[Xelion]

You know they are reaching the limit on how small they can make wafers. I checked out ATI's Xpress 1100 series and it does seem promising, but the clock speed could be higher. As for it saying that it'll be able to run Windows Vista in full hardware acceleration mode only means that it will, it doesn't say how fast or how long loading times will be. Your not going to run some commercial applications smoothly with an integrated chip either. Another point worth mentioning is that on most boards the northbridge and southbridge already have heat sinks, soon they'll need fans, its just a matter of time for integrated chips.

But that's no use at all if data flows to and from it at a quarter of the speed.

Where did you get that from? The only limit for speed between the graphics card and the cpu is the AGP/PCI-E speed(transfer rate), since I'd imagine that the speed from the northbridge, the FSB and the memory are faster.

[Nuke]

the smallest theoretical size you can cut a chip is to a point where wires on the chip are one or two atoms thick. it will take many advances in nanorobotics to acheace that though, but its theoretically possible. that wall is still a ways away. the wall were at now is the 3 ghz barrier. increasing the clock is pointless if you cant feed the chip data fast enough. bus technolody needs to be improved. all the speed boosting advances lately have been in improving the flow of data to the core. its why using multiple processors doesnt increase the performance in perportion to the number of cpus. they bipass this limitation with dual cores, by transfering data between them at the chips clock rate and by sharing cache. still its not a 2 x increase. i look forward to the high end quad core chip intel is working on, but if its a 3x speed increase over a single core then it will be a good thing. im curious how far they can push multiple cores before they hit another wall.

[Grey Wolf]

OK, let's take a look at what you've just said.

1. The 3GHz barrier. We've passed that barrier.
2. Why increasing the clock speed is useless is not because of the bus, but rather because to increase the clock speed they increase the length of the pipeline, making brand prediction errors more costly and lowering the IPC (Instructions Per Clock cycle) of the CPU.
3. Increasing the bus is a concern, but more for transferring data to and from the RAM, not for transferring data between the processors.
4. You fail at microelectronics. It's not nanorobotics that are creating a wall to decreasing feature size on chips, it's things like pattern transfer, index of refraction, etc.

In conclusion, use real facts, not ones you made up.

[Nuke]

facts have a tendancy to be misleading

[Descenterace]

But that's no use at all if data flows to and from it at a quarter of the speed.

Where did you get that from? The only limit for speed between the graphics card and the cpu is the AGP/PCI-E speed(transfer rate), since I'd imagine that the speed from the northbridge, the FSB and the memory are faster.

Actually. it was a general comment on the tendancy of gimmickery to ignore the real bottlenecks, but now that you mention it... what's the bandwidth of PCI-E 4x? Memory bandwidth is something like 3-4GB/sec these days. Then we have to consider latency; it's no good shifting 16GB/sec if it's taking so long the CPU's already working on the next frame by the time the PPU is finished with the current one.

So for the PPU to be worthwhile, it must not only be able to do the calculations bust must also be connected to a suitable bus. I'm guessing PCI-E 4x is adequate; graphics cards need to shift a lot of data and they always worked fine on AGP.
Even so, the CPU's bottleneck these days is memory speeds. That is going to be worse for a device on the PCI-E bus. To make the PPU worthwhile, it'd need a substantial amount of on-board memory for caching objects, but since the CPU needs access to those objects as well that stuff still needs writing back to memory.

I can see the PPU being useful when the CPU is close to maximum load, but otherwise it's just adding latency to the loop.

[Fury]

The answer to high bandwidth and low latency needs is HTX. Many major players are supporting HTX, so  Intel will either have to license HyperTransport including HTX or develop their Common Systems Interconnect (CSI) to include 100% compatible connector similar to HTX. But Intel's CSI is still in development and will probably be launched sometime 2008.

[ZmaN]

Theres an Anaandtech article for this too...

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2807&p=1

Honestly, Im totally lost.  I guess I can see why AMD is doing this.

In short, AMD is losing money because of Intel's Core Duo 2, and AMD is running out of resources, so they decided to pur 5.4 billion dollars into buying ATI and theyre hoping it will all pay off when "GPUs and CPUs become one".  I dont see that happening for at least 5 years, maybe 10.  Thats a total waste of money, time and investment.

Intel seems totally fine on this, they have market share, and enouch cash to continue, and with the Core 2 Duo just released, and with all the stuff Dell has bought from them, theyre good so far.  (I did hear that Dell was contemplating buying AMD parts because Intel wasnt delivering for them, but thats changed).

Nvidia is im assuming at this point probably going to be making more for Intel, as AMD is acquiring ATI.  Apparently there is supposed to be an evening out of chipset sales per company sometime around, from what I can see in the article.

ATI is all game for this i guess.  I think that they are going to lose all ties with Intel because AMD is (or at least was) prime competition for Intel.  If ATI joins up with AMD, I dont think intel would want and ATI chips OR support, since that would be like feeding directly to your enemy (AMD).

Man times have changed...  All I know is I'm totally happy with Intel's performance lately and if they end up working more with Nvidia then they have with ATI, im game.  Though I do favor ATI cards, I was never really an AMD fan and now that Intel has control of the road for at least 5 years or so and AMD is probably not going to gain much more control, Nvidia working with Intel would probably work better for both companies, as both are top notch, compared to their counterpart companies battling them.

I'll still buy ATI Cards.

Someone said that ATI is going to be renamed?  EWWW...

AMD Radeon?  That sounds horrible..

[vyper]

If you're buying a card based on it's name, you deserve to get shafted.

[ZmaN]

If you're buying a card based on it's name, you deserve to get shafted.

no i didnt say that.

I meant that I like ATI cars better than Nvidias.

but the name AMD Radeon is lame...

[vyper]

All I can say is, vroom?

[Fury]

AMD is strongly considering open-sourcing at least a functional subset of ATI’s graphics drivers. It’s time for X Window System, OpenGL, and client virtualization for which ATI binary drivers aren’t available to escape the ghetto of the 1980s-era framebuffer. And what a boon for PR. If AMD’s graphics cards were the only ones with open device drivers, it might affect a buying decision or two.

http://www.infoworld.com/article/06/08/02/32OPcurve_1.html

[aqws]

http://lwn.net/Articles/195061/

Boy, this merger must be fun.

[neoterran]

This isn't a merger. AMD bought ATI. That's not the same thing as two companies merging, FYI.