Intel to beat AMD by 20%

[Duke3d87]

I think Intel caught up nicely. AMD really pulled ahead but Intel's newest chips are really competitive. I buy what performs the best though and I guess we'll just wait and see.

http://news.zdnet.com/2100-9584_22-6043587.html

AMD does not believe that Conroe is really that good. The believe that they still have the upper hand. They still belive that their integerated memory controller is better. The thing is, the Conroe has a 4 MB L2 cache, and that helps performance, as from what i can see, stuff that does not fit in L2 caches goes to ram. So, being that they have shared L2, i would guess that the 4 MB L2 cache allows stuff to fit in the cache for Intel that would AMDs would require to go to memory...

 

[robberbaron]

AMD is planning on upping the FX dual cores to 2mb IIRC. That would be 2mb per core, and the cache coherency works across the chip, almost effectively a 4mb shared L2 cache, heh.

 

[covertclocker]

http://news.zdnet.com/2100-9584_22-6043587.html

AMD does not believe that Conroe is really that good. The believe that they still have the upper hand. They still belive that their integerated memory controller is better. The thing is, the Conroe has a 4 MB L2 cache, and that helps performance, as from what i can see, stuff that does not fit in L2 caches goes to ram. So, being that they have shared L2, i would guess that the 4 MB L2 cache allows stuff to fit in the cache for Intel that would AMDs would require to go to memory...

Just because you stick 2 more MB of cache on the chip, it doesn't mean that Intel will be able to access the memory any faster. Without an on-chip memory controller, the Intels front side bus is still going to get congested with crap. When comparing current 200 AMD FSB with Intel's 333 FSB, there is no comparison. Even with the slower FSB, you have the on-board mem controller and the HT, it still blows Intel out of the water. This has been proven time and time again. Intel's arch is old skool, AMD is new skool. And for everyone touting Intel's gains by going to the 65nm process, you have to understand that AMD is on the 90nm process. Once AMD gets on the 65nm bandwagon with DDR2, it will be AMD - 1 Intel - 0

 

[markku]

AMD is planning on upping the FX dual cores to 2mb IIRC. That would be 2mb per core, and the cache coherency works across the chip, almost effectively a 4mb shared L2 cache, heh.

actually l2 cache size for "new" fx seems to be still 2x1M but they'll add 4M of L3 to some opterons and eventually, to fx..

as for conroe being 20% faster than amds offering.. realistic estimate.
it's gonna be 2.8 FX vs 3.33/1333 conroe EE (rumours) and 2.6 X2 vs 2.66/1066 conroe.
just look at yonah.. with it's relatively low clockspeed and fsb speed it's doing quite fine.
now add 2x cache and increase fsb by 1.6x.. plus any other improvements..

btw this 4MB L3 FX move kinda reminds me the first EE cpu..

 

[Nman]

Well come next week or so it should be very informative for both sides. With what is known about thier next gen. I don't think that amd's on die controller will be all that advantageous. When you look at the reviews that have come out as of yet with even the yonah core on what I wouldn't consider top end chipsets. It still does very well in most areas. The area I am looking forward to hearing about hopefully next week is any work they have done in the fp arena. I am hoping that they did some major restructuring to boost fp. If they can do that they have a good chance of at least breaking even. While it is fun to hear all the rhetoric from both sides... I want to see actual reviews before making any decision. I think that they could do the 20% if they included something along the lines of spec precomp. Is it included? We'll see. Hopefully they elucidate some of the finer points next week at idf.

 

[Duke3d87]

Just because you stick 2 more MB of cache on the chip, it doesn't mean that Intel will be able to access the memory any faster. Without an on-chip memory controller, the Intels front side bus is still going to get congested with crap. When comparing current 200 AMD FSB with Intel's 333 FSB, there is no comparison. Even with the slower FSB, you have the on-board mem controller and the HT, it still blows Intel out of the water. This has been proven time and time again. Intel's arch is old skool, AMD is new skool. And for everyone touting Intel's gains by going to the 65nm process, you have to understand that AMD is on the 90nm process. Once AMD gets on the 65nm bandwagon with DDR2, it will be AMD - 1 Intel - 0

once it's AMD = 65nm, Intel will start ramping up the 45nm process. Why do you think the XeonMP and the Itaniums have L3? Becuase they are bottlenecked processors. Anandtech once did an article on the XeonMP and talked about how L3 sat between L2 and the FSB and sped up access to the ram. It makes sense. And the architecture that Intel will be using is "new skool"

 

[robberbaron]

btw this 4MB L3 FX move kinda reminds me the first EE cpu..

Except that the cache would actually add up, whereas the first EE really had about 1.5mb effective cache

 

[Legenic]

Just because you stick 2 more MB of cache on the chip, it doesn't mean that Intel will be able to access the memory any faster. Without an on-chip memory controller, the Intels front side bus is still going to get congested with crap. When comparing current 200 AMD FSB with Intel's 333 FSB, there is no comparison. Even with the slower FSB, you have the on-board mem controller and the HT, it still blows Intel out of the water. This has been proven time and time again. Intel's arch is old skool, AMD is new skool. And for everyone touting Intel's gains by going to the 65nm process, you have to understand that AMD is on the 90nm process. Once AMD gets on the 65nm bandwagon with DDR2, it will be AMD - 1 Intel - 0

this is really ignorant imo. no offense.

1. I agree that the integrated memory controller is superior to fsb design, but that doesn't mean intel still can't do good things without it. look at core duo for example, it's performance is practically on-par with x2's at the same clock speed, all without an integrated memory controller. fsb will hurt intel most in quad core plus server environments, but other than that I'm willing to bet they pull through fine.

2. everyone assumes that amd is far superior because they're matching intel on a larger manufacturing process. amd 90nm = intel 65nm? that's bs. the reason is because of architecture and clock speed.

lower clock speed = lower power consumption.

look at the pentium m for example. it had no issues on 90nm like the p4 did, because of it's design. similar to A64s. lower clock speed and more efficient architecture.

I'm willing to bet when amd goes 65nm you'll see little performance increase. what you will see is amd able to scale to higher clock frequencies, and manufacturing chips for less cost (more cores per wafer).

 

[elm669]

I've been out of the intel loop for a while now but is netburst dead yet? If it's not dead when will the P6/ Pentium-M based desktop chips come out?

 

[robberbaron]

I've been out of the intel loop for a while now but is netburst dead yet? If it's not dead when will the P6/ Pentium-M based desktop chips come out?

They won't really make desktop chips based on that. They'll be coming out with NGMA which will replace both Banias and Netburst.

 

[empoy]

Just because you stick 2 more MB of cache on the chip, it doesn't mean that Intel will be able to access the memory any faster. Without an on-chip memory controller, the Intels front side bus is still going to get congested with crap. When comparing current 200 AMD FSB with Intel's 333 FSB, there is no comparison. Even with the slower FSB, you have the on-board mem controller and the HT, it still blows Intel out of the water. This has been proven time and time again. Intel's arch is old skool, AMD is new skool. And for everyone touting Intel's gains by going to the 65nm process, you have to understand that AMD is on the 90nm process. Once AMD gets on the 65nm bandwagon with DDR2, it will be AMD - 1 Intel - 0

by the time AMD hits 65nm bandwagon Intel's process will be 45nm
when we talk about microarchitecture, it does not revolve around the FSB
and the "on-board memory controller".

 

[Duke3d87]

by the time AMD hits 65nm bandwagon Intel's process will be 45nm
when we talk about microarchitecture, it does not revolve around the FSB
and the "on-board memory controller".

It has been said that Intel's cloverton will not be as fast as Intel would like people to believe it is. Is that true?