Originally Posted by hmm
I remain highly skeptical.
That is all right, I'm not real strong on this. Here is the problem, unless Intel pulls a rabbit out of the hat not much is getting updated at WWDC. I just don't see a lot of chips suitable for Apples line up being ready for June.
Here is what I find implausible about the use of E3s. Their primary difference in the context of a workstation compared to the i7s used in imacs would be the ability to use ECC ram. E3s could be used in an xmac type machine if they felt the need to maintain some kind of solution that preserves ECC ram. Most people never experience problems from bit flipping or even understand that parity based correction only applies to single bit errors.
I was under the impression that modern ECC codes could detect and correct more bad bits than that. It doesn't really matter as you say only a few people even understand the benefit. However whatever does replace the Mac Pro needs to support ECC or Apple will have yet another big gap in its line up.
It's really there in workstations for things like CAD and engineering applications. I don't see Apple moving toward E3 just for those customers.
The biggest problem for E3 in a Mac Pro replacement is addressable RAM.
If thunderbolt really is a high priority to Apple as a part of their core product strategy, the i5 and i7 versions make more sense than their E3 counterparts. This is because they include some form of embedded graphics, so Apple can use their current solution to thunderbolt there rather than rely on internal switches or whatever is necessary to include it. I don't see them as providing a high net benefit compared to the elusive headless imac. Am I missing something there?
TB implements a form of a cross bar switch to route single out to the TB bus. That chip can connect to integrated GPUs or discrete GPUs from what in understand. As you know finding information on Intels sight is very difficult these days.
I'm not that well read on Xeon Phi. I know you mentioned cheaper variants. The ones I've read about were the chips that cost several thousand yet provide a simpler development path than GPGPU. I'm also not completely familiar with their advantages relative to the typical ever increasing core counts on Xeon EP chips.
Well you get around 60 cores. Performance is as much as ten times what you can get on a dual socket XEON machine. That of course depends upon the code. I mention Phi because it is an alternative, I suspect that Apple will prefer GPU compute.
I do know that Xeon EP is aimed at servers first with workstations being more of a leveraged solution. What little I have read about Xeon Phi has been in the form of tech articles. They typically suggest it's intel's answer to GPGPU use in HPC solutions. The proposed advantage is that they're easier to develop for than NVidia's Tesla cards. There's also the issue that NVidia subsidizes the Teslas somewhat through their higher volume solutions, which Intel has encroached upon with their own integrated gpu solutions.
That is pretty much it. Right now they are in effect co processors that are easier to program than a GPU.
In any event I just find it interesting that it is so difficult to find info on Intel initiatives that they freely had on their web site while back. It is like they have adopted Apples security concepts and now hide everything in development from the public.