AMD plans native 8-core cpu

[reviewhunter]

We were surprised to learn that despite the rumors going around, AMD's eight-core CPU is actually a native one. AMD can do native octa-core just as well as Intel can with its Nehalem, and the codename for this eight-core monster is Montreal. This comes from a highly ranked source at AMD.

So, Montreal is not just two Shanghai quad-cores stitched together, it's a native octa-core 45nm part. This makes sense for servers initially and eventually it will also launch for the ultra high-end gaming market.

Of course, this all happens in 2009.


Source: http://www.fudzilla.com/index....=view&id=6154&Itemid=1

 

[Arkaign]

All the 'native'/'true' *-Core arguments can go DIAF. In reality, it makes little to zero difference. The only thing that matters are tangible factors like :

Price / Performance
Scaling
Heat
Etc

For example, the Phenom is native quad, but it's a pile of useless shit at this point. I'd much rather have an Athlon X2, any C2D, or any C2Q.

 

[GuitarDaddy]

This "Core Wars" thing is getting pretty ridiculous, it's great for the server market but for desktop usage Intel and AMD will both have 8-core CPU's out long before any normal software even takes advantage of 4-cores

Kinda sad for those like me who don't do folding or video encoding. I have a strong dual core CPU running 3+ghz and I see nothing worthy of an upgrade in the next year or so.

 

[nullpointerus]

Native octa-core?

I'd be happy if they got native quad-core working. The TLB issues, platform immaturity, and overclocking issues have really cut into AMD's support within the enthusiast community.

I want a strong AMD!

 

[Extelleron]

Originally posted by: Arkaign
All the 'native'/'true' *-Core arguments can go DIAF. In reality, it makes little to zero difference. The only thing that matters are tangible factors like :

Price / Performance
Scaling
Heat
Etc

For example, the Phenom is native quad, but it's a pile of useless shit at this point. I'd much rather have an Athlon X2, any C2D, or any C2Q.

It doesn't make zero to no difference, there is a scaling advantage to having all your cores on one die and this has been shown by Anandtech (Phenom displayed around 7% better scaling than Intel C2Q).

Phenom is not a "pile of useless shit," it's a perfectly good processor in its price bracket. At $190, you can't beat the 9500 if the applications you use benefit from multi-core. In those apps the Phenom 9500 will beat the E8400 which costs more.

I don't think AMD's problems arise from Phenom being Quad-core on one die, I think the problem is AMD's 65nm process sucks. Even G2 Brisbane can't hit the same clocks F3 Windsor can hit on 90nm. Hopefully 45nm will be much better.

With their 45nm CPUs, AMD is going to have a much smaller die than Nehalem and they will likely position Tri-core/Quad-core w/o L3 against Nehalem dual-cores, and there's no doubt they will be able to compete in the mainstream market. Nehalem Quad and Octal-core CPUs will dominate the high-end sector, but that is a small part of the market.

 

[Arkaign]

Originally posted by: Extelleron

Originally posted by: Arkaign
All the 'native'/'true' *-Core arguments can go DIAF. In reality, it makes little to zero difference. The only thing that matters are tangible factors like :

Price / Performance
Scaling
Heat
Etc

For example, the Phenom is native quad, but it's a pile of useless shit at this point. I'd much rather have an Athlon X2, any C2D, or any C2Q.

It doesn't make zero to no difference, there is a scaling advantage to having all your cores on one die and this has been shown by Anandtech (Phenom displayed around 7% better scaling than Intel C2Q).

Phenom is not a "pile of useless shit," it's a perfectly good processor in its price bracket. At $190, you can't beat the 9500 if the applications you use benefit from multi-core. In those apps the Phenom 9500 will beat the E8400 which costs more.

I don't think AMD's problems arise from Phenom being Quad-core on one die, I think the problem is AMD's 65nm process sucks. Even G2 Brisbane can't hit the same clocks F3 Windsor can hit on 90nm. Hopefully 45nm will be much better.

With their 45nm CPUs, AMD is going to have a much smaller die than Nehalem and they will likely position Tri-core/Quad-core w/o L3 against Nehalem dual-cores, and there's no doubt they will be able to compete in the mainstream market. Nehalem Quad and Octal-core CPUs will dominate the high-end sector, but that is a small part of the market.

In what situation did the Phenom scale better than the C2Q? The data I've seen shows Phenom scaling worse than the X2, which was already running getting diminishing returns around the 3ghz range.

http://www.tomshardware.com/20...ng_compared/index.html

"Performance per core does not scale as well as with an Athlon 64 X2 core between 2.2 and 2.8 GHz. This means that the performance gains of Phenom at future clock speeds will not be as significant as they have been with Athlon 64 X2 in the past. Let me give you some numbers to give you a better feeling: Athlon 64 X2 wins in 18 of our benchmarks, while Phenom 9000 only scales better in four categories. I would also like to emphasize that we used Asus's BIOS version 0603, which does not include a fix to the Phenom's TLB bug. Hence Phenom runs without any performance limitations."

And this :

http://enthusiast.hardocp.com/...w2LCxoZW50aHVzaWFzdA==

Shows your idea that Phenom scales better than C2Q to be laughable. Look at the 2.3 to 3.0ghz Phenom results compared to the jump from 2.4 to 3.0 on the C2Q side.

At ~$190 price range, provided you can locate one, you can get a Microcenter Q6600, or an E8400 (slim availability admittedly at this time). Comparing the $190 9500 is also dumb because it's so slow.

AMD's best deals are still in the X2 market.

I never said Phenom's problems were because they were 'native' 4-core procs. The Phenom's problem is that it runs like shit. The async L3 that runs at a locked speed causes scaling to actually be horrible, and that means that unless they fix that issue, higher-speed Phenoms will become LESS competitive.

Phenom as it stands is almost unbuyable outside of brand loyalists. It's like the Pentium D of AMD. It serves no purpose other than to be a flaky, slow, late, mess.

 

[Nedder]

Originally posted by: GuitarDaddy
This "Core Wars" thing is getting pretty ridiculous, it's great for the server market but for desktop usage Intel and AMD will both have 8-core CPU's out long before any normal software even takes advantage of 4-cores

Kinda sad for those like me who don't do folding or video encoding. I have a strong dual core CPU running 3+ghz and I see nothing worthy of an upgrade in the next year or so.

I don't agree.

Encoding video like x264, my q6600 can do 6-8fps where my e4500 does 3-4fps. That's a huge reduction in time for the average movie (8 hours instead of 16).

Pro Tools LE - all the RTAS plugins use cpu and things like reverb and convolution gobble up cpus. Having 4 cores vs 2 lets me use more instances of things like Altiverb and Izotope trash/spectron.

Gaming - only a few games like SC support 4 cores but for those that don't you can play them while doing something like encoding video without any hiccups.

8 cores would be fantastic and I look forward to such a system.

 

[Extelleron]

Originally posted by: Arkaign

Originally posted by: Extelleron

Originally posted by: Arkaign
All the 'native'/'true' *-Core arguments can go DIAF. In reality, it makes little to zero difference. The only thing that matters are tangible factors like :

Price / Performance
Scaling
Heat
Etc

For example, the Phenom is native quad, but it's a pile of useless shit at this point. I'd much rather have an Athlon X2, any C2D, or any C2Q.

It doesn't make zero to no difference, there is a scaling advantage to having all your cores on one die and this has been shown by Anandtech (Phenom displayed around 7% better scaling than Intel C2Q).

Phenom is not a "pile of useless shit," it's a perfectly good processor in its price bracket. At $190, you can't beat the 9500 if the applications you use benefit from multi-core. In those apps the Phenom 9500 will beat the E8400 which costs more.

I don't think AMD's problems arise from Phenom being Quad-core on one die, I think the problem is AMD's 65nm process sucks. Even G2 Brisbane can't hit the same clocks F3 Windsor can hit on 90nm. Hopefully 45nm will be much better.

With their 45nm CPUs, AMD is going to have a much smaller die than Nehalem and they will likely position Tri-core/Quad-core w/o L3 against Nehalem dual-cores, and there's no doubt they will be able to compete in the mainstream market. Nehalem Quad and Octal-core CPUs will dominate the high-end sector, but that is a small part of the market.

In what situation did the Phenom scale better than the C2Q? The data I've seen shows Phenom scaling worse than the X2, which was already running getting diminishing returns around the 3ghz range.

http://www.tomshardware.com/20...ng_compared/index.html

"Performance per core does not scale as well as with an Athlon 64 X2 core between 2.2 and 2.8 GHz. This means that the performance gains of Phenom at future clock speeds will not be as significant as they have been with Athlon 64 X2 in the past. Let me give you some numbers to give you a better feeling: Athlon 64 X2 wins in 18 of our benchmarks, while Phenom 9000 only scales better in four categories. I would also like to emphasize that we used Asus's BIOS version 0603, which does not include a fix to the Phenom's TLB bug. Hence Phenom runs without any performance limitations."

And this :

http://enthusiast.hardocp.com/...w2LCxoZW50aHVzaWFzdA==

Shows your idea that Phenom scales better than C2Q to be laughable. Look at the 2.3 to 3.0ghz Phenom results compared to the jump from 2.4 to 3.0 on the C2Q side.

At ~$190 price range, provided you can locate one, you can get a Microcenter Q6600, or an E8400 (slim availability admittedly at this time). Comparing the $190 9500 is also dumb because it's so slow.

AMD's best deals are still in the X2 market.

I never said Phenom's problems were because they were 'native' 4-core procs. The Phenom's problem is that it runs like shit. The async L3 that runs at a locked speed causes scaling to actually be horrible, and that means that unless they fix that issue, higher-speed Phenoms will become LESS competitive.

Phenom as it stands is almost unbuyable outside of brand loyalists. It's like the Pentium D of AMD. It serves no purpose other than to be a flaky, slow, late, mess.

That's not the kind of scaling I, or any one else, is talking about when they mention native vs non-native Quad. Phenom scales better from one to four cores than C2Q; read the Anandtech review.

The reason Phenom scales bad in their review is because they aren't increasing L3/NB speed as well. As AMD increases Phenom clocks, NB speed will increase as well. If you leave the NB @ 1.8GHz, then you should expect to see lower scaling.

Phenom isn't great at this point but it competes in the market it is in. Q6600 is not $190, @ Microcenter it will be around $210-220 after tax and that's if you have a Microcenter in your back yard. Otherwise, a new Q6600 will set you back $250. An E8400 will cost around the same if you can find one.

AMD will also be launching the Phenom 9100e, Quad-core @ 1.8GHz w/ 65W TDP, for a very low price ($130-140).

 

[heyheybooboo]

I've been waiting 10-12 years now for multithread parallelization to reach the desktop in any practical or significant manner ...

It's coming ... it's coming ... it's coming...

Someday never comes.

There will remain for the foreseeable future limited off-the-shelf software capable of running parallel threads across 2 cores, much less 4 (or 8, 16, 32...)

 

[Idontcare]

Originally posted by: reviewhunter
We were surprised to learn that despite the rumors going around, AMD's eight-core CPU is actually a native one. AMD can do native octa-core just as well as Intel can with its Nehalem, and the codename for this eight-core monster is Montreal. This comes from a highly ranked source at AMD.

So, Montreal is not just two Shanghai quad-cores stitched together, it's a native octa-core 45nm part. This makes sense for servers initially and eventually it will also launch for the ultra high-end gaming market.

Of course, this all happens in 2009.


Source: http://www.fudzilla.com/index....=view&id=6154&Itemid=1

I would love to drop four of these into a quad socket mobo. And then I'd like to set that computer on a desk next to 3 more just like it. Yes, that would make me happy indeed.

 

[Arkaign]

Originally posted by: Idontcare

Originally posted by: reviewhunter
We were surprised to learn that despite the rumors going around, AMD's eight-core CPU is actually a native one. AMD can do native octa-core just as well as Intel can with its Nehalem, and the codename for this eight-core monster is Montreal. This comes from a highly ranked source at AMD.

So, Montreal is not just two Shanghai quad-cores stitched together, it's a native octa-core 45nm part. This makes sense for servers initially and eventually it will also launch for the ultra high-end gaming market.

Of course, this all happens in 2009.


Source: http://www.fudzilla.com/index....=view&id=6154&Itemid=1

I would love to drop four of these into a quad socket mobo. And then I'd like to set that computer on a desk next to 3 more just like it. Yes, that would make me happy indeed.

128-core cluster FTW?

 

[Extelleron]

Originally posted by: heyheybooboo
I've been waiting 10-12 years now for multithread parallelization to reach the desktop in any practical or significant manner ...

It's coming ... it's coming ... it's coming...

Someday never comes.

There will remain for the foreseeable future limited off-the-shelf software capable of running parallel threads across 2 cores, much less 4 (or 8, 16, 32...)

Err... Just any about any application that demands performance already supports 2+ cores. The vast majority of games shipping today use 2 cores and a few are starting to take advantage of 4 cores.

If you're doing video/audio editing, 3D rendering, etc... then you need a multi-core CPU.

 

[Denithor]

Originally posted by: Extelleron
With their 45nm CPUs, AMD is going to have a much smaller die than Nehalem and they will likely position Tri-core/Quad-core w/o L3 against Nehalem dual-cores, and there's no doubt they will be able to compete in the mainstream market. Nehalem Quad and Octal-core CPUs will dominate the high-end sector, but that is a small part of the market.

...Huh? Excuse me? AMD's 45nm CPUs will have a much smaller die than Nehalem?

What kind of crack have you been smoking?

Intel Tick-Tock Roadmap

Penryn (out now) is the 45nm die shrink of Conroe. Nehalem will be produced on those same 45nm lines.

"Scratches head."

So AMD's 45nm process is smaller than Intel's 45nm process? :roll:

 

[Extelleron]

Originally posted by: Denithor

Originally posted by: Extelleron
With their 45nm CPUs, AMD is going to have a much smaller die than Nehalem and they will likely position Tri-core/Quad-core w/o L3 against Nehalem dual-cores, and there's no doubt they will be able to compete in the mainstream market. Nehalem Quad and Octal-core CPUs will dominate the high-end sector, but that is a small part of the market.

...Huh? Excuse me? AMD's 45nm CPUs will have a much smaller die than Nehalem?

What kind of crack have you been smoking?

Intel Tick-Tock Roadmap

Penryn (out now) is the 45nm die shrink of Conroe. Nehalem will be produced on those same 45nm lines.

"Scratches head."

So AMD's 45nm process is smaller than Intel's 45nm process? :roll:

Before you post such a heated reply, make sure you actually know what you are talking about. Read up, get educated, before you start lecturing other people.

You make it sound as if every chip manufactured on the same process will be the same size... I don't know where you are getting that idea. Look at AMD's Brisbane CPU... it's 65nm with 154.8M transistors and is 126mm^2 in die size. Intel's Conroe is 65nm with 291M transistors and is 143mm^2 in die size. Note that the reason Conroe has so many more transistors but is only slightly larger is that it has much more cache, which is made up of a lot of transistors but takes up minimal space on the die.

If you look at AMD's current Agena core, it is 285mm^2 in die size, virtually the same size as Intel's Kentsfield, which is 286mm^2 (143mm^2 * 2). Intel's Penryn shrink reduced the die size from 143mm^2 to 107mm^2, so Quad-core Yorkfield is 214mm^2 (107mm^2 * 2).

Nehalem is a new core and will be much more complex; the core has been built from the ground up and it also brings an integrated NB and memory controller. Estimates based on images of the Nehalem die & wafer put the Quad-core version at around 270-280mm^2.

Deneb is the 45nm shrink of Agena core and contains some improvements versus the 65nm core, just like Penryn improves upon 65nm Conroe. It will be smaller than Agena because of the 45nm process; I'd estimate it will be around 220-230mm^2. The Deneb core without the 6MB of L3 should be much smaller than that; probably 180-190mm^2.

Last I checked, 220-230mm^2 < 270-280mm^2.

Check your facts before posting next time and you won't make a fool of yourself.

 

[Cheex]

I haven't read the posts above mine but I will say this....

WTF?!!
They haven't even gotten native quad running good yet!!
WTF?!!

 

[heyheybooboo]

Originally posted by: Extelleron

Originally posted by: heyheybooboo
I've been waiting 10-12 years now for multithread parallelization to reach the desktop in any practical or significant manner ...

It's coming ... it's coming ... it's coming...

Someday never comes.

There will remain for the foreseeable future limited off-the-shelf software capable of running parallel threads across 2 cores, much less 4 (or 8, 16, 32...)

Err... Just any about any application that demands performance already supports 2+ cores. The vast majority of games shipping today use 2 cores and a few are starting to take advantage of 4 cores.

If you're doing video/audio editing, 3D rendering, etc... then you need a multi-core CPU.

You are not making any distinction whatsoever between load-balancing across cores and true parallelism of threads across cores.

 

[Arkaign]

Originally posted by: heyheybooboo

Originally posted by: Extelleron

Originally posted by: heyheybooboo
I've been waiting 10-12 years now for multithread parallelization to reach the desktop in any practical or significant manner ...

It's coming ... it's coming ... it's coming...

Someday never comes.

There will remain for the foreseeable future limited off-the-shelf software capable of running parallel threads across 2 cores, much less 4 (or 8, 16, 32...)

Err... Just any about any application that demands performance already supports 2+ cores. The vast majority of games shipping today use 2 cores and a few are starting to take advantage of 4 cores.

If you're doing video/audio editing, 3D rendering, etc... then you need a multi-core CPU.

You are not making any distinction whatsoever between load-balancing across cores and true parallelism of threads across cores.

No matter, it's been proven that Dual-Core makes huge gains in today's software and multitasking environments. It will only get more efficient with time.

 

[myocardia]

Originally posted by: Extelleron
Before you post such a heated reply, make sure you actually know what you are talking about. Read up, get educated, before you start lecturing other people.

You make it sound as if every chip manufactured on the same process will be the same size... I don't know where you are getting that idea. Look at AMD's Brisbane CPU... it's 65nm with 154.8M transistors and is 126mm^2 in die size. Intel's Conroe is 65nm with 291M transistors and is 143mm^2 in die size. Note that the reason Conroe has so many more transistors but is only slightly larger is that it has much more cache, which is made up of a lot of transistors but takes up minimal space on the die.

If you look at AMD's current Agena core, it is 285mm^2 in die size, virtually the same size as Intel's Kentsfield, which is 286mm^2 (143mm^2 * 2). Intel's Penryn shrink reduced the die size from 143mm^2 to 107mm^2, so Quad-core Yorkfield is 214mm^2 (107mm^2 * 2).

Nehalem is a new core and will be much more complex; the core has been built from the ground up and it also brings an integrated NB and memory controller. Estimates based on images of the Nehalem die & wafer put the Quad-core version at around 270-280mm^2.

Deneb is the 45nm shrink of Agena core and contains some improvements versus the 65nm core, just like Penryn improves upon 65nm Conroe. It will be smaller than Agena because of the 45nm process; I'd estimate it will be around 220-230mm^2. The Deneb core without the 6MB of L3 should be much smaller than that; probably 180-190mm^2.

Last I checked, 220-230mm^2 < 270-280mm^2.

Check your facts before posting next time and you won't make a fool of yourself.

Do you happen to have links to all of these "facts" you're seeming to pull out of thin air? BTW, here's something you should research, while you're researching: which company is able to produce cache on a much denser scale than the other? That fact alone will play a significant part in the total die size of each of the processors you mention above. You do realize that a Kentsfield has ~twice as much cache as a Phenom, don't you?

 

[Extelleron]

Originally posted by: myocardia

Originally posted by: Extelleron
Before you post such a heated reply, make sure you actually know what you are talking about. Read up, get educated, before you start lecturing other people.

You make it sound as if every chip manufactured on the same process will be the same size... I don't know where you are getting that idea. Look at AMD's Brisbane CPU... it's 65nm with 154.8M transistors and is 126mm^2 in die size. Intel's Conroe is 65nm with 291M transistors and is 143mm^2 in die size. Note that the reason Conroe has so many more transistors but is only slightly larger is that it has much more cache, which is made up of a lot of transistors but takes up minimal space on the die.

If you look at AMD's current Agena core, it is 285mm^2 in die size, virtually the same size as Intel's Kentsfield, which is 286mm^2 (143mm^2 * 2). Intel's Penryn shrink reduced the die size from 143mm^2 to 107mm^2, so Quad-core Yorkfield is 214mm^2 (107mm^2 * 2).

Nehalem is a new core and will be much more complex; the core has been built from the ground up and it also brings an integrated NB and memory controller. Estimates based on images of the Nehalem die & wafer put the Quad-core version at around 270-280mm^2.

Deneb is the 45nm shrink of Agena core and contains some improvements versus the 65nm core, just like Penryn improves upon 65nm Conroe. It will be smaller than Agena because of the 45nm process; I'd estimate it will be around 220-230mm^2. The Deneb core without the 6MB of L3 should be much smaller than that; probably 180-190mm^2.

Last I checked, 220-230mm^2 < 270-280mm^2.

Check your facts before posting next time and you won't make a fool of yourself.

Do you happen to have links to all of these "facts" you're seeming to pull out of thin air? BTW, here's something you should research, while you're researching: which company is able to produce cache on a much denser scale than the other? That fact alone will play a significant part in the total die size of each of the processors you mention above. You do realize that a Kentsfield has ~twice as much cache as a Phenom, don't you?

Exactly what facts am I "pulling out of thin air"? Everything I said is common knowledge and well known except for my estimation of Shanghai's die size. The only thing I can say for sure is it will be smaller than Barcelona, so it will be <285mm^2. My estimate is around 230mm^2; I would bet that it will be within 5% of that.

As for Intel having lower cache density, that was true before, but not so much now: http://www.chip-architect.com/...19_Various_Images.html

With Barcelona, AMD is just about equal to Intel with 11.2-12.1mm^2/MB on the 65nm node. Brisbane's cache was larger than Intel's, but apparently AMD has made improvements since then. I should note that AMD's Barcelona is made up of L2 & L3; it is possible that L3 might be a bit denser than L2, although I don't think this would be drastic since both are the same SRAM.

I'm well aware of how much cache Kentsfield has versus Agena, and this is nothing new. AMD's CPUs, in recent times, have always been composed more of logic than cache. Intel has been the one to stick a huge amount of cache on a chip, partly because it has been capable of a smaller SRAM cell size and its CPUs rely on cache to perform well.. for the most part, AMD CPUs see little performance increase with more cache.

I'm not the one who needs to be doing the researching; some other people in this thread need to be getting their facts straight before posting.

 

[Idontcare]

http://aceshardware.freeforums...-shanghai-t405-15.html

Nehalem = 246mm^2

Shanghai = 243mm^2

Hans de Vries FTW.

 

[nyker96]

please no more native x-cores. It just ain't work ... I thought they already knew that, guess not!

 

[Extelleron]

Originally posted by: Idontcare
http://aceshardware.freeforums...-shanghai-t405-15.html

Nehalem = 246mm^2

Shanghai = 243mm^2

Hans de Vries FTW.

That contradicts what I had seen before regarding Nehalem die sizes... I had previously seen ~270mm^2 estimated.

If that is accurate, then AMD has seen some pretty bad SRAM scaling (63%+ of 65nm size) but good logic scaling (60% of 65nm size).

 

[dmens]

The 270mm2 estimate was based on a photo of an entire wafer, which is completely inadequate for figuring out die sizes. Inq really screwed that one up.

I also noticed the jackasses at amdzone making the same claim up to 290mm2, lol.

 

[Idontcare]

Originally posted by: Extelleron

Originally posted by: Idontcare
http://aceshardware.freeforums...-shanghai-t405-15.html

Nehalem = 246mm^2

Shanghai = 243mm^2

Hans de Vries FTW.

That contradicts what I had seen before regarding Nehalem die sizes... I had previously seen ~270mm^2 estimated.

If that is accurate, then AMD has seen some pretty bad SRAM scaling (63%+ of 65nm size) but good logic scaling (60% of 65nm size).

Checkout Hans post with the embedded graphics to back up the math. Hans is pretty much legend for doing the analysis of CPU die shots.

Per Hans, Nehalem L2$ is 7.1mm2/MB while Shanghai is 7.5mm2/MB...not the end of the world difference but yes Intel has better SRAM density.

Intel really takes off with their L3$ density...5.7mm2/MB for Nehalem versus 7.5mm2/MB for Shanghai.

 

[JavaMomma]

Originally posted by: heyheybooboo
I've been waiting 10-12 years now for multithread parallelization to reach the desktop in any practical or significant manner ...

It's coming ... it's coming ... it's coming...

Someday never comes.

There will remain for the foreseeable future limited off-the-shelf software capable of running parallel threads across 2 cores, much less 4 (or 8, 16, 32...)

This is very true, until now software developers have had little reason to write multi-threaded code other then for little trivial things, like say the UI run in a separate thread, this is because so few people had the hardware. Even now there are still few things that are worth the effort of multi threading with the current set of tools developers have. I mean most software is poorly written anyways, and if I'm given a piece of code which isn't performing well enough ... there are usually other easier things I can do first to speed it up.

In the last year I've written two routines that I wrote to take advantage of and scale with the number of CPUs and it was a lot of effort, it did result in a nice boost in performance on my dual core system, but still ... software developers need some easier tools ... I've been keeping an eye on Parallel Extensions for .NET
http://forums.microsoft.com/MS...?ForumID=1986&SiteID=1

BTW I write business software in C# VS2005, so very different from gaming or video editing software where those guys might write a lot of multi-threaded stuff, I don't know...