Intel adapts Performance Rating. A sign of slowing clockspee

[Deadmeat]

Intel 'to adopt performance ratings'
By Tony Smith
Posted: 15/03/2004 at 09:44 GMT
The Register Mobile: Find out what the fuss is about. Take the two week trial today.

Intel appears to have conceded at last that a processor's clock frequency isn't the be-all and end-all of chip performance. It is to begin adding performance ratings to its processors, the better to distinguish one model from another.

The sound you hear in the background is AMD's workforce dancing for joy. Having been forced to adopt what are essentially AMD's 64-bit extensions to the 32-bit x86 ISA, Intel is now borrowing AMD's scheme for naming chips - or at least something along similar lines.

The chip giant hasn't confirmed the move or denied it, preferring not to comment. But CNET, picking up on rumours circulating on the Net late last week, cites a source "familiar" with Intel's plans as admitting that it has performance ratings in its sights.

Next quarter's launch of Dothan, the 90nm version of the Pentium M, will see the new scheme debut. Dothan has 2MB of L2 cache, but is unlikely to offer radically higher clock speeds than the current line of 130nm Pentium Ms. Performance ratings would allow Intel to demonstrate the effect of the extra 1MB of cache rather than show a small jump in clock frequency. However, Intel will still tout clock speeds.

The move will also help push Prescott, the 90nm Pentium 4. Prescott has proved something of a disappointment, shipping at clock speeds no different from older, 130nm versions of the chip. Again, performance ratings would allow Intel marketeers to stress the advantage of the part's larger cache, updated architecture and frontside bus speed. According to the CNET source, Intel will put the scheme in place this summer.

Curiously, the performance metric will be used to compare chips in the same family, not across the board, presumably to leave clear blue water between, say, Xeon and Pentium 4 processors. AMD does the same thing with its Opteron and Athlon 64 processors.

Of most use to consumers would be a consistent cross-vendor metric. The result: processors could all be rated using a measure as standard as clock frequency but more relevant to today's superscalar, multi-threading CPUs. Finally, the 'megahertz myth' could be laid to rest.

But AMD's attempt to build an industry consensus behind such a metric was quietly dropped last year. The company simply could not find enough chip vendors to back the move. ®

This is the clearest sign that the clockspeed improvement rate of CMOS devices is slowing down. Think about it. AMD's devices clock at 1.8~2.4 Ghz range, while fastest PPC runs at 2 Ghz. You would expect CELL to top out at 2 Ghz at most in single PE configuration, good for 128 GFLOPS tops or less.

 

[hey69]

i didn't expect to read CELL and 2ghz at end of the post when i was reading this article

 

[archie4oz]

I'm afraid I'm going to have to disagree...

I don't think there's been a truly measurably consistent rate of clock speed improvement to base this assumption, especially if you're measuring actual performance.

Clock speeds have been all over the place from process enhancements to architectural improvements. I remember an offhand comment from an Intel engineer mentioning that the clock scalability of the P6 core benefitted more layout changes to the core than the Coppermine process shrink. So far the P4's been the only design that's really emphasized clock speed seemingly over anything else, and recently it seems as though that may have been only half of the coin as it may look as though the overly deep design was intentionally designed to suffer from pipeline bubbles to accommodate SMT...

 

[Saem]

Deadmeat you're way off.

This is because of Banias/Centrino. They chip clocks lower than P4 Mobile chips, but beats them interms of performance. So they need to fix this disparity.

This doesn't have to do with slowing clock rates. Additionally, your calculations for clock scaling seem WAY off, it seems your methods of extrapolation are to blame. I'd say that it's simillar for your thoughts on transistor density or you're just keeping your estimates exceptionally conservative.

 

[jvd]

Deadmeat you're way off.

This is because of Banias/Centrino. They chip clocks lower than P4 Mobile chips, but beats them interms of performance. So they need to fix this disparity.

This doesn't have to do with slowing clock rates. Additionally, your calculations for clock scaling seem WAY off, it seems your methods of extrapolation are to blame. I'd say that it's simillar for your thoughts on transistor density or you're just keeping your estimates exceptionally conservative.

While i disagree with what most of deadmeat is saying , I also disagree with your reasoning .

THe p4 (netburst) tech is reaching its celling. This tech was allways meant to scale very high. Intel knew that consumers bought pcs because of mhz . Its been drilled into consumers. They saw amd had a chip that was faster or on par with the p3 and was clocking higher. Since they couldn't build a chip that would clock as high as the athlon and be as fast per mhz as the athlon they went with the p4 tech because it would scale high enough to dim the athlons advantage.

But now the p4 is stalling . They have been stuck at 3.2 ghz for almost 8 months now. The ees are vapor ware at best and are expensive beasts and barely compete against a ghz slower athlon 64 .

The hyped up prescott core is a hot beast. It has trouble clocking as high as the last core . So intel knows it can't just increase clock speeds whenever amd starts to catch up now. So they are going to increase the work done by the chip per mhz .

Thus where rating system comes in. Which will of course be used across the board all the way to the p3 line (bannas and centrino)

But this doesn't mean that other core designs and chip designs can't clock higher .

Just like when the p3 reached its limits of the time didn't mean the athlon or p4 would stop climbing .

The cell chip could very well clock at 4ghz although I doubt it. It is very complex and very big from what I can gather .

 

[Tahir2]

I believe Prescott will have a few more tweaks made to it on the process level and power level when it goes to its new socket form factor which will allow Intel to reach its goals of 4-5GHz for Prescott.

However Intel realises that clockspeed increases at the cost of massively increased power consumption is not the way to go.

Edit: P.S. I have no idea what this topic is doing in the Console Forum.

 

[Guden Oden]

Jvd,
I'd say you're the one that's way off. The "netburst" (geez, speak of a corny PR-name) architecture's just been extended by almost 50%, meaning there's now a lot more headroom to scale into. Prescott's just come onto the market; so the first revision runs a bit hot, they're fixing that, but really, there's not much point for them to bump up the clockspeed at the moment. They already got the biggest number, so why compete with themselves?

It doesn't have trouble at all to match current northwoods, in fact, there's plenty of headroom in the chip even as it is in its present hot incarnation, judging from our happy overclockers from around the world. Heck, there's no reason at all it should have, not with that obnoxiously long pipeline it's got.

Not that any of this's got much to do with consoles, but I didn't start this topic so don't blame me.

 

[Megadrive1988]

There are no Intel CPUs slated to power any of the forthcoming nextgen consoles, as far as I know. not counting "console" PCs like Phantom, DISCover, etc.

unless Microsoft decides to put an entire Xbox1 chipset, including the Intel P3/Celeron hybrid, in Xbox2.

 

[Deadmeat]

...

But now the p4 is stalling . They have been stuck at 3.2 ghz for almost 8 months now. The ees are vapor ware at best and are expensive beasts and barely compete against a ghz slower athlon 64 .

You answered your own question. Like you said the NetBurst was built for speed, and is fabricated on the best fab in the indsutry, yet they have trouble gaining clockspeed. Why is this happening? Because the power consumption level of individual transistor won't go down anymore due to current leakage even if they become smaller and smaller. Since the power consumption is not going down, it becomes difficult to clock higher.

Now, if something like NetBurst built expressively for clockspeed has trouble gaining clockspeed, then imagine the trouble SCEI engineers are in trying to make CELL run at any usuable clock... That 4 Ghz 4 PE CELL is somebody's fat dream not realized until the end of this decade. In the meanwhile, the PSX3 must do with a 2 Ghz 1 PE version...

 

[Tahir2]

How are AMD able to release processors that are 100 million transistors plus and scale to 2Ghz and beyond yet only consume 35W?

Deadmeat.. you are deadwrong.. again and again and again.

 

[Guden Oden]

Re: ...

You answered your own question. Like you said the NetBurst was built for speed, and is fabricated on the best fab in the indsutry

You keep harping this on and on, yet you have no shred of evidence backing that up. IBM does strained silicon, low-k AND soi in their G5s, Intel's not even tried SOI yet. Best fab in the industry my ass.

yet they have trouble gaining clockspeed.

No they don't, wtf are you talking about???

Why is this happening? Because the power consumption level of individual transistor won't go down anymore due to current leakage

Blaha blaha. TLA that will shut you up good: SOI. Besides, there's other tech too under development that will counter leakage, so it's not as if you're right in your predictions or anything... We have nothing to worry about here! Intel has it's double-gated transistors or something like that for example.

In the meanwhile, the PSX3 must do with a 2 Ghz 1 PE version...

I knew you'd manage to get in a swipe at Sony here even though YOU started the topic talking about Intel... You're just too funny man...

Besides, you're not a CMOS engineer, so how would you know anyway?

 

[jvd]

Re: ...

But now the p4 is stalling . They have been stuck at 3.2 ghz for almost 8 months now. The ees are vapor ware at best and are expensive beasts and barely compete against a ghz slower athlon 64 .

You answered your own question. Like you said the NetBurst was built for speed, and is fabricated on the best fab in the indsutry, yet they have trouble gaining clockspeed. Why is this happening? Because the power consumption level of individual transistor won't go down anymore due to current leakage even if they become smaller and smaller. Since the power consumption is not going down, it becomes difficult to clock higher.

Now, if something like NetBurst built expressively for clockspeed has trouble gaining clockspeed, then imagine the trouble SCEI engineers are in trying to make CELL run at any usuable clock... That 4 Ghz 4 PE CELL is somebody's fat dream not realized until the end of this decade. In the meanwhile, the PSX3 must do with a 2 Ghz 1 PE version...

I also said they had the same problem once before with the p3 . Remember tomshardware finding that at 1.1 ghz it was unstable and they were recalled . Then nothing for 6 months while amd hit 1.2 ghz and vola a p4 came out that would eventualy scale to over 2 ghz faster than the p3 which stalled.

Its just this design of chip is stalling . They will need to move onto another one to get more out of it

 

[jvd]

Jvd,
I'd say you're the one that's way off. The "netburst" (geez, speak of a corny PR-name) architecture's just been extended by almost 50%, meaning there's now a lot more headroom to scale into. Prescott's just come onto the market; so the first revision runs a bit hot, they're fixing that, but really, there's not much point for them to bump up the clockspeed at the moment. They already got the biggest number, so why compete with themselves?

It doesn't have trouble at all to match current northwoods, in fact, there's plenty of headroom in the chip even as it is in its present hot incarnation, judging from our happy overclockers from around the world. Heck, there's no reason at all it should have, not with that obnoxiously long pipeline it's got.

Not that any of this's got much to do with consoles, but I didn't start this topic so don't blame me.

Yea I know i didn't really want to lock another thread today .

But anyway . I'm not wrong. This isn't the first revision of the presocott. THe prescott was over 8 months late and was revised two times to get it to clock at its current speed.

Do you rally think intel would be sitting around letting amd have have the fastest chip on the block with out countering. And the ee is not a counter its a save face move. They aren't out in any numbers . The 3.2 ghz versions aren't even out in siginifgaint numbers

 

[zidane1strife]

Because the power consumption level of individual transistor won't go down anymore due to current leakage even if they become smaller and smaller. Since the power consumption is not going down, it becomes difficult to clock higher.

From what I've heard some companies have come up with ways to effectively deal with leakage at the smaller processes... SoN, and the like.

Ibm, tosh...

Do you rally think intel would be sitting around letting amd have have the fastest chip on the block with out countering. And the ee is not a counter its a save face move. They aren't out in any numbers . The 3.2 ghz versions aren't even out in siginifgaint numbers

Maybe... it is time for intel to fall...

 

[jvd]

Well i highly doubt intel will help in destroying themselves. They aren't nvidia .

 

[nobie]

Of course power consumption will go down as transistors become smaller. Will leakage be an issue? Yes. But that doesn't mean power consumption won't go down. IBM has said that with current technologies leakage won't be a crippling factor until the 30nm node. Intel says 16nm is possible if different materials are used. Beyond that point maybe power consumption won't go down, but that point is in the distant future (10+ years). For the foreseeable future there aren't any insurmountable hurdles to CMOS scaling.

 

[Guden Oden]

Yea I know i didn't really want to lock another thread today .

You could just punt it over to hardware talk where it belongs, you know... We'd actually get more people contributing that way I might add, as many B3Ders don't read the console forum (for whatever reasons I cannot understand... )

But anyway . I'm not wrong. This isn't the first revision of the presocott.

It's the first spin that's been RELEASED.

and was revised two times to get it to clock at its current speed.

Source?

Do you rally think intel would be sitting around letting amd have have the fastest chip on the block with out countering.

Hm, they've been in that position more or less all from the Athlon was introduced until the P4 started to out-scale its AMD counterpart (around P4 @ 2.4GHz or so), and that didn't kill them. AMD is too small to be much of anything but a minor annoyance anyway.

And the ee is not a counter its a save face move.

Well, of course it is, with the price tag it carries. Still doesn't change that AMD is a very minor player.

The 3.2 ghz versions aren't even out in siginifgaint numbers

Source?

 

[jvd]

Yea i could


what does being released have to do with anything .

Doesn't change the fact that there weren't already respins of it to get greater performance. The first spins of it were getting 2.8 ghz. Source aceshardware forums .

THe p4 on rambus platforms have pretty much over about 70% of the time been faster than the athlons with intel responding promtly whenever amd took the lead. Which is what the prescott was meant for.

What does amd being a minor player have anything to do with this convo. Amd can take market share away from intel. The more it does the less money intel makes and the bigger amd becomes .

I dunno about you but i wouldn't want a minior player to start challengine me and grow from that minor player to something that can beat me .



as for the 3.2 ghz versions not out in siginifgaint numbers that is based on searching 48 stores (online for it) and asking on forums if anyone has seen it in stock as i needed to buy one . This is valid as of a week ago. As of 3 weeks ago you can see aces general forum for the reports from others of them being in limited supply .

Not what you'd expect from a 800$ cpu .

 

[zidane1strife]

maybe...

 

[Rugor]

Precott is like Willamette, released at too low a clockspeed to really show its performance.

Having said that, I do think the Netburst was the wrong way to go.