Intel gets back on their horse - several months from now

[Fozzik]

The Pentium is dead, long live Core!

Intel announced today at the Spring 2006 Intel Developer forum that their new architecture would be known simply as Core. I'm not sure they could have thought of anything more boring...it's like naming your dog... dog.

But it is looking like the name is definitely going to be the only boring thing about these new *cough* Core cores.

I'll get over the name in a bit. Although we're going to hear much more in-depth details about the new architecture over the next couple days (and mostly likely see many more benchmarks), the first numbers are in. A few simple details about the new architecture were also released, to go with what we have already known for some time.

These new processors are based on the Pentium M architecture which has really been the one strong point in Intel's lineup for quite a while now. Good for them figuring out where the solid product was, and running with it. The new architecture will be available in server, desktop, and mobile chips...and all seem to show a vast improvement in performance with an almost equally vast decrease in power consumption. In short, this architecture isn't just an small improvement over the current Pentium... it's a full 360 degree roundhouse kick to the face.

New features we know about so far (yes, I know this is marketing speak, and doesn't mean a whole lot...more coming soon).

Four issue-wide, 14-stage pipeline. This is wider than the netburst architecture, with much fewer stages...which will allow for much more agility, and higher instructions per clock...similar to the Athlon 64.
Shared Cache. The new Core architecture uses a single large cache for both cores.
Micro-fusion. This is the ability to take micro-operations (small bits of things the processor is working on) and combine them in various ways to run at the same time...the Pentium M has had this all along.
Macro-fusion. This one is new to the Core architecture, and most likely just means combining instructions on a larger scale than micro-fusion... basically filling up the available resources on the processor more efficiently each clock.
Single-clock execution of SSE instructions. All SSE instructions are now run in a single clock, which should speed up... well...SSE instructions and programs that use them. ;)


The first performance numbers are up at Anandtech here (http://www.anandtech.com/tradeshows/showdoc.aspx?i=2713) (thanks rabb1t), and in the name of all that's holy...it appears Intel has done it. We'll have to reserve final judgment of course until more independent sources can verify...but it looks like Intel is back on their horse and riding hard. AMD could be in big trouble in all areas of desktop performance for quite some time if these numbers hold up.

Keep in the back of your mind as you get washed in the beautiful hype and massive numbers over the next few days, that none of these chips are going to be available for sale for several more months. Intel is announcing and hyping nice and early. We won't call it a paper launch, since only graphics card companies do those, right?

Way to go Intel. Now if they can just get passed the empty feeling in their stomachs... knowing that their unimaginable amounts of highly-paid engineers didn't design this core. They bought a company in Israel who designed it...but hey, whatever works. *sorry, had to get my shot in as a wounded AMD fan...things are looking a bit bleak for the underdogs at the moment* :D

[Traldan]

Sweeeet. I know what my next processor's probably going to be. :p

[rabb1t]

But it is looking like the name is definitely going to be the only boring thing about these new *cough* Core cores.

No no no. I believe it is pronounced like this (http://www.everbecoming.com/temp/core.wav) :p

[rabb1t]

Shared Cache. The new Core architecture uses a single large cache for both cores.

Hasn't this been mentioned as a bad thing by most reviewers though?

Single-clock execution of SSE instructions. All SSE instructions are now run in a single clock, which should speed up... well...SSE instructions and programs that use them. ;)

um... um... *blink* um... but what does that mean?

- - - -

The Anandtech article mentions it is running on an LGA775. Hopping over to NewEgg to make a "wish list" *sighs at tech addiction* I notice that currently the boards only go up to DDR2-800. Isn't the board capable of 1 gig speed ram? I know somewhere I saw that OCZ just announced they have 1 gig low latency ram. Can't seem to find the link though. Here is an announcement on OCZ's site (http://www.ocztechnology.com/aboutocz/press/2006/153)

Looks like Corsair has one out too (5-4-4-9 latency).

[Giftmacher]

Yup agree with you about the duff name, for example "a dual core Core processor" not slipping off the tongue is it?

As for all the improvements... well I wonder what AMD's answer to this will be? Could get interesting now the opening salvo of another processor war has been fired.

For now though this looks like a really positive move by intel.

For a start I'm glad to see power consumption, and therefore heat go down. I just about lost interest in intel chips when the Prescotts came out, far too hot IMO, cases are crammed with ever hotter components as it is without cramming the CPU equivalent of a nuclear reactor in there. (Ok I exaggerate a bit :p) Anyway cooler is better.

Other things I'm not so clear about though:

Intel seem to be sticking with off chip RAM control. Anyone know why? I thought doing this introduced a major bottle neck, am I misinformed?

Dual/Quad cores etc. We seem to have topped out in the GHz stakes now and I noticed the "Kentsfield" (presumably for servers?) quad core chip in that link, perhaps where servers go desktops will follow... This has me wondering what it will all mean for future OSs and apps? Sticking multiple cores in a machine won’t necessarily make it go faster unless the software is geared to take advantage of the extra power, so are we going to see software effectively rewritten every few years to cope with extra cores? (at least until we manage to break silicon imposed limitations.)

Gift.

[Razorwire]

The Pentium is dead, long live Core!
About freaking time, how long have we had the lame pentium name and can we lose the celery chips as well?

Way to go Intel. Now if they can just get passed the empty feeling in their stomachs... knowing that their unimaginable amounts of highly-paid engineers didn't design this core. They bought a company in Israel who designed it...but hey, whatever works. *sorry, had to get my shot in as a wounded AMD fan...things are looking a bit bleak for the underdogs at the moment* :D
Intel hasn't won until you can do a side by side shopper compaision ;)

[Fozzik]

Hasn't this been mentioned as a bad thing by most reviewers though?


No. If you saw that somewhere, that person was mistaken. sharing cache is a good thing when we're talking multiple cores. The more high-speed memory an individual core can access directly, the better.


um... um... *blink* um... but what does that mean?



SSE is a set of instructions designed to speed up multi-media type tasks like encoding, applying filters, etc. SSE was designed by Intel, as a way to try and separate themselves from the competition. They built hardware support for these special instructions, and then tried to get software makers to use them (being the 800-pound gorilla helps a little, but not too many applications make heavy use, even today). AMD began incorporating support for SSE in hardware...and for quite a while now both AMD and Intel have had equal support for SSE. This new core basically makes these instructions run faster, by being able to process them in a single clock cycle (most took two clock cycles to run in the past).

This has me wondering what it will all mean for future OSs and apps? Sticking multiple cores in a machine won’t necessarily make it go faster unless the software is geared to take advantage of the extra power, so are we going to see software effectively rewritten every few years to cope with extra cores? (at least until we manage to break silicon imposed limitations.)

Software is already being re-written and redesigned to allow much more parallelism. It won't be a complete overhaul each time more cores are added to a CPU...because once applications are written to utilize multiple simultaneous threads, more cores will just speed things up. It is a massive change for the software side, but it was about time for programmers to take more responsibility for writing efficient and optimized code (not saying everyone was bad at this, but a lot of the big players were just adding more and more bloat and letting the faster processors sort it out). Increases in computer speed over the next few years will be much more dependant on how well the software is written.

I too definitely will reserve judgment at least until these new chips are released. There are still a lot of questions to be answered...and I'm sure by the time Conroe actually hits retail, we'll have heard something from the AMD side about how they plan to strike back.

Quad core, at the moment, is planned by both AMD and Intel for 2007, and will come to the desktop sometime next year. AMD is pretty much ready, and we might see quad-core sooner if there isn't any other way they can answer the performance of Intel's new Cores *cough*.

Intel seem to be sticking with off chip RAM control. Anyone know why? I thought doing this introduced a major bottle neck, am I misinformed?

An on-die memory controller is much better from a latency standpoint...but it also means a lot of architectural changes, and a decent amount of die space. Intel probably doesn't want to do it partly because AMD did it first, and they hate appearing to be following their much smaller rival (although they often are)... but it's also most likely a matter of money. Design changes like that cost big money, and it also means pulling a big part out of the motherboard chipset...and decreasing the possible performance differences between chipsets. Intel also makes chipsets, so they have to worry about robbing Peter to pay Paul (as it were).

[Giftmacher]

Thanks for that Fozzik, all is clear now. Really good point about Intel manufacturing mobo chipsets too, when you put it like that it makes a lot of sense.

Gift.

[rabb1t]

FATALITY!!!

As if more power wasn't enough, Intel peeps have announced they have a liquid system that "will enable the latest Pentium Extreme Edition chips to hit 5GHz with ease" - not only that, the price, ~$50!

:eek:

http://www.bit-tech.net/news/2006/03/08/intel_liquid_cooler/


/cry /cry

It is going to be very hard to resist upping to Conroe, esp. if it releases right around Vista. (It would be a lot easier to upgrade the core then install Vista than to do it later. :rolleyes: )

[rabb1t]

Another set of marks for Conroe. This time vs. OC FX-60, standard FX-60, standard P4(?) XE 955.

http://www.bit-tech.net/news/2006/03/08/intel_conroe_benchmark_fear/

Oh um... guess it is just that one page heh.

[Fozzik]

Looking at those results, you can already see there's something mildly fishy going on with the Intel-configured machines. I don't think there is any way the benchmarks we've seen are unbiased, since they were completely set up and controlled by Intel. There's no way an overclocked FX-60 should perform worse than a stock FX-60.

I'm guessing things will look closer when we see real independant benchmarks...but Intel definitely has a lead that will require some serious attention from AMD. They are going to have to pull out all the stops to close this gap. No matter how you look at it...Intel's got a good, smart architecture on the desktop for the first time in quite a few years.

No matter my feelings for Intel as a company, it's very nice to see netburst die...and competition for quality architecture to begin again.

[Nubb]

I agree with Fozz, initial red-flag waved when I read "Intel configured AMD...".

Plus the systems don't look fair enough to me (not much info about the systems in the first place). Xfires for Intel, but what did the AMD run? Memory is at odds too, all I see is it was on DDR400 for AMD and Intel's system on DD2.

Definitely this new core is going to be fun to watch! I just hope AMD has something special up it's sleeve to surprise us all in the near future -- and squash the inferior brand!

Thanks,
Nubb

[rabb1t]

More marks... Conroe still 'kicking @zz and taking names'.

http://www.anandtech.com/tradeshows/showdoc.aspx?i=2716&p=1

[Giftmacher]

That's impressive stuff, I really can't wait to see what AMD do about this. Conroe really does look powerful in those benchmarks and I doubt AMD will take that kind of beating sitting down. The next couple of months are going to be an interesting watch I think.

Gift.

[rabb1t]

I won't have the $ to buy into Conroe unless I get a super good job... /cry /cry /cry... but this has been in the back of my mind since I saw it...

http://www.anandtech.com/tradeshows/showdoc.aspx?i=2716
Intel also made it a point to mention that by the time Conroe ships DDR2-800 will be the memory of choice, however dual channel DDR2-667 already offers more memory bandwidth than Conroe’s 1066MHz FSB can use so the fact is meaningless.

Questions...


What exactly does that mean?
If the DDR2-667 can spit out info faster than the Conroe can absorb it, why add on the DDR2-800 capability?
Does this mean the same for an AM2 socket? I know they handle memory differently, but I'd think if that bandwith were too high for Conroe wouldn't that hold true of AM2?

[Fozzik]

The difference is the onboard memory controller on AMD chips (or that's the biggest difference). The front-side bus is the limiting factor for the Intel processors. More memory bandwidth doesn't do any good when the bus between the processor and the memory controller is saturated. The lowest latency on a traditional platform (with the memory controller on the north bridge) is when the front-side bus and memory bus are running at the same clock. Most northbridge memory controllers contain their own caching to hang on to things until the front-side bus opens up for more info... so higher speed memory can give a bit more performance.

In Intel's case, it will take bumping up the clock on the FSB to really utilize DDR2-800. Otherwise, it's just a marketting gimick... to sucker in those that think that faster memory automatically means more performance.

[Razorwire]

So from what I am seeing on these tests we have Core vs 939 chips right? Have we seen the performance stats of Core vs AM2 gen CPUs? That is sort of like compairing a CBR600 vs my little no brand 125, totally different classes of chips. Need to get Gen vs Gen compairisions before we can have any idea of the real next gen winner.

[rabb1t]

I think Intel hasn't compared to the AM2 'cause they couldn't get one. (They set up the Conroe vs. 939 tests we've seen so far.) I don't think a Conroe vs. AM2 comparison can appear until both are released to reviewers.

Shouldn't be long though. AM2 is due for actual release in 3 months, and Conroe in approx 5.

[rabb1t]

um... words... and more words... there are a lot of words here... (http://www.tomshardware.com/2006/03/13/idf_spring_2006/)

[Giftmacher]

Well I didn't understand large portions of that but wow Kentsfield is for desktops! Multiple (as opposed to just two) cores are a lot closer than I realised.

I'm still not likeing keeping the memory controler seperate, it's good for integrated graphics be damned! But seriously, I think they're going to have to work hard to solve this as a issue because the more cores they dump on a chip the bigger the nightmare will get. Intel had better get clever in this regard, or else buy another clever company :)

Also, was interested to see flash drive assistance for boot up, nice if a little limited in scope and application right now.

Gift.

[rabb1t]

Looks like Conroe isn't likely to make the previous Q3 release date. I just found an article that mentions it is being tied to the Vista launch, and as such, might be an early Q4 release. (There was a recent 'leak' that Vista has been bumped back to November.)

[rabb1t]

Looks like the 1st gen Conroe cores are still expected to hit in Q3, as Q1'07 is slated for the quad cores; Yet another point in Intel's favor.

http://www.dailytech.com/article.aspx?newsid=1471

Guess I won't be so sad about it being very unlikely to upgrade to Conroe soon. Guess this means I can just target Q3'07 and move straight to quad core. :p

[Giftmacher]

Hmmm Quad core goodness...

I've got to get a better job by the new year! :D

Gift.

[Annuna]

I still find those benchmark tests to be completely worthless until performed by a trustworthy 3rd party with truly comparable AMD products. Intel could claim their chips were running at 30 petahertz for all I care, until its delivered and validated independantly, Intel is still behind in the race, as they have been the last couple years.

I think this is a desperate Intel's attempt at stirring up excitement because they have indeed been hammered the last couple of years by AMD. I don't doubt their new chips will be a vast improvement, but will it really live up to what they are claiming? If so, why did they have to perform the benchmarks themselves?

[rabb1t]

http://www.dailytech.com/article.aspx?newsid=1619

The 2.67GHz Conroe CPU featured in the IDF benchmarks last month will have an official MSRP of $530 at launch.

I have no idea what this one means... SuperPi in 21.454 seconds?
http://www.xtremesystems.org/forums/showthread.php?t=95021