I hate what the CPU is becoming!

[VirtualLarry]

He wants it the size of a Wii give or take a few inches.

Look at the Acer Revo. Atom+ION, very small package, good for media playback, not so hot for gaming.

 

[DrMrLordX]

Yep...me has a sneeking suspicion that Intel planned to have a QPI socket interface to their MIC platform, much as you are thinking with HTX, to address the network topology barrier (enabling access to those fine-grained and mild-grained apps and marketspace)...which is why there was a specific clause included in the FTC/DOJ stuff precluding Intel from moving away from PCIe for the next 6 or 7 yrs.

(who wins there? not us consumers )

I agree 100%, PCI-e is too latency-intensive for where desktop computing is going. With non-coherant HT connections being possible, it would make sense to move most if not all peripherals off PCI-e and onto HT links instead via HTX slots (on AMD platforms). I can't point to anything in publicly-available QPI specs to support the idea of a similar slot for QPI platforms, but you're right, Intel had to be thinking about doing that.

It's nice to have a common interface that can be used between platforms which is why I think PCI-e will continue to be supported anyway (definitely because of the DoJ), but the option to move stuff off PCI-e and onto HT/QPI Links would be very nice.

Maybe Intel will do something to help spruce up future versions of PCI-e to make it more useful.

 

[lopri]

Intel boards can't have QPI sockets/slots and PCIe slots at the same time? And what would be the "win" for the consumers? Being able to choose as needed, or having to throw away expensive Intel Fiber NICs (PCIe-based) and purchase a new one based on QPI? At the same time limiting their choices to products based on QPI (which would be "licensed" no doubt)?

It's really getting old. I think it's time for me to take a break from the forums.

 

[StrangerGuy]

I predict the GPU portion will take up even more die space than the CPU portion in the following years. Why? CPU performance is already a non-issue for gaming now let alone your usual MS Office, so that only leaves rendering and encoding which can realistically benefit from more cores/clocks. However, both are also massively parallel which lends themselves very suited for GPU processing. With Intel and AMD fully embracing GPU on CPU the switch will be inevitable.

I dare say for the consumer side the multicore era will soon draw to a close; probably at around 8 to 12 cores for the top-end and after that, replaced paradoxically by a GPU-dominated architecture.

 

[DominionSeraph]

PC gaming has been dead for years. What's the issue?

 

[DrMrLordX]

Intel boards can't have QPI sockets/slots and PCIe slots at the same time?

That depends on the wording and interpretation of the ruling that now binds Intel. If they introduced QPI slots and vendors liked the idea of moving their product lineup to qpi cards, then you'd have a situation where vendors would probably move most of their upcoming cards to qpi and partially (if not fully) abandon PCI-e. I'm sure they'd release PCI-e equivalents, but you'd probably see the same card with a different connector and some kind of bridge hardware on the card. PCI-e slots might start to decrease in number, if not disappear, from motherboards.

Nvidia/JHH and/or someone else could then argue that Intel is effectively killing the PCI-e standard by rendering it completely obsolete. What AMD would do is anyone's guess, but there's always HTX, and they could move their discrete GPU lineup to that slot if necessary.

In other words, if Intel does not actively continue to work on developing the PCI-e standard (as it has in the past) as their preferred interconnect for discrete video/hpc cards, it could be argued that they would be in violation of the FTC's ruling.

And what would be the "win" for the consumers?

Being able to hook up a video/HPC card to a QPI slot and then offload fp-intensive calculations to the card with reasonable turnaround times.

Being able to choose as needed

Right now you can't choose an interconnect for discrete graphics/hpc cards that makes them useful for fine-grained applications, as Idontcare indicated.

, or having to throw away expensive Intel Fiber NICs (PCIe-based) and purchase a new one based on QPI? At the same time limiting their choices to products based on QPI (which would be "licensed" no doubt)?

Again, I think Intel will continue to support PCI-e regardless of whether or not the gub'ment forces them to do so, in much the same fashion that PCI was supported for years. I think it is the card vendors that would abandon PCI-e in favor of QPI/HT if they had the chance, and if that happened, motherboard vendors would put less of them (and/or slower ones) on boards ASAP.

It's really getting old. I think it's time for me to take a break from the forums.

What's getting old? This thread doesn't seem particularly contentious at all . . . certainly not by this point.

PC gaming has been dead for years. What's the issue?

Netcraft confirms it.

 

[mechwarrior1989]

I don't really understand the hate for the integrated GPU.

The integrated GPU is not meant to replace dedicated GPU's, disabling the GPU will prevent it from using any additional power and thus prevent it from producing any additional heat. The fact that these are produced in mass quantities means that the integrated GPU costs next to nothing to add to the CPU. You definitely will not be able to get additional processing cores in the same space as the gpu is taking up.

Think about motherboards. The cheapest motherboards are all mATX motherboards with integrated graphics. Higher cost motherboards all tend to be ATX motherboards without integrated graphics. Yet you'll gladly get one of those higher end motherboards for your new gaming computer despite the fact that it SHOULD come with some video ports on the back since it's just as capable of handling video as the mATX motherboard. Yet no one is concerned about that. So why should you be concerned about the negligible cost of an integrated GPU in a CPU?

 

[deimos3428]

I think a good fusion would be moving the GPU *NEAR* the CPU, maybe even on-die, but with a huge amount of super-fast RAM near it. We need less than 20 nm to be able to move huge GPUs on die with CPU however. And I'm not even sure that will ever compete with discrete solutions.

An on-die CPU could help discrete GPU performance quite a bit, actually...there's no reason this can't work on both sides of the bus.