Sign in with
Sign up | Sign in
Your question

2,3Ghz Barcelona Vs. 3Ghz Clovertown

Last response: in CPUs
Share
February 22, 2007 2:48:47 PM

First of all, sheez this forum annoys me.. Takes like 2 minutes to get the page to show ! Feels like I'm taking crazypills waiting so long ..

To the case.

http://www.hkepc.com/bbs/itnews.php?tid=746541

So 2,3Ghz top bin for Barcelona. June mass production, if not delay.

But notice, only 2,6Ghz Barcelona scheduled for 2Q08, and with a 120W envelope. That's much, ain't it? Better hope it performs like heck.


http://www.xbitlabs.com/news/cpu/display/20070221162530...

Intel plans a 65nm 3Ghz 1600 PSB (Processor System Bus) in second half sometime.

Xeon 45nm DP processors are also second half.


The bad: How will things stack up in 2Q08? 2,6Ghz Barcelona against, what 3,4 - 3,7Ghz 45nm Xeon?

Thoughts? And anyone knows the difference between PSB and FSB? Also, I'm wondering what processors AMD used in their Barcelona vs Clovertown 40%+ "comparison" :) 
February 22, 2007 2:51:46 PM

Quote:
And anyone knows the difference between PSB and FSB?


Im pretty sure its just another name for the FSB, processor side bus instead of front side bus.
February 22, 2007 3:06:07 PM

Is that 120W still using the maximum peak power draw, as AMD has always done. Intel uses an *average* power draw. In other words, the average for the AMD is always well below their "TDP". This is why good reviews are careful to do a platform average over a lengthy standardized use benchmark with varying cpu loads. In this real world test, you see why Google has gone with Opterons, etc.

Another interesting point, the correct benchmarks for servers are *very different* from the usual benchmarks Tom's and other consumer sites use.
Related resources
February 22, 2007 3:19:30 PM

The fastest Barcelona planned so far is a 2.6GHz processor, model 8272SE and 2272SE. If on schedule, this would be releasing on Q2 08.

That's at least 14 months away! For a 2.6GHz as the top end? That's it? AMD should consider seppuku. At least they would die with honour! 8O
February 22, 2007 3:27:19 PM

I have a feeling 2.6Ghz in Q2 '08 must be wrong. It just doesn't make sense.

I'm not horribly interested in quad-cores at the moment, though. Dual-core processors still strike my fancy for anything short of heavy multimedia encoding.
February 22, 2007 6:46:06 PM

Quote:
I have a feeling 2.6Ghz in Q2 '08 must be wrong. It just doesn't make sense.


Apparently this info is coherent with this Feb. 5 chart from Daily Tech:

February 22, 2007 6:48:37 PM

By the time we are ready for the Barcelona Vs Clovertown showdown, it will be Barcelona vs Penryn, or Barcelona vs Nehalem
February 22, 2007 6:58:26 PM

Quote:
By the time we are ready for the Barcelona Vs Clovertown showdown, it will be Barcelona vs Penryn, or Barcelona vs Nehalem


Might be Barcelona vs. Intel Qubit! :D 

If the Barcelonas that have to hold the fort for the next 14 months are running in the low 2GHz range where are all these zippydoo wonderkin features that are going to make it such an "awesome" CPU? Everything I've seen makes it seem as if it will be roughly equal to Intel's offerings clock per clock, give or take 5%.
February 22, 2007 7:01:03 PM

The 3800+ X2 @ 2.0GHz wins in MOST things against a 965EE a 3.73GHz so clock speed isn't the only factor in determining IPC.

As far as the clockspeed of Barcelona the first dual core chips from AMD were several hundred MHz slower than the single core chips. That's just how physics works. You can't get a 2.8GHz quad at the same envelope as a 2.8GHz dual.
February 22, 2007 7:02:54 PM

Quote:
Is that 120W still using the maximum peak power draw, as AMD has always done. Intel uses an *average* power draw. In other words, the average for the AMD is always well below their "TDP". This is why good reviews are careful to do a platform average over a lengthy standardized use benchmark with varying cpu loads. In this real world test, you see why Google has gone with Opterons, etc.

Another interesting point, the correct benchmarks for servers are *very different* from the usual benchmarks Tom's and other consumer sites use.


According to everyone's favorite rag AMD is going to start using average instead of max so the TDP will come down by around 20% or more.
February 22, 2007 7:05:37 PM

Quote:
The 3800+ X2 @ 2.0GHz wins in MOST things against a 965EE a 3.73GHz so clock speed isn't the only factor in determining IPC.

As far as the clockspeed of Barcelona the first dual core chips from AMD were several hundred MHz slower than the single core chips. That's just how physics works. You can't get a 2.8GHz quad at the same envelope as a 2.8GHz dual.

QFT. I think we all know how little clockspeeds matter when comparing chips with different architectures.
February 22, 2007 7:12:04 PM

Quote:

QFT. I think we all know how little clockspeeds matter when comparing chips with different architectures.


Granted, but to cut to the chase: IN BENCH WE TRUST. One bench will tell the tale that 10,000 speculative posts can't! I can hardly wait for the true performance numbers on K10. It will be anything but dull around here that day! :D 
February 22, 2007 7:19:09 PM

you sure do know how to rain on my hopefull "Go AMD!" party... not necesarilly cuz i am an AMD fanboy, i just know that if AMD comes out with an Intel raping CPU, the prices for CPUs will drop again. Cant wait to see AMDs future Desktop CPU line-up.
February 22, 2007 7:23:33 PM

From one Ninja to the next.. things are going to get reaaaaaally cheap. Wallets ready, prepare to pay... /Green Velvet..
February 22, 2007 7:29:53 PM

Quote:
By the time we are ready for the Barcelona Vs Clovertown showdown, it will be Barcelona vs Penryn, or Barcelona vs Nehalem


Actually it will be Shanghai vs. Nehalem or perhaps K11(?) vs. Nehalem.
February 22, 2007 7:35:55 PM

Quote:
The fastest Barcelona planned so far is a 2.6GHz processor, model 8272SE and 2272SE. If on schedule, this would be releasing on Q2 08.

That's at least 14 months away! For a 2.6GHz as the top end? That's it? AMD should consider seppuku. At least they would die with honour! 8O



I guess if you compare a multithreaded app using a 2.2GHz dual core and a 3.0GHz single core what do you think will win?

More cores allows for lower clock speed. For server apps more cores are better than fewer faster cores.
February 22, 2007 7:39:01 PM

Quote:
I guess if you compare a multithreaded app using a 2.2GHz dual core and a 3.0GHz single core what do you think will win?

More cores allows for lower clock speed. For server apps more cores are better than fewer faster cores.


Of course, but I'm referring to clock per clock K10 vs. C2Q. I can't see anything spectacular that K10's arch has that will allow it to shut down a C2Q of similar clock. And I have to repeat that IN BENCH WE TRUST! The benches will certainly shut this argument down, one way or the other. Until then, it's all speculation.
February 22, 2007 7:41:51 PM

Quote:
Is that 120W still using the maximum peak power draw, as AMD has always done. Intel uses an *average* power draw. In other words, the average for the AMD is always well below their "TDP". This is why good reviews are careful to do a platform average over a lengthy standardized use benchmark with varying cpu loads. In this real world test, you see why Google has gone with Opterons, etc.


Too bad this generalization is no longer valid in relationship to Intel's current processors:

http://www.gamepc.com/labs/view_content.asp?id=o2000&pa...

Quote:

Another interesting point, the correct benchmarks for servers are *very different* from the usual benchmarks Tom's and other consumer sites use.

And Woodcrest/Clovertown dominates almost all of the key enterprise benchmarks. From TPC-C, SAP-SD, specJBB, etc.
February 22, 2007 7:47:27 PM

Is K10 a rename of K8l?
February 22, 2007 7:57:44 PM

Quote:
I guess if you compare a multithreaded app using a 2.2GHz dual core and a 3.0GHz single core what do you think will win?

More cores allows for lower clock speed. For server apps more cores are better than fewer faster cores.


Of course, but I'm referring to clock per clock K10 vs. C2Q. I can't see anything spectacular that K10's arch has that will allow it to shut down a C2Q of similar clock. And I have to repeat that IN BENCH WE TRUST! The benches will certainly shut this argument down, one way or the other. Until then, it's all speculation.

Then you're the only one since the general consensus is that Barcelona will kill in FP and maim in integer.
February 22, 2007 7:59:08 PM

Quote:
Is K10 a rename of K8l?



That's the word, though truthfully AMD never said K8L until it came out in the news. Supposedly K8L was the code name for Turion X2.
February 22, 2007 8:00:19 PM

Quote:

Then you're the only one since the general consensus is that Barcelona will kill in FP and maim in integer.

It's far more likely that Barcelona may beat Conroe in FP and very likely lose in integer.
February 22, 2007 11:16:01 PM

Quote:
The fastest Barcelona planned so far is a 2.6GHz processor, model 8272SE and 2272SE. If on schedule, this would be releasing on Q2 08.

That's at least 14 months away! For a 2.6GHz as the top end? That's it? AMD should consider seppuku. At least they would die with honour! 8O

I don't find it to be very far away. This is how AMD is doing it with their current opterons. Took until now after rev-F to get the faster ones released. But notice that 2.8 ghz dual core cpu with 1 mb cache in the fx and even x2 form was released a good deal of time before the opterons. Although the quad core will come first to opteron, the fastest cpus, the 2.6 ghz quad cores, I would bet will show up in Q3 under the Agena FX codename.
Summary: 2.6ghz will hit the FX front way before it hits the opteron front, because thats the way AMD does their opteron! In my opinion, look for a 2.6Ghz FX part in Q3?
February 22, 2007 11:23:45 PM

Quote:
The fastest Barcelona planned so far is a 2.6GHz processor, model 8272SE and 2272SE. If on schedule, this would be releasing on Q2 08.

That's at least 14 months away! For a 2.6GHz as the top end? That's it? AMD should consider seppuku. At least they would die with honour! 8O

I don't find it to be very far away. This is how AMD is doing it with their current opterons. Took until now after rev-F to get the faster ones released. But notice that 2.8 ghz dual core cpu with 1 mb cache in the fx and even x2 form was released a good deal of time before the opterons. Although the quad core will come first to opteron, the fastest cpus, the 2.6 ghz quad cores, I would bet will show up in Q3 under the Agena FX codename.
Summary: 2.6ghz will hit the FX front way before it hits the opteron front, because thats the way AMD does their opteron! In my opinion, look for a 2.6Ghz FX part in Q3?


That's actually not true. Sun was getting 2.8 Opterons maybe even before FX62. They were called SE. AMD always leads with the server parts. FX is just trickle down because of the "specialty" nature.
February 23, 2007 1:53:51 AM

this will not be a fair test for AMD cause as soon as this Barca chip wins some benchmarks against Intel's 65 nm quad offerings, Intel's 45 nm will kick its ass silly 8O
February 23, 2007 1:56:24 AM

Quote:
this will not be a fair test for AMD cause as soon as this Barca chip wins some benchmarks against Intel's 65 nm quad offerings, Intel's 45 nm will kick its ass silly 8O

AMD will have 2 quarters to get it's k10 machine moving. I don't see 45nm kicking 65nm's ass. I don't see anything because there is so much left to question right now. Barcelona performance? 45nm performance? Penyrn?
You've got no basis for what you say, pretty much all of you... SPECULATION!?!?!
February 23, 2007 3:01:23 AM

Quote:
Then you're the only one since the general consensus is that Barcelona will kill in FP and maim in integer.


Not only do I subscribe to Jack's statement with regards to clock speed differential, but continuing speculation about the performance of K10 is as ridiculous as arguing about how WinFS in Vista is such a killer feature. The bottom line is that for years Longhorn had WinFS but when Vista showed up, it was stripped out. I'm sure that I can hire a few engineers here at CRA Semiconductors and come up with some really impressive CPU specs on paper. But until the benchys come in, it's all brainfarts.

Quote:
In my opinion, look for a 2.6Ghz FX part in Q3?


OK, of what year though? 07 or 08?
February 23, 2007 3:17:20 AM

Quote:
Is that 120W still using the maximum peak power draw, as AMD has always done. Intel uses an *average* power draw. In other words, the average for the AMD is always well below their "TDP". This is why good reviews are careful to do a platform average over a lengthy standardized use benchmark with varying cpu loads. In this real world test, you see why Google has gone with Opterons, etc.
Another interesting point, the correct benchmarks for servers are *very different* from the usual benchmarks Tom's and other consumer sites use.


Uuummm, yoo hoo...... clue to you:


Quote:
Intel seems to be on the rebound not only on the desktop side of the picture, but the server side as well. Google has begun buying Intel server components in high volume, Pat Gelsinger, a co-general manager of Intel's Digital Enterprise Group, wrote on his blog. Gelsinger said Intel had to create custom equipment to win back the business. Intel design teams "have been maniacal as we designed a unique board for them, developing a unique memory module with them, working every angle of the cost equation and engaging with our sales teams to get the business," Gelsinger said. "We bought a small number of chips from Intel recently, but we continue to be supplied by more than one vendor," Google said in a statement but didn't discuss motherboard purchases or other details.



Quote:
Intel inside again for new Google servers
Resurgent chipmaker has reclaimed an exacting and prestigious customer that had chosen AMD-based x86 servers for the last year.
By Stephen Shankland
Staff Writer, CNET News.com

Published: January 25, 2007, 12:13 PM PST
TalkBack E-mail Print del.icio.us Digg this
Intel, armed with a custom-designed motherboard, has reclaimed Google as a server customer after a year watching the search powerhouse give its business to Advanced Micro Devices, CNET News.com has learned.

Google has begun buying Intel server components in high volume, said Pat Gelsinger, a co-general manager of Intel's Digital Enterprise Group, speaking about the Google relationship on an internal Intel blog entry Wednesday seen by CNET News.com. "We're in business with the volume systems ramp under way," he said.


So much for Hals theory
February 23, 2007 3:23:00 AM

07... That's the last roadmap I saw with agena on it
February 23, 2007 3:42:00 AM

Quote:
07... That's the last roadmap I saw with agena on it


Unfortunately, DailyTech disagrees with you. The latest roadmap shows no 2.6GHz before Q2 08.
February 23, 2007 3:42:11 AM

Quote:
Actually, we do have some data --- Intel has published several high K papers in the past and clocking up to 3.5-3.7 GHz range is not only doable but the data suggests this is where Penryn may well land.... this is at stock conditions and at 25% of the source to drain leakage (i.e. lower power).
http://www.intel.com/technology/silicon/HighK.pdf
http://www.intel.com/technology/silicon/ieee/IWGI2003.p...

It certainly remains in the land of speculation, but we also certainly have more concrete data from the Intel side of the equation than the AMD.


Let's just hope Intel can get their clocks above the 2GHz Penryn demos. :wink:

I sure they will do it. It just makes you wonder what happened.
February 23, 2007 4:06:08 AM

r600 delay along with k10...so maybe they are going to build something that puts the 680 to shame as far as a chipset...if intel is claiming huge gains with a new chipset maybe amd has done the same
February 23, 2007 6:07:01 AM

I think maybe you missed this
February 23, 2007 6:24:19 AM

I really have trouble buying into the hype this early in the game but.
AMD has said thier barcelona will be 40% faster than Intel's chips. They do not say clock for clock, or top bin to top bin. Let's just say clock for clock.
If Intel's penryn can max out at 3.5, AMD would only have to come out with a 2.5 part to match it.
That is assuming that Intel puts out a chip at it's top possible bin. Especially for servers, Intel has stayed well back from theoretical max. If the chips can only run @ 3.5, do you really think they will sell them above 3.2?
If 3.2 is thier top bin, then the basic 2.3 is well 2.3 * 1.4=3.22.

My only problem is that I find the 40% faster a little hard to believe.
February 23, 2007 6:35:28 AM

Quote:
I think maybe you missed this


What is Google doing with all those "disks and memory"??? I smell a story here... :lol: 

As for your 40% calculation, yes, it is theoretically possible, but I agree with you that it's a little hard to believe. If AMD is playing performance/watt games again, then maybe. But otherwise in sheer ba!!s-out horsepower, I say no way. But we will see. I'm seriously thinking of adding to my signature: IN BENCH WE TRUST! 8)
February 23, 2007 8:36:50 AM

Quote:
I think maybe you missed this


What is Google doing with all those "disks and memory"??? I smell a story here... :lol: 

As for your 40% calculation, yes, it is theoretically possible, but I agree with you that it's a little hard to believe. If AMD is playing performance/watt games again, then maybe. But otherwise in sheer ba!!s-out horsepower, I say no way. But we will see. I'm seriously thinking of adding to my signature: IN BENCH WE TRUST! 8)

Do it.

Doooooooo itttttttttttttttttttttt.

Dooooooooooooooooooooooooooooooooooooooooo itttttttttttttttttttttttttttttttttttttttttttttttttttttttttt.

You know you want to.
February 23, 2007 9:22:42 AM

Quote:
i think more fitting for you would be
"In Wench i trust"


Nah. Never trust wenches. I use them and abuse them, but never trust them! :twisted:

When you say 40% faster with respect to price/perf., I'm not too sure. Maybe perf/watt, which we all know is a total snore of a quality. Who really cares???

(Even though nobody helped me out with my formula on actually calculating the realworld current draw of differing TDP CPUs. :cry: 

Quote:
Do it.

Doooooooo itttttttttttttttttttttt.

Dooooooooooooooooooooooooooooooooooooooooo itttttttttttttttttttttttttttttttttttttttttttttttttttttttttt.

You know you want to.


Damn. How could I resist. OK, I give in! :lol: 
February 23, 2007 9:40:15 AM

This threads seems to be an exercise in trying to squeeze water out of a stone. But I don't understand why all of the obvious things keep getting overlooked.

For example, why do people keep mentioning that 2.6Ghz won't be out until Q2 08? You get 2.5Ghz in Q3 07. The extra 100Mhz isn't going to matter that much.

Since the highest clocks will probably draw too much power for large scale servers the preferred clock would probably be 2.66Ghz for Intel and 2.5Ghz for AMD. This would make the Intel chip only about 6% faster. This is not a huge advantage.

The 40% number is not IPC. The only area where AMD will have a sizable lead over Intel is in legacy FP operations. These are typically fairly small in terms of running time so even if AMD is 50% faster it could mean as little as 5% in terms of actual speed. SSE and Integer should roughly match at the same clock. So, if Intel has the faster clock they will have a faster speed . . . sort of. AMD has had an excess of memory bandwidth or, in other words, it wasn't able to move the data and instructions from memory to the pipeline fast enough. However, doubling the L1 bus width and prefetch size will certainly help. Look for AMD to gain a bit in memory intensive operations. Also, power draw is a huge factor on any large scale server installation. And, AMD should have the advantage on power draw since Intel will still be using FBDIMM and the massive quad FSB northbridge will also draw a lot of power. This is where the 40% number comes from. If this still isn't clear then let me say this a different way:

Upper desktop and small scale servers: Advantage - Intel
Low to mid desktop: Advantage - neither
Mobile: Advantage - neither
Medium to large scale servers: Advantage - AMD

Intel will benefit the most on the upper desktop where power draw isn't as much of a factor (single socket and DDR2) and its higher clock will put it ahead in many benchmarks. Intel will be at its greatest disadvantage on 4-way and up servers where its lower FSB speed, I/O volume, and FSB controller contention put it at a disadvantage along with the higher power draw of its quad northbridge and FBDIMM modules. Everything else should fall pretty much in the middle.

BTW, I believe they are incorrect about Barcelona's using HT 1.0; I think it has HT 2.0 even though it still only has DC 1.0. This is interesting though since we can clearly see the space for the 4th HT link on the Barcelona die. This does make one wonder why AMD is putting off DC 2.0 then until mid 2008 if it is on the die in Q2 07.

There really isn't much point in talking about high clocks. I tend to agree with JJ that 20% more should be possible for Intel and this would indeed be 3.6ghz. I could see these speeds being released sometime in 2008. However, AMD will also have 45nm which should give lower power draw if not higher clocks plus DC 2.0 and a mature stream processor. This should make quad Opteron a pretty heavy contender in servers. I'm guessing Itanium and Power will both start feeling the heat.

I'm still not entirely certain that Intel is trying to match Opteron in this segment since a really heavy duty quad Xeon combined with quad Opteron could essentially be the death of Itanium even on the new socket. The installed base of Power is too high to die this easily and IBM will prefer it in-house but there is no doubt that IBM will start feeling the R&D cost versus return pinch as well. I have a suspicion that Intel may be drawing the Xeon out a bit to give the new socket Itanium some breathing room.

Barcelona is a server chip. I assume most people here care more about the desktop. I doubt that Intel will lose the upper desktop lead anytime in 2007 or the first half of 2008. On the other hand, I doubt Intel will be able to stop AMD from slowly taking share during 2007. I believe AMD will finish 2007 with a bit more share than it has now. It remains to be seen if Intel will try to match the all out quad FX system which I assume is way too expensive for most people here.

BTW, the reason AMD used maximum TDP per family in the past was to prevent a tight power budget on 3rd party motherboards from preventing a cpu upgrade. It may be the case that now that AMD can guarantee a working chipset from its ATI subsidiary that it can be less conservative about TDP ratings.

Intel should give us a much better idea what it has in mind in April at the Spring IDF, and AMD should give us a better idea what it has in mind at the June Technology Analyst Day. This should give us a better comparison for 2008.

Finally, Penryn didn't tapeout until the beginning of January. It typically takes 12 months from tapeout to release. The pull in cannot be more than 2 months at most.
February 23, 2007 10:13:17 AM

Quote:

Then you're the only one since the general consensus is that Barcelona will kill in FP and maim in integer.

It's far more likely that Barcelona may beat Conroe in FP and very likely lose in integer.


Except I'm basing my statements on the specs of both chips, not a gut feeling.
But we'll see. AMD hasn't blinked yet and I don't think Barcelona will cause them to either.
February 23, 2007 10:24:24 AM

Scientia, having 2.5Ghz sometime in the foreseeable future and then getting a tiny increment much later on is the point. I'm not arguing that 2.5 will be slow and/or 2.6 is going to be the only worthwhile CPU. I'm stating the fact that even if we allow that K10 is competitive on launch day, by the time it gets around, over a year from now, to reach its first upwards increment of only 100Mhz, it's gonna be a day late and a dollar short. And I'm not seeing anything tangible in the foreseeable future from the AMD camp to exceed that K10 2.6.

The rumblings are emanating from the Intel camp that they have some monsters about to be unleashed. By the time Q2 '08 (or later) comes around, the extra 100Mhz pumped into K10 might be like trying to compete in the Daytona 500 by pedalling your kart faster.

Barcelona is a server chip, however with the lines to desktop blurring day by day, we have to take the enthusiast-prosumer like me into consideration. I have repeatedly stated that I am in the market, hopefully by Q3 07 if the CPUs are available, to get a 2xQuad. 771, 775 or QFX doesn't matter one whit. The only rationale is going to be price/performance. I'm just as open to K10 as I am to Clovertown or anything else, including Caneland (in my wildest poweruser dreams...). I'd love to see how a 775 dual socket would stack up against QFX. That would give me two valid options that would save me the expense of FB-DIMMs.

Again, as I have just incorporated into my sig: In Bench We Trust! (Others pay cash!) :lol: 
February 23, 2007 11:13:11 AM

Quote:

Then you're the only one since the general consensus is that Barcelona will kill in FP and maim in integer.

It's far more likely that Barcelona may beat Conroe in FP and very likely lose in integer.


Except I'm basing my statements on the specs of both chips, not a gut feeling.
But we'll see. AMD hasn't blinked yet and I don't think Barcelona will cause them to either.

Here is my suggestions. If K10 chips are slower in the desktop deparment you should be forced to eat 10kg of corndog in one day. Infront of a webcam of course :) 
February 23, 2007 12:09:59 PM

Jack,

I think some people still don't understand what High-K gate material will buy Intel 8O If this process technology comes in with high yields, the gap between Intel & AMD will get bigger. Intel has changed the game by switching to new gate material, they effectively opened a new front in this competition. You add the new architecture with IMC and AMD will be behind on 2 fronts: process technology shrink node + gate material. Chip design at that point will be a draw or actually in Intel's favor. AMD's Fuzion is not a high end competitor and most likely we will not see it till 2010. Intel has an answer in the works for that one too. Intel has been executing since C2D announcement roughly about a year ago, AMD has had nothing to offer other than hype :roll:
February 23, 2007 12:20:19 PM

Quote:
First of all, sheez this forum annoys me.. Takes like 2 minutes to get the page to show ! Feels like I'm taking crazypills waiting so long ..

To the case.

http://www.hkepc.com/bbs/itnews.php?tid=746541

So 2,3Ghz top bin for Barcelona. June mass production, if not delay.

But notice, only 2,6Ghz Barcelona scheduled for 2Q08, and with a 120W envelope. That's much, ain't it? Better hope it performs like heck.


http://www.xbitlabs.com/news/cpu/display/20070221162530...

Intel plans a 65nm 3Ghz 1600 PSB (Processor System Bus) in second half sometime.

Xeon 45nm DP processors are also second half.


The bad: How will things stack up in 2Q08? 2,6Ghz Barcelona against, what 3,4 - 3,7Ghz 45nm Xeon?

Thoughts? And anyone knows the difference between PSB and FSB? Also, I'm wondering what processors AMD used in their Barcelona vs Clovertown 40%+ "comparison" :) 


PSB sounds like there answer to HT
February 23, 2007 12:52:03 PM

Nope. PSB has been around for a while. It stands for Processor System Bus.
February 23, 2007 12:56:08 PM

Quote:
Nope. PSB has been around for a while. It stands for Processor System Bus.

I think the PSB is named because Intel started to use dedicated bus for each processor-to-northbridge connection. :wink:
February 23, 2007 1:06:02 PM

Quote:
Jack,

I think some people still don't understand what High-K gate material will buy Intel 8O If this process technology comes in with high yields, the gap between Intel & AMD will get bigger. Intel has changed the game by switching to new gate material, they effectively opened a new front in this competition. You add the new architecture with IMC and AMD will be behind on 2 fronts: process technology shrink node + gate material. Chip design at that point will be a draw or actually in Intel's favor. AMD's Fuzion is not a high end competitor and most likely we will not see it till 2010. Intel has an answer in the works for that one too. Intel has been executing since C2D announcement roughly about a year ago, AMD has had nothing to offer other than hype :roll:


AMD/IBM also has the High-k metal gates. They are planning to use on there 45nm process.
February 23, 2007 1:07:19 PM

Ok my bad.
February 23, 2007 1:13:06 PM
February 23, 2007 1:18:09 PM

Quote:
This threads seems to be an exercise in trying to squeeze water out of a stone. But I don't understand why all of the obvious things keep getting overlooked.

For example, why do people keep mentioning that 2.6Ghz won't be out until Q2 08? You get 2.5Ghz in Q3 07. The extra 100Mhz isn't going to matter that much.

Since the highest clocks will probably draw too much power for large scale servers the preferred clock would probably be 2.66Ghz for Intel and 2.5Ghz for AMD. This would make the Intel chip only about 6% faster. This is not a huge advantage.

The 40% number is not IPC. The only area where AMD will have a sizable lead over Intel is in legacy FP operations. These are typically fairly small in terms of running time so even if AMD is 50% faster it could mean as little as 5% in terms of actual speed. SSE and Integer should roughly match at the same clock. So, if Intel has the faster clock they will have a faster speed . . . sort of. AMD has had an excess of memory bandwidth or, in other words, it wasn't able to move the data and instructions from memory to the pipeline fast enough. However, doubling the L1 bus width and prefetch size will certainly help. Look for AMD to gain a bit in memory intensive operations. Also, power draw is a huge factor on any large scale server installation. And, AMD should have the advantage on power draw since Intel will still be using FBDIMM and the massive quad FSB northbridge will also draw a lot of power. This is where the 40% number comes from. If this still isn't clear then let me say this a different way:

Upper desktop and small scale servers: Advantage - Intel
Low to mid desktop: Advantage - neither
Mobile: Advantage - neither
Medium to large scale servers: Advantage - AMD

Intel will benefit the most on the upper desktop where power draw isn't as much of a factor (single socket and DDR2) and its higher clock will put it ahead in many benchmarks. Intel will be at its greatest disadvantage on 4-way and up servers where its lower FSB speed, I/O volume, and FSB controller contention put it at a disadvantage along with the higher power draw of its quad northbridge and FBDIMM modules. Everything else should fall pretty much in the middle.

BTW, I believe they are incorrect about Barcelona's using HT 1.0; I think it has HT 2.0 even though it still only has DC 1.0. This is interesting though since we can clearly see the space for the 4th HT link on the Barcelona die. This does make one wonder why AMD is putting off DC 2.0 then until mid 2008 if it is on the die in Q2 07.

There really isn't much point in talking about high clocks. I tend to agree with JJ that 20% more should be possible for Intel and this would indeed be 3.6ghz. I could see these speeds being released sometime in 2008. However, AMD will also have 45nm which should give lower power draw if not higher clocks plus DC 2.0 and a mature stream processor. This should make quad Opteron a pretty heavy contender in servers. I'm guessing Itanium and Power will both start feeling the heat.

I'm still not entirely certain that Intel is trying to match Opteron in this segment since a really heavy duty quad Xeon combined with quad Opteron could essentially be the death of Itanium even on the new socket. The installed base of Power is too high to die this easily and IBM will prefer it in-house but there is no doubt that IBM will start feeling the R&D cost versus return pinch as well. I have a suspicion that Intel may be drawing the Xeon out a bit to give the new socket Itanium some breathing room.

Barcelona is a server chip. I assume most people here care more about the desktop. I doubt that Intel will lose the upper desktop lead anytime in 2007 or the first half of 2008. On the other hand, I doubt Intel will be able to stop AMD from slowly taking share during 2007. I believe AMD will finish 2007 with a bit more share than it has now. It remains to be seen if Intel will try to match the all out quad FX system which I assume is way too expensive for most people here.

BTW, the reason AMD used maximum TDP per family in the past was to prevent a tight power budget on 3rd party motherboards from preventing a cpu upgrade. It may be the case that now that AMD can guarantee a working chipset from its ATI subsidiary that it can be less conservative about TDP ratings.

Intel should give us a much better idea what it has in mind in April at the Spring IDF, and AMD should give us a better idea what it has in mind at the June Technology Analyst Day. This should give us a better comparison for 2008.

Finally, Penryn didn't tapeout until the beginning of January. It typically takes 12 months from tapeout to release. The pull in cannot be more than 2 months at most.




Pretty god assessent as usual and also as usual I have to disagree abotu a few things. First, I don't think that Conroe is gong to stay ahead of Kuma and Agena will be Opteron for the desktop so it should have the same specs vs. Kentsfield.

AMD is not mentiong particulars about desktop benches because Barcelona IS a server part. but Ibelieve that just like X2 Kuma will have the same effect even staying at dual core.

I am expecting SPEC numbers soon as it did tape out last summer(Aug?). The way they described the improvements (at least with FP - this includes SSE4A which allows 4 FLOPS/cycle) it will be 1.8X the speed of dual core (when comparing two K10 cores to two K8 cores).
The doubling of the L1 and loads per cycle not to mention OoO will ncrease int performance by at least 15%. Especially since they are officially suporting 1066 RAM and hopefully the split planes for IMC and cores will allow async RAM that will actually run at its rated sped at default.
As it is now the HT bus would have to be 533Mz rather than 400MHz to make an even number not above the rated speed.

I can agree though that this is like beating a dead horse. I can also agree that there seems to be a large concentratin of Intel employees around here.

It's off topic but one guy said if you have a 10% margin at $300 per sale you can make the sme amount with $200 per sale ( I tried to say that you would need 30% more sales to make up for the 30% per sale price).

Anyway, I'm more than confident that Dirk Meyer and those guys now what they're doing. They worked to improve on K8 and not beat C2D. 80% better than Opteron is serious horsepower.

ALL HAIL THE DUOPOLY!!!!
!