Second-class Intel to trail AMD for years

[Hyperlite]

The Register

 

[LTC8K6]

Curly:

Oh no!
Not that!

 

[theMan]

"This should worry folks at Dell, HP and SGI, as they're most vulnerable to Intel's shortcomings."

hahaha, suckers.

 

[Aenslead]

dissapointing, but not unexpected.

 

[dmens]

Guess the reporter missed the point, because Whitefield is about as futuristic as Woodcrest. And he obviously can't read the roadmap, because in 18 months, the Nehalem cores are will be about ready for their big debut, which should run nicely.

And before that is in the market, just because Woodcrest uses FSB doesn't make it non-competitive. With DDR3 and FBD1, the latency difference is reduced, and there are architectural to reduce memory bandwidth requirements, and server chips tend to have those enhancements. And one must think about the design tradeoffs... with the converged approach to core design, IMO one should spend more time on a more flexible approach.

 

[SickBeast]

I really don't know what is stopping intel from making an on-die memory controller. Even Wingznut (an intel engineer) has stated that he wishes they would go with such a design.

Maybe it's pride holding them back.

 

[dmens]

Who says there is no integrated mem controllers in the works? LOL.

 

[Markfw]

Originally posted by: dmens
Guess the reporter missed the point, because Whitefield is about as futuristic as Woodcrest. And he obviously can't read the roadmap, because in 18 months, the Nehalem cores are will be about ready for their big debut, which should run nicely.

And before that is in the market, just because Woodcrest uses FSB doesn't make it non-competitive. With DDR3 and FBD1, the latency difference is reduced, and there are architectural to reduce memory bandwidth requirements, and server chips tend to have those enhancements. And one must think about the design tradeoffs... with the converged approach to core design, IMO one should spend more time on a more flexible approach.

I know you work there, and its the register, but Intel deserves most of what is coming at them lately, as they missed the ball, and admitted it, that for 12-18 months they would be behind. I will believe it when I see non-prejudiced benchmarks, IE Anandtech.com. Until then, its all speculation.

 

[dmens]

Oh yeah, 12-18 months sounds about right. But 2009? Nah.

As for missing the ball, management needs to get its act together.... still bickering over politics and agendas while behind. Yeesh.

 

[stratman]

Nehalem's still coming out? IIRC that was the extremely high-frequency netburst proc?

I thought it was cancelled after the ghz race ended, and merom/conroe were taking it's place.

Thanks for your posts dmens, they're always appreciated (by most of us at least).

 

[dmens]

Nah, new and improved: http://www.theinquirer.net/?article=24439

 

[Markfw]

Originally posted by: dmens
Oh yeah, 12-18 months sounds about right. But 2009? Nah.

As for missing the ball, management needs to get its act together.... still bickering over politics and agendas while behind. Yeesh.

That means sometime in 2007.. I think we all knw that is probably the timeframe. Also, that stement from management was 6 months ago or so, but I still think 2007 will be Intels big comback, unless AMD pulls a rabbit out of their hat. Bottom line, is I will wait and see, no speculation here, except AMD will be around a little longer to fight and compete !

 

[Viditor]

But Nehelam will not be supporting CSI, which is the whole point. As far as on-die memory controllers, that is a feature of CSI (for those that don't know, CSI is to be a ring form of HTransport). Even with multiple FSBs, the quadcore and higher chips will be quite limited in bandwidth when compared to the Opterons...
I think the Inq has got it right this time...

That said, Intel may be quite formidable in the low end server sector...which is after all the largest segment by far.

 

[dmens]

Nehalem won't have CSI? What?

Also, CSI is neither FSB, HT nor IMC, but it can be all three, or one, or two. Let's just say CSI shouldn't be bandwidth limiting, imo of course.

 

[irwincur]

It is funny to see how much a small company like AMD changed the entire market in two short years.

- Won the GHz race by effectively letting Intel run it into the ground. AMD wins by sitting aside and suckering Intel into creating blast furnaces. Win AMD.
- Won the design race. This one is definitely clear.
- Won the x86-64 race, to the point where the "never going to compete against Itanium, A64" has pretty much outmoded it wihtout ever directly competing.
- Won the innovation race. With 100x less R&D, AMD managed to build not only a better processor, but a better framework (hypertransport) than anything Intel is willing to attempt.
- Won the Intel looks stupid race for beating the crap out of their old master. When is the last time the clone has become the clonee in this industry? Intel is essentially going to be nothing more than an A64 clone before long. How long until those on the edges of the industry see this?

All around, Intel has been getting killed. For a company that is still making record profits, and spending money left and right - how is this happening. Furthermore, how long until all of these mistakes finally catch up to them. I suspect that in about six to ten months, some of the rosy financial news at Intel may begin to look a bit dull. Seriously, you cannot get beat this bad and hold on forever when you are still bickering and offering total crap.

I really think that Intel is riding a bubble right now, and it is just a matter of time before it pops. Unless of course, they can take corrective action NOW. From the looks of it, this is not happening.

 

[AnnoyedGrunt]

Intel made some poor architechtural decisions with the P4, and then some poor marketing decisions (IMO) by not making the P-M a desktop processor (yeah, it's slower in certain areas than a P4 and A64, but overall it's a great processor).

However, these shortcomings were "management decisions" and while costly, they often lead to new understandings in other areas.

I think AMD made better decisions (I'd like to think they listened to the engineers) and ended up getting a more robust and future-proof architechture the first time around. However, they are just now getting on the mobile bandwagon, and while their desktop chip was low power enough to be used as a mobile chip, I wonder how long that will last.

The problem is that Intel now will be focusing on lower clock speeds, but at the same time they have learned many tricks regarding increasing the clock speed. There may come a time when that knowledge becomes important once again.

Also, Intel has traditionally been setting the manufacturing curve, and is able to reduce process size and therefore cost sooner than everyone else. This gives them a large advantage in price strategy, that may come in handy if they start losing too much market to AMD.

Anyhow, I'm a proud owner of an A64 3200+ and I hope to see AMD keeping the pressure on Intel for several years to come. However, Intel has huge resources, and while it may take them some time to change direction, once they get their momentum they can make huge strides very quickly. AMD will need to keep choosing the "correct" architecture the first time through a design, because they can't afford to absorb a design change the way Intel can.

-D'oh!

 

[dmens]

Originally posted by: AnnoyedGrunt
The problem is that Intel now will be focusing on lower clock speeds

Nitpick on that statement, it is not exactly true. Clock speed still has a linear relationship with performance, there is no focus on "lower clock speeds", rather clock speed will no longer play as important a role in microarchitecture and pipelining as it did with the P4 designs.. but it remains a very important factor nonetheless.

Many of the tricks used to boost frequencies have a very low ROI from a power pov, so they will be abandoned with the new focus on efficiency, but sometimes it is worth sacrificing some power to gain slack on critical paths.

 

[fatty4ksu]

Trust me. Intel will be mighty, mighty fine.

 

[Hyperlite]

Originally posted by: AnnoyedGrunt
Intel made some poor architechtural decisions with the P4, and then some poor marketing decisions (IMO) by not making the P-M a desktop processor (yeah, it's slower in certain areas than a P4 and A64, but overall it's a great processor).

However, these shortcomings were "management decisions" and while costly, they often lead to new understandings in other areas.

I think AMD made better decisions (I'd like to think they listened to the engineers) and ended up getting a more robust and future-proof architechture the first time around. However, they are just now getting on the mobile bandwagon, and while their desktop chip was low power enough to be used as a mobile chip, I wonder how long that will last.

The problem is that Intel now will be focusing on lower clock speeds, but at the same time they have learned many tricks regarding increasing the clock speed. There may come a time when that knowledge becomes important once again.

Also, Intel has traditionally been setting the manufacturing curve, and is able to reduce process size and therefore cost sooner than everyone else. This gives them a large advantage in price strategy, that may come in handy if they start losing too much market to AMD.

Anyhow, I'm a proud owner of an A64 3200+ and I hope to see AMD keeping the pressure on Intel for several years to come. However, Intel has huge resources, and while it may take them some time to change direction, once they get their momentum they can make huge strides very quickly. AMD will need to keep choosing the "correct" architecture the first time through a design, because they can't afford to absorb a design change the way Intel can.

-D'oh!

Excellent points. While Intel appears to have some amazing products on the horizon, we have to keep in mind that AMD will not be sitting idle. AMD is growing quickly as it is, and as market share and revenue continue to increase over the next 18 months, R&D amount and speed will increase. Hopefully, AMD can do nothing but become even more competitive and continue to succeed. Just to cover my a$$, i have no binding alliance with AMD. I currently have a Northwood system and an Athlon XP system. Simply put, AMD currently has a supirior product and Intel is tripping over themselves. We will have to see how things pan out in the next few years. Regardless of the company that produces it, i will buy the product that best suits my needs and budget.

 

[Viditor]

Originally posted by: dmens
Nehalem won't have CSI? What?

Also, CSI is neither FSB, HT nor IMC, but it can be all three, or one, or two. Let's just say CSI shouldn't be bandwidth limiting, imo of course.

According to every roadmap and rumour published, Nehalem won't have CSI (it was supposed to have a 1200 Mhz FSB)...are you saying that you have inside information that this is incorrect?
While Whitefield was to be the first of the Nehalem chips to use both CSI and the joint Itanium/Xeon chip interface, it has been cancelled...

 

[TuxDave]

Originally posted by: Viditor

Originally posted by: dmens
Nehalem won't have CSI? What?

Also, CSI is neither FSB, HT nor IMC, but it can be all three, or one, or two. Let's just say CSI shouldn't be bandwidth limiting, imo of course.

According to every roadmap and rumour published, Nehalem won't have CSI (it was supposed to have a 1200 Mhz FSB)...are you saying that you have inside information that this is incorrect?
While Whitefield was to be the first of the Nehalem chips to use both CSI and the joint Itanium/Xeon chip interface, it has been cancelled...

Out of curiousity... which roadmaps has Nehalem in them? (aside from the old 10GHz one)

 

[Viditor]

Originally posted by: TuxDave

Out of curiousity... which roadmaps has Nehalem in them? (aside from the old 10GHz one)

That's just it...the old 10GHz roadmap had Nehelam as a very fast FSB design. It just doesn't make sense that it's now become a completely different architecture with the same name...
In addition, CSI isn't the type of thing you just "spring" on the server community...it takes ~2 years to qualify such a new design! And by that I mean it takes 2 years of public release where server builders can use and study and measure it, prior to actually putting it in servers...
If Intel released a CSI based platform tomorrow, we wouldn't see any sales till 2007.

 

[dexvx]

Originally posted by: Viditor
That's just it...the old 10GHz roadmap had Nehelam as a very fast FSB design. It just doesn't make sense that it's now become a completely different architecture with the same name...
In addition, CSI isn't the type of thing you just "spring" on the server community...it takes ~2 years to qualify such a new design! And by that I mean it takes 2 years of public release where server builders can use and study and measure it, prior to actually putting it in servers...
If Intel released a CSI based platform tomorrow, we wouldn't see any sales till 2007.

It does not take 2 years to get certification in the server community. It has taken AMD 2 years, but that was because Opteron was brand new to the market (it has also taken Itanium 2 years because it was brand new to its market). As for servers in the intended Xeon range, once third party Q&A gives it its certifications, it is more or less ready to go. Such certifications take a long time, but not 2 years of public release.

 

[TuxDave]

Originally posted by: Viditor

Originally posted by: TuxDave

Out of curiousity... which roadmaps has Nehalem in them? (aside from the old 10GHz one)

That's just it...the old 10GHz roadmap had Nehelam as a very fast FSB design. It just doesn't make sense that it's now become a completely different architecture with the same name...
In addition, CSI isn't the type of thing you just "spring" on the server community...it takes ~2 years to qualify such a new design! And by that I mean it takes 2 years of public release where server builders can use and study and measure it, prior to actually putting it in servers...
If Intel released a CSI based platform tomorrow, we wouldn't see any sales till 2007.

Ah, ok. Yeah, I last remember seeing Nehalem on that old 10Ghz roadmap. Haven't seen it much again so I was just curious if Intel released some new information on it.

 

[Zebo]

The article sounds pretty intelligent but don't be mislead. Intel will have conroe mid to late 2006 which does'nt need a intergrated memcontroller to compete. I expect it to be faster than equiv A64's running at same speed in about 75% of benchmarks. And more importantly it will ramp higher due to intels 65nm processes which anand has already given us a preview of. If intel releases is able to release 3.2 Ghz Conroe and we see OC's to 4Ghz... Good night AMD in retail/enthusiast sector.

AMD has problems past 2.8 and running on 90nm all next year. At best we will see a 3Ghz single core and 2.8Ghz dual core all next year.