them. You have to account for things such as concurrency or how scalable a workload is for a GPU. You even have to know what types of rendering methods they and what drawbacks that it will pose. There are many parts to a GPUs performance such as functional units like shaders, ROPS, TMUs, and even bandwidth as well as memory. Oh and that's just the surface of things! CPUs have not become very complex in their design because they benefit more from serial based workloads and the laws of physics prevents them from going into higher frequencies so chip designers altogether except for intel stopped trying to make to make significantly more powerful and even then intel only does incremental boosts to their CPU performance, not the 20% like we used to get back in the day with a new process node! I can't even begin to start programming for GPUs yet even though microsoft made it easier with the C++ AMP library because it's still too hard for me!