OSDev.org

So, there we are: the i7 is here.

On Wikipedia
HotHardware review
sorry, but i'm in a hurry, so i let you to debate this.

This is some of a change from the quard core, but realy
optimizing windows would decrease the need for such "high" processors ..
but does it includes new instructions ?, that would be a good thing, some "new" thing ...

KMT dk

It includes the X2APIC and is basically (from what I gather) an Intel XEON with hyper-threading and Turbo Boost. Also, the Quick Path technology is an interesting addition as well.

I've already priced out an i7 custom machine.

kmtdk wrote:optimizing windows would decrease the need for such "high" processors ..

That's not the problem. More CPU (and GPU) power allows us to have more realistic games, which is the main reason hardware keeps developing at the pace it does.

When prices come down, I'll certainly be buying a Core i7 machine for myself. Custom built, of course.

JackScott wrote: When prices come down, I'll certainly be buying a Core i7 machine for myself. Custom built, of course.

Relative to the Core2Quad 9xxx series it is not much more pricey. It costs ~$300 for the i7 920 on NewEgg.com, and in comparison to previous computer builds is really not that much.

JackScott wrote:

kmtdk wrote:optimizing windows would decrease the need for such "high" processors ..

That's not the problem. More CPU (and GPU) power allows us to have more realistic games, which is the main reason hardware keeps developing at the pace it does.

Damn, and I thought it was the Porn industry that was driving such technological advancements. After all, they're widely credited with being one of the main driving forces behind the adoption of the good ol' DVD format, and are frequently early adopters of new technology.

Yes, I know all too much about this stuff.

Oooh, shiny processor... Must get mobile version...

In other news, I just received my copy of WoW: Wrath of the Lich King and I am installing it now.

Hi,

For OS developers, Core i7 is extremely interesting - for the first time it's possible to end up with NUMA, multi-core and hyperthreading in the same commodity system. In terms of scheduler design/optimization and power management it'll be a very interesting challenge. On top of that there's scalability issues to think of - 4 chips with 8 cores per chip and 2 logical CPUs per core can add up to a lot of lock contention and cache line bouncing...

JackScott wrote:

kmtdk wrote:optimizing windows would decrease the need for such "high" processors ..

That's not the problem. More CPU (and GPU) power allows us to have more realistic games, which is the main reason hardware keeps developing at the pace it does.

AFAIK the scheduler in Windows (Vista and all it's predecessors) treats all CPUs the same, which means in some situations it'll suck badly. For example, if there's 2 quad core chips and only 2 threads to run, then Vista could decide to run both threads on the same core and cause a "shared resources/hyperthreading" performance loss, or it could decide to run both threads on different cores in the same chip and miss out on the "I'm the only CPU in the chip that's running" turbo boost. A good scheduler could easily give 50% better performance than a Microsoft scheduler under certain loads...

Of course Intel have been mentioning hyperthreading optimizations in their manuals ever since Pentium 4 was released (8 years) and Microsoft have done nothing, and AMD have been NUMA for a long time (7 years) and Microsoft have done even more nothing.

Most games are still mostly single threaded. The Turbo Boost feature might help games a little, but teaching game programmers how to write software would help a lot more (even for current dual and quad core systems).

JackScott wrote:When prices come down, I'll certainly be buying a Core i7 machine for myself. Custom built, of course.

I'll be waiting at least until Intel release their multi-processor chips (first quarter of next year?). The CPUs they have released are "single chip systems only" because they lack addition QuickPath interconnects needed for 2-way and 4-way systems; and there aren't any motherboards for it yet either. To be honest, I'm tempted to spend the extra $$$ and go for a 4-way motherboard with eight core chips, but I'm not sure I can wait that long (or that it'll fit my budget when it arrives).


Cheers,

Brendan

I just purchased an i7 940 core and it should be here in 5 days. Once I grab a new rackmount case for it and shove it in, I'd be happy to give people access to the kvm so they can use it to test their OS's on.

Brendan wrote: Most games are still mostly single threaded. The Turbo Boost feature might help games a little, but teaching game programmers how to write software would help a lot more (even for current dual and quad core systems).

The problem with multithreading games is that pretty much everything a game needs to do happens on the same set of data structures.

Both the graphics and physics engines need to access the information on all the objects. Theoretically, you could run the physics on a different thread. However, even on multi core systems, it's been shown that the locking overhead makes everything run slower than just doing it all in one thread. Add to this the fact that physics engines are now starting to run on the GPU (nVIDIA has made PhysX run on CUDA on GeForce 8s and above, for example), and the graphics and physics are now contented arround the same hardware.

This is all why the actual complexity (Number of batches) of a scene goes up much slower than it's apparent complexity (Visual quality); graphics cards are getting better faster than processors. For years now, the main contention in games has been the CPU. No matter what you do, the CPU is contented arround the same set of data structures. Multiple threads operating on these structures just causes cache thrashing and lock contention. 95% of the advancement of graphics is tricks performed by shaders on the GPU.

Even on the PS3, where you have to think laterally to fully exploit the SPUs, their role is limited to mostly generating input to vertex and pixel shaders.

Anything more than dual core is really superflous outside a server, or perhaps a HTPC (For video decoding). And games would really prefer you just had a fast single core.

Anything more than dual core is really superflous

AI algorithms only need to modify local data during their decision cycles, so if you don't touch the world state in that time, you can dedicate a core to each unit.

And even in physics, there are few cases where all objects interact with all other objects, you can cluster event groups to sets where objects could possibly collide, and compute each set on a different core since they don't interact with one another.