AMD - Bulldozer Chitchat

[CAT-THE-FIFTH]

I'm confused. Nvidia were trounced by ATI

I edited my last post as I think the words I said were unfair.

However,remember that Fermi was always just a few weeks away for many months according to Nvidia.

I could be entirely and utterly wrong here but what if Intel is doing the same thing so they can stall sales of the competition? The roadmap says 1H 2012 not early 2012 like mentioned earlier.

[scaryjim]

I take CATs point - Intel knows Bulldozer is just around the corner, so they "announce" that Sandy Bridge E desktop processors are going to be launched in a few weeks and will be better than Bulldozer so everyone should wait before buying new computers. Then a few weeks down the line they "announce" that the chips will be delayed by a week or two - but you've all waited this long, so you might as well wait a bit longer.... and so-on ad infinitum. People who would otherwise have bought Bulldozer at launch are now two months down the line and still haven't bought Bulldozer...

[watercooled]

I don't think there's anything unfair about what you've said, I've seen lots of companies do it over the years and that's what I was implying in my post.

[Terbinator]

I just want new stuff, even though SNB has only recently came out i need my tech appetite feeding

[shakerist]

Im really confused atm so does my M4a89GTD PRO support all am3+ processors or only a few crappy ones ?

[scaryjim]

If it has an AM3+ socket it'll support all AM3+ processors. If it has an AM3 socket, it won't support any of them. I believe the AM3+ are black sockets, and the AM3 are white, but check what it actually says on the socket before making any snap purchases

[CAT-THE-FIFTH]

This is interesting news:

http://forums.hexus.net/mobile-devic...ransistor.html

From the article:
"Drew also pointed out that ARM has already announced test chips at 22 and 20nm already, with foundry partners TSMC and GlobalFoundries also working on those processes, and that IBM is already working on 14nm"

[watercooled]

Yeah as we were discussing earlier it's just a die shrink. You don't have to think too far back to remember the hype over 32nm and how it was going to be so incredibly power efficient and allow much higher clocks. I mean yes, it does both of those things to a degree like the die shrinks before it, but it's nothing world-changing.

[DanceswithUnix]

Yeah as we were discussing earlier it's just a die shrink. You don't have to think too far back to remember the hype over 32nm and how it was going to be so incredibly power efficient and allow much higher clocks. I mean yes, it does both of those things to a degree like the die shrinks before it, but it's nothing world-changing.

The last couple of process shrinks at Intel have fixed problems they had before in controlling their leakage. They went from a leaky process plus dodgy P4 cpu design to a fixed process and a good cpu design. Not earth shattering, but I think they did better than put themselves back on track there.

This "3d" transistor is something that has to happen eventually, but it looks like Intel are first with it. This is one of those things that promise to be a big win. Remember when AMD went to copper interconnect, it was like they got 2 shrinks at once with their performance bump. This could be another moment like that, so no it won't herald world peace or an end to global warming, but it will be nice.

For AMD, their SOI was a step towards solving the same leakage problems, so perhaps the pressure isn't on them yet, but they will lose some advantage they had there.

[Max Tractor]

Im really confused atm so does my M4a89GTD PRO support all am3+ processors or only a few crappy ones ?

Asus http://event.asus.com/2011/mb/AM3_PLUS_Ready/


Now it is all very confusing as some posts suggest, the info I have gleamed from support debate is as folows, the asus boards as per the link will support because the extra pins that am3+ will use are already powered in the current sockets of the motherboards mentioned, the black socket motherboards being released on current gen chipsets will also support am3+, AMD have however said they will only be supporting am3+ on am3+ motherboards. This debate has raged for sometime now, any other info from anybody please input


cheers

[CAT-THE-FIFTH]

The last couple of process shrinks at Intel have fixed problems they had before in controlling their leakage. They went from a leaky process plus dodgy P4 cpu design to a fixed process and a good cpu design. Not earth shattering, but I think they did better than put themselves back on track there.

This "3d" transistor is something that has to happen eventually, but it looks like Intel are first with it. This is one of those things that promise to be a big win. Remember when AMD went to copper interconnect, it was like they got 2 shrinks at once with their performance bump. This could be another moment like that, so no it won't herald world peace or an end to global warming, but it will be nice.

For AMD, their SOI was a step towards solving the same leakage problems, so perhaps the pressure isn't on them yet, but they will lose some advantage they had there.

Another factor to consider is that AMD is looking to decrease the size of their cores. The Bulldozer module and Bobcat are examples of this and IIRC it was mentioned somewhere that this was partially to offset the advantage Intel has in transistor size.

[CAT-THE-FIFTH]

I was on another forum and someone noticed something interesting about the claims Intel are making.

The following image is from the Anandtech article on the new transistors:

http://www.anandtech.com/show/4313/i...ing-in-2h-2011



At higher voltage the improvement over a 32NM planar transistor starts to diminish. AFAIK,the current Sandy Bridge processors have a VID over 1V. The planar transistors Intel use are produced on a 32NM bulk process.

It would be interesting to see the advantage when compared to planar transistors produced on a 32NM SOI process.

[DanceswithUnix]

That is an interesting observation, though if 22nm goes pop at 1.1v then it is all a bit mute.

I remember the days when CPUs ran off 5v, and all this were fields mumble mumble...

That graph doesn't have the other arrows on it, saying you can get the same performance for 0.2v less, which I think is the most impressive one.

[CAT-THE-FIFTH]

That is an interesting observation, though if 22nm goes pop at 1.1v then it is all a bit mute.

I remember the days when CPUs ran off 5v, and all this were fields mumble mumble...

Forums are awash with people saying AMD is screwed but they cannot compare the 32NM process AMD is using since it is different.

AMD uses a 32NM SOI process for Llano and Bulldozer. AFAIK,the improvement over such a process with the 22NM finfets is probably going to be less when compared to the 32NM bulk process Intel uses.

There is also no graph comparing 22NM planar transistors with 22NM finfets either.

That graph doesn't have the other arrows on it, saying you can get the same performance for 0.2v less, which I think is the most impressive one.

The 18% decrease in gate delay is probably where this metric came from though.

However,you need to consider the voltage the transistors in the current CPUs run at. AFAIK,at load the current Sandy bridge Core i5 and Core i7 processors have a VID of around 1.1V to 1.2V AFAIK. Unless the transistors internally run at a lower voltage you need to extend the graph some more past 1.0V to get gate delay for the 32NM transistors Intel are using in one of their current processors.

No doubt there is an improvement but I suspect it is not as much as Intel is making it out to be for a desktop processor.

I suspect the main improvement will down to transistor density and Intel can beef up the IGP more and/or make the CPU smaller.

[watercooled]

These people are a bit daft, that's all there is to it - they have seen the early marketing, believed every word of it and added their own bits to the end as fact. As I was saying I heard all the same nonsense about the 32nm die shrink.

Edit: That's also a good reason why I like this forum.

[Kumagoro]

Is gate delay the only factor which determines switching speed in the cpu as a whole though.