AMD talks Barcelona

[Viditor]

News.com article

A bit more info and predictions from AMD...

"We expect across a wide variety of workloads for Barcelona to outperform Clovertown by 40 percent," Allen said. The quad-core chip also will outperform AMD's current dual-core Opterons on "floating point" mathematical calculations by a factor of 3.6 at the same clock rate, he said.

It's AMD's first chip with a built-in level-three cache

AMD redesigned the Barcelona core, marking the biggest changes since the company made its 2003 transition from its 32-bit Athlon chips to the current 64-bit lineup

On FP: At a given clock frequency, a Barcelona core outperforms a current Opteron core by a factor of 1.8

On quad core speeds: Barcelona's clock frequency will be lower than that for the company's dual-core chips

 

[Markfw]

Cool. I hope that is before H2, not after, but either way, thats looks promising.

 

[Sylvanas]

Interesting, start Q2 would be great, seeing as thats when I plan to make my next purchase

 

[hectorsm]

I wonder if it's 10% greater performance per core (10X4=40%) than clovertown. If so, then in reality it is just about 10% more efficiency per core. Still...very good performance if it turns out to be true.

 

[Viditor]

Originally posted by: hectorsm
I wonder if it's 10% per core (10X4=40%). If so, then in reality it is just about 10% more efficiency per core and Core2. Still...very good performance if it turns out to be true.

Possible, but they are calling for an 80% improvement over current Opteron cores (per core at the same clockspeed) in FP.
That said, I wouldn't expect near as much improvement in integer performance...

 

[Nemesis 1]

? A faster floating-point engine performs mathematical calculations--long an Opteron strong suit, though not as important a part of the chip as that for integer operations. At a given clock frequency, a Barcelona core outperforms a current Opteron core by a factor of 1.8. By going quad-core, a Barcelona chip overall will provide a boost factor of 3.6, Allen says
Not all things are better, though. Specifically, Barcelona's clock frequency will be lower than that for the company's dual-core chips. That's a common situation because quad-core chips require more circuitry, and more circuitry means more power consumption and waste heat, unless the chips run slower

Ya Its not looking that rosey. Lower clock rates. When K8l is only 1.8 faster at the same clock. Than factor in that clock rate . were looking at maybe only 1.4 than amd's X2 @ 2.8. kinda looks like it will only be,< Intels 3.0 GHz processor

 

[Duvie]

I need to see hard numbers and not marketing hype.....

I really interested in seeing the 3 levels of cache as well.

 

[Viditor]

Originally posted by: Nemesis 1
? A faster floating-point engine performs mathematical calculations--long an Opteron strong suit, though not as important a part of the chip as that for integer operations. At a given clock frequency, a Barcelona core outperforms a current Opteron core by a factor of 1.8. By going quad-core, a Barcelona chip overall will provide a boost factor of 3.6, Allen says
Not all things are better, though. Specifically, Barcelona's clock frequency will be lower than that for the company's dual-core chips. That's a common situation because quad-core chips require more circuitry, and more circuitry means more power consumption and waste heat, unless the chips run slower

Ya Its not looking that rosey. Lower clock rates. When K8l is only 1.8 faster at the same clock. Than factor in that clock rate . were looking at maybe only 1.4 than amd's X2 @ 2.8. kinda looks like it will only be,< Intels 3.0 GHz processor

They are comparing dual core and quad core clockspeeds...that's why it's lower (just as it will be for Intel).
It's not K8L, it's K10...and a factor of 1.8 faster is 80% faster. That seems pretty substantial to me...

 

[Hard Ball]

Originally posted by: Sylvanas
Interesting, start Q2 would be great, seeing as thats when I plan to make my next purchase

Highly unlikely; the last I have heard (I know people who have worked on the project), the server parts will come around 2nd week of June 07 or so; with desktop parts to follow sometime later. They could possibly pull it forward, but it would be difficult to see how they can pull that into April.

 

[Viditor]

Originally posted by: Duvie
I need to see hard numbers and not marketing hype.....

I really interested in seeing the 3 levels of cache as well.

I agree...I think we both said the exact same thing when the C2D hype came out a year ago.
Nonetheless, it's encouraging to see that at least the hype matches the expectations most of us had (based on the design notes we've seen).

 

[Viditor]

Originally posted by: Hard Ball

Originally posted by: Sylvanas
Interesting, start Q2 would be great, seeing as thats when I plan to make my next purchase

Highly unlikely; the last I have heard (I know people who have worked on the project), the server parts will come around 2nd week of June 07 or so; with desktop parts to follow sometime later. They could possibly pull it forward, but it would be difficult to see how they can pull that into April.

Agreed...and I've been hearing those release dates as well.

 

[Nemesis 1]

Not really when you factor in the lower clockspeeds. Its only 1.8 faster at same clock.

I doubt we will see K8L at 2.5 ghz for quit some time after its first release.

Got a link where AMD says its K10

 

[Duvie]

Here is the problems I see......


? It's AMD's first chip with a built-in level-three cache. Cache memory can respond faster than main memory, and Intel has relied on large amounts of cache to improve its processors' performance. Each Barcelona core has its own 64-kilobyte first-level cache and 512KB second-level cache, and the four cores together share a 2MB third-level cache, though AMD has said that size can be increased.


The L2 cache is too small...plus L3 cache is likely not going to run at same speed as the L2 cache so the paltry 2mb the 4 cores share is not an upgrade to what Intel has now.

I was hoping for 1mb per each core and 4mb of L3 to share.

I can tell you right now even this new chip may struggle with the new Folding at Home units I am seeing that want 2mb of cache minimum to really speed up performance. 1mb doesn't cut it and takea HUGE i mean HUGE hit.


? A faster floating-point engine performs mathematical calculations--long an Opteron strong suit, though not as important a part of the chip as that for integer operations. At a given clock frequency, a Barcelona core outperforms a current Opteron core by a factor of 1.8. By going quad-core, a Barcelona chip overall will provide a boost factor of 3.6, Allen said.

In what app did they run to come up with this?

2 things here.

1) I thought there are still better at FPU now and they get their asses handed to them in the test I run. So what apps would I see this in, becasue I dont see it now.

2) 1.8x factor versus same speed dual core now but then they admit they will run slower clocks. So what are we going to expect? 2.8ghz clock now with an opteron but maybe the quads come out at top end near 2ghz or 40% slower?



seems like a lot of marketing hype. I hope they dont all base it on those spec scores, because those things almost never translate into things us desktop users see. I dont know the name of the tes but they usually run it on the servers.



If they are better I will get some. But like the C2D I waited until I saw it was a clear winner

 

[Nemesis 1]

Originally posted by: Viditor

Originally posted by: Duvie
I need to see hard numbers and not marketing hype.....

I really interested in seeing the 3 levels of cache as well.

I agree...I think we both said the exact same thing when the C2D hype came out a year ago.
Nonetheless, it's encouraging to see that at least the hype matches the expectations most of us had (based on the design notes we've seen).

No comparsion to last years C2D hype. Intel showed a working C2D last year. Still many wouldn't except the facts and called it hype

 

[Hard Ball]

Originally posted by: Nemesis 1
? A faster floating-point engine performs mathematical calculations--long an Opteron strong suit, though not as important a part of the chip as that for integer operations. At a given clock frequency, a Barcelona core outperforms a current Opteron core by a factor of 1.8. By going quad-core, a Barcelona chip overall will provide a boost factor of 3.6, Allen says
Not all things are better, though. Specifically, Barcelona's clock frequency will be lower than that for the company's dual-core chips. That's a common situation because quad-core chips require more circuitry, and more circuitry means more power consumption and waste heat, unless the chips run slower

Ya Its not looking that rosey. Lower clock rates. When K8l is only 1.8 faster at the same clock. Than factor in that clock rate . were looking at maybe only 1.4 than amd's X2 @ 2.8. kinda looks like it will only be,< Intels 3.0 GHz processor

I don't want to say too much. But keep in mind, the clock difference referenced in the article refers to chips within the same thermal envelope, which currently for rev F opterons stands at 68W(2.4GHz top), 95W(2.6GHz top), and 120W(2.8GHz+). For most of the current Socket F infrastructure, it's pegged at 95W. That should give you a more accurate calculation for upgrades on each type of infrasctructure with differing VRM standard.

 

[imported_Questar]

I guess they need something to prop up margins.

Half a billion dollar loss - seems the one thing AMD is consistently good at is wiping out years of profit in one quarter.

 

[Viditor]

Originally posted by: Nemesis 1
Not really when you factor in the lower clockspeeds. Its only 1.8 faster at same clock.

I doubt we will see K8L at 2.5 ghz for quit some time after its first release.

Got a link where AMD says its K10

1.8 faster = 80% faster at the same clockspeed...
Currently the fastest dual core Opteron is 2.8 GHz. If the quad core maxes out at 2.6 GHz at launch (likely), then you have a clockspeed difference of 8%.
Are you saying that an 80% increase is less than an 8% clockspeed difference???

As to K8L, you need to do some research...the term K8L was invented by The Inquirer and has never once been used by AMD.

 

[TuxDave]

Originally posted by: Viditor

Originally posted by: Nemesis 1
Not really when you factor in the lower clockspeeds. Its only 1.8 faster at same clock.

I doubt we will see K8L at 2.5 ghz for quit some time after its first release.

Got a link where AMD says its K10

1.8 faster = 80% faster at the same clockspeed...
Currently the fastest dual core Opteron is 2.8 GHz. If the quad core maxes out at 2.6 GHz at launch (likely), then you have a clockspeed difference of 8%.
Are you saying that an 80% increase is less than an 8% clockspeed difference???

As to K8L, you need to do some research...the term K8L was invented by The Inquirer and has never once been used by AMD.

Actually from what I heard, K8L was a name Intel dubbed on AMD's next generation processor for their 'expected performance competition' analysis. Then it must've leaked to the Inquirer (as all Intel secrets have) and then adoped by the geek community.

 

[Viditor]

Originally posted by: Nemesis 1

Originally posted by: Viditor

Originally posted by: Duvie
I need to see hard numbers and not marketing hype.....

I really interested in seeing the 3 levels of cache as well.

I agree...I think we both said the exact same thing when the C2D hype came out a year ago.
Nonetheless, it's encouraging to see that at least the hype matches the expectations most of us had (based on the design notes we've seen).

No comparsion to last years C2D hype. Intel showed a working C2D last year. Still many wouldn't except the facts and called it hype

And of course AMD will show a working K10, probably before Q2.
Intel began the hype in December/January, and showed the first chip in April/May...the time between those periods is where we are now for AMD.

 

[Nemesis 1]

Ya I know where K8l name came from and it wasn't the Inquirer. As for K8l being released at 2.6 ghz your dreaming. Duvie I believe was correct in what he believes it will be released at. That would hold true with AMD past releases.

Also keep in mind that. These first K8l's will be server. We won't see desktop processors till late in the 3rd or 4 qt.

Intel will have Wolfdales and Yorkfields out by than. With better sse4 instructions and High K with metal gates with the 4 core models at 3.5 ghz and 12mb. of shared cache between all 4 cores. The wolfdale is what I want clocked @ 4ghz with 6mb of shared cache.

I am sure Duvie will be eyeballing the Yorkfields.

Intel showed C2D at spring IDF

Intel Developer Forum
March 7-9, 2006 | San Francisco | Moscone Center West

 

[imported_Questar]

As to K8L, you need to do some research...the term K8L was invented by The Inquirer and has never once been used by AMD.

Actually AMD has never said this is K10.

K10 was supposed to have SMT, "huge" caches, and run at 10Ghz.

In this interview, AMD themselves call this K8L:
http://www.digitimes.com/news/a20060628VL201.html

 

[Nemesis 1]

Originally posted by: Viditor

Originally posted by: Nemesis 1

Originally posted by: Viditor

Originally posted by: Duvie
I need to see hard numbers and not marketing hype.....

I really interested in seeing the 3 levels of cache as well.

I agree...I think we both said the exact same thing when the C2D hype came out a year ago.
Nonetheless, it's encouraging to see that at least the hype matches the expectations most of us had (based on the design notes we've seen).

No comparsion to last years C2D hype. Intel showed a working C2D last year. Still many wouldn't except the facts and called it hype

And of course AMD will show a working K10, probably before Q2.
Intel began the hype in December/January, and showed the first chip in April/May...the time between those periods is where we are now for AMD.

Intel showed the c2d at spring idf .

Intel Developer Forum
March 7-9, 2006 | San Francisco | Moscone Center West

I can't help but notice you always overstate AMD and understate Intel. Why is that?

 

[Hard Ball]

Originally posted by: Nemesis 1
Ya I know where K8l name came from and it wasn't the Inquirer. As for K8l being released at 2.6 ghz your dreaming. Duvie I believe was correct in what he believes it will be released at. That would hold true with AMD past releases.

Also keep in mind that. These first K8l's will be server. We won't see desktop processors till late in the 3rd or 4 qt.

Intel will have Wolfdales and Yorkfields out by than. With better sse4 instructions and High K with metal gates with the 4 core models at 3.5 ghz and 12mb. of shared cache between all 4 cores. The wolfdale is what I want clocked @ 4ghz with 6mb of shared cache.

I am sure Duvie will be eyeballing the Yorkfields.

It seems obvious that you are not basing any of your comments on anything more than your own speculation.

 

[Nemesis 1]

Originally posted by: Hard Ball

Originally posted by: Nemesis 1
Ya I know where K8l name came from and it wasn't the Inquirer. As for K8l being released at 2.6 ghz your dreaming. Duvie I believe was correct in what he believes it will be released at. That would hold true with AMD past releases.

Also keep in mind that. These first K8l's will be server. We won't see desktop processors till late in the 3rd or 4 qt.

Intel will have Wolfdales and Yorkfields out by than. With better sse4 instructions and High K with metal gates with the 4 core models at 3.5 ghz and 12mb. of shared cache between all 4 cores. The wolfdale is what I want clocked @ 4ghz with 6mb of shared cache.

I am sure Duvie will be eyeballing the Yorkfields.

It seems obvious that you are not basing any of your comments on anything more than your own speculation.

On K8l I am going by what AMD has done in the past.

As for Intel going by what articles say.

http://www.xbitlabs.com/news/cpu/display/20060930234444.html

 

[Duvie]

Originally posted by: Hard Ball

Originally posted by: Nemesis 1
Ya I know where K8l name came from and it wasn't the Inquirer. As for K8l being released at 2.6 ghz your dreaming. Duvie I believe was correct in what he believes it will be released at. That would hold true with AMD past releases.

Also keep in mind that. These first K8l's will be server. We won't see desktop processors till late in the 3rd or 4 qt.

Intel will have Wolfdales and Yorkfields out by than. With better sse4 instructions and High K with metal gates with the 4 core models at 3.5 ghz and 12mb. of shared cache between all 4 cores. The wolfdale is what I want clocked @ 4ghz with 6mb of shared cache.

I am sure Duvie will be eyeballing the Yorkfields.

It seems obvious that you are not basing any of your comments on anything more than your own speculation.

He is right about me eyeballing the Yorkfields!!!