About five years ago Intel introduced their (at the time) next-generation micro-architecture, what we've come to know as NetBurst. The NetBurst introduction actually took place over two IDFs, the first one was where we got brief glimpses of the new architecture and the second one where we got the full scoop on what was new. As it turns out, quite a bit was new back then, and almost all that was new has been since stripped out of the next-generation architecture that was first introduced yesterday.
Going along with history, Intel hasn't divulged too much about their next-generation architecture here at IDF, instead I'm expecting to get the rest of the story at Spring IDF 2006. That's unfortunate for those of us who want the information now, but then again we've got pretty much a full year between now and when Conroe, Merom and Woodcrest are actually released, so it isn't unreasonable. But I will say that at the very least it'd be nice for Intel to have at least released the name of the new architecture, because it's a pain writing next-generation and new architecture all the time :)
There have been some interesting morsels here at IDF already, despite the lack of information about the new architecture. If you're curious as to what they are, take a look at our IDF coverage on the front page or just head over to the Trade Shows section.
Every IDF I make it a point to stop by the Rambus booth at the Technology Showcase and say hi to a good friend, Dr. Steven Woo. We rarely talk about anything related to Rambus, but he's an incredibly bright engineer and a great person to talk to about just anything technology related.
Yesterday, on my bi-annual trip to the Rambus booth we talked a bit about Intel's new architecture. And the conversation helped me realize something, the transition to getting peak performance out of Intel's next-generation architecture may once again depend very heavily on the adaptation of Conroe/Merom/Woodcrest-friendly compilers. The idea is that it is going to be very difficult to feed a 4-issue core, and most of the code that has been compiled for the Pentium 4 or Athlon 64 won't be optimized to extract the levels of parallelism necessary for Intel's next-generation.
While I'm not sure if the battle for software optimized for Conroe/Merom/Woodcrest will be as tough as it was with the Pentium 4, simply because we are starting with a natively much higher IPC core, there will definitely be a struggle. But thanks to the nature of the architecture, even without optimized software, it shouldn't have a problem outshining the Pentium 4 (at least better than the Pentium 4 did the Pentium III...which it didn't really at first).
The other thing that I've been thinking about is an article I did a while ago where I compared clock-for-clock performance of the Athlon 64 and the Pentium M. In that piece I came to the conclusion that at the same clock speed, the Pentium M very close to the performance of the Athlon 64, despite the fact that AMD had an on-die memory controller. Areas where FP/SSE2 performance really mattered were the places where the Pentium M didn't fare so well, although Intel hopes to address those with the improvements they've scheduled for Yonah.
So I look at Conroe, Merom and Woodcrest as faster, higher IPC versions of Yonah, which is already a higher IPC version of Dothan. Thinking of the processors that way, it looks like we may actually see some pretty serious competition from Intel at the end of next year. And with AMD having no plans to update their micro-architecture in the next 1.5 years, the race may end up getting very close by this time next year.
The first keynote of today is just starting, so I'm going to get back to paying attention. More coverage later today.
Take care.
Going along with history, Intel hasn't divulged too much about their next-generation architecture here at IDF, instead I'm expecting to get the rest of the story at Spring IDF 2006. That's unfortunate for those of us who want the information now, but then again we've got pretty much a full year between now and when Conroe, Merom and Woodcrest are actually released, so it isn't unreasonable. But I will say that at the very least it'd be nice for Intel to have at least released the name of the new architecture, because it's a pain writing next-generation and new architecture all the time :)
There have been some interesting morsels here at IDF already, despite the lack of information about the new architecture. If you're curious as to what they are, take a look at our IDF coverage on the front page or just head over to the Trade Shows section.
Every IDF I make it a point to stop by the Rambus booth at the Technology Showcase and say hi to a good friend, Dr. Steven Woo. We rarely talk about anything related to Rambus, but he's an incredibly bright engineer and a great person to talk to about just anything technology related.
Yesterday, on my bi-annual trip to the Rambus booth we talked a bit about Intel's new architecture. And the conversation helped me realize something, the transition to getting peak performance out of Intel's next-generation architecture may once again depend very heavily on the adaptation of Conroe/Merom/Woodcrest-friendly compilers. The idea is that it is going to be very difficult to feed a 4-issue core, and most of the code that has been compiled for the Pentium 4 or Athlon 64 won't be optimized to extract the levels of parallelism necessary for Intel's next-generation.
While I'm not sure if the battle for software optimized for Conroe/Merom/Woodcrest will be as tough as it was with the Pentium 4, simply because we are starting with a natively much higher IPC core, there will definitely be a struggle. But thanks to the nature of the architecture, even without optimized software, it shouldn't have a problem outshining the Pentium 4 (at least better than the Pentium 4 did the Pentium III...which it didn't really at first).
The other thing that I've been thinking about is an article I did a while ago where I compared clock-for-clock performance of the Athlon 64 and the Pentium M. In that piece I came to the conclusion that at the same clock speed, the Pentium M very close to the performance of the Athlon 64, despite the fact that AMD had an on-die memory controller. Areas where FP/SSE2 performance really mattered were the places where the Pentium M didn't fare so well, although Intel hopes to address those with the improvements they've scheduled for Yonah.
So I look at Conroe, Merom and Woodcrest as faster, higher IPC versions of Yonah, which is already a higher IPC version of Dothan. Thinking of the processors that way, it looks like we may actually see some pretty serious competition from Intel at the end of next year. And with AMD having no plans to update their micro-architecture in the next 1.5 years, the race may end up getting very close by this time next year.
The first keynote of today is just starting, so I'm going to get back to paying attention. More coverage later today.
Take care.
Facebook
Twitter
RSS
9 Comments
View All Comments
knitecrow - Monday, August 29, 2005 - link
AMD’s success is in large part due because of Intel’s miscalculations which, in turn, are a result of Intel letting their marketing department dictate processor design. Hence, the insane netburst architecture built around the "gigahertz" myth to fool consumers. Intel has never lacked good engineers or good ideas. Internal politics seem to get in the way. I remember Bob Colwell talking about moving north bridge stuff into the cpu die.... and the chipset guys where like "what you trying to do, I’ll loose my job."Now that Intel is focused again, AMD is going to face serious problems. The real question is... what has AMD been working for the past 3 years? All I have seen so far is minor tweaks of the original clawhammer.
I seriously hope AMD has been working some new technology or else I fear this will be a tale of the grasshopper and the ant.
creathir - Thursday, August 25, 2005 - link
What is with the Texas cities??And the VIIV logo looks like the SA Bus System (VIA) logo.
http://www.viainfo.net/Ride/Default.aspx">http://www.viainfo.net/Ride/Default.aspx
Strange.
- Creathir
GTaudiophile - Thursday, August 25, 2005 - link
Anand: Could you look into this and write a brief article?http://forums.anandtech.com/messageview.aspx?catid...">http://forums.anandtech.com/messageview.aspx?catid...
Anand Lal Shimpi - Thursday, August 25, 2005 - link
I'll play with it when I get back to CT this weekend.Take care,
Anand
HardwareD00d - Friday, August 26, 2005 - link
that would be great! I'm having the same problems with game speedups and studdering on my X2-4800+. GTA San Andreas and Postal2 get really screwy after a few minutes of gameplay and rebooting is the only thing that fixes it.kleinwl - Wednesday, August 24, 2005 - link
It's good to hear that Intel is making real progress in updating there chips. As an (Forced=Dell) Corporate User, I've been disappointed by the poor performance per dollar of the Intel system. Our corporate police spend the least amount of money on systems (including CAD/CFD) so a cost-performance competitive Intel chip will make things much easier buying needed equipment. As it is we are running single Xeon Dell workstations, but paying the same cost as if we were getting a dual-core Optieron. It's very sad.Now if we can only get the IT police to upgrade from Windows 2000 to XP-64, we could make some progress feeding our ram intensive programs.
GhandiInstinct - Wednesday, August 24, 2005 - link
"It's good to hear that Intel is making real progress in updating there chips"Well, son, it's been a bit more than five years so I believe we should be expecting a bit more than just "progress" on that front. I mean this is the next generation, or so they're calling it. I was expecting a revolution, a NiNtenDo Revolution. = )
ceefka - Friday, September 2, 2005 - link
It's good that Intel is making steps in the right direction, but we'll have to see if they will come back with a vengeance. Perhaps after the introduction of Intel's new generation will 4-way Opterons finally meet some real competition. Scalability wasn't very high on those Xeons.Even so,
1. AMD will still have reasonable headroom to raise the frequencies once they are on 65nm.
2. They will also introduce 4-core architecture with perhaps even an upgraded memory controller, the latter being my wild guess.
3. HTT is said to be highly scalable, so I expect that to hit 2GHz with X4.
4. New chipsets could still unleash power from even current AMD64 systems. It's not only the CPU that determines the strength of a system.
5. How many (home) apps are going to take full advantage of 4 core architecture in order to express this technical progress.
jeffrey - Wednesday, August 24, 2005 - link
First post.What are the chances of RMBS used in desktop graphics cards?