News Conroe coming sooner than you think

[WilHarris]

http://www.bit-tech.net/news/2006/04..._release_date/

[The Bart Man]

Nice. I alreaddy a long time waiting fore a new sstems. First thought of stikking to AMD but since the Conroe is a better performer for less and less watts it will be a simple dicission!
As long as there will be some nice Mobo's with nice ati chips fore crossfire support I like it!

[RotoSequence]

Hmm, looks like Intel is going to hold the performance crown for a bit longer than originally anticipated (provided its actually able to get parts out in volume).

This is going to be interesting

[LoneArchon]

I wait till it comes out to compare the pricing. I thinking these chips will start around $400. If that is true then i going for AMD for the lower price

[Bindibadgi]

The E6700, which is a 2.67GHz chip with 4MB of cache and a 1066 front side bus will cost $529 while the E6600, a 2.40GHz version will cost $315. The other two members of the Intel Cointreau family have 2MB of cache apiece, with the E6400 (2.13GHz) costing $240, while the E6300 (1.86GHz) will cost $210.

From the inq. The 2.4 4meg or the 2.13 2meg looks to be the most "middle ground affordable". Whether an extra meg of cache per core makes a lot of difference and is worth the extra $100 would be nice to know.

[RotoSequence]

I have a feeling it will; unlike AMD, Intel doesnt have a super high bandwidth bus with memory bandwidth out the wazoo. As such, the extra L2 (What one AMD employee has called "compensation cache") will probably make a fairly substantial performance difference, as Conroe will be chewing through information rather quickly - I would imagine much more so than the previous Netburst architecture

[The Bart Man]

Also the Conroe is next gen. wille the AM2 is just a socket 939 with DDR2 support. So probably more new good features!

[Bindibadgi]

Originally Posted by RotoSequence

I have a feeling it will; unlike AMD, Intel doesnt have a super high bandwidth bus with memory bandwidth out the wazoo. As such, the extra L2 (What one AMD employee has called "compensation cache") will probably make a fairly substantial performance difference, as Conroe will be chewing through information rather quickly - I would imagine much more so than the previous Netburst architecture

It DOES have a super-high bandwidth bus though, that's what the 1066/1333 bus IS. It's only because OF netburst they had to have compensation cache, but now with effectively a "P3/P-M" pipeline lengh AND uber fsb, will >2meg cache make that much of a difference? At what point do you get diminishing returns?

[RotoSequence]

Originally Posted by Bindibadgi

It DOES have a super-high bandwidth bus though, that's what the 1066/1333 bus IS. It's only because OF netburst they had to have compensation cache, but now with effectively a "P3/P-M" pipeline lengh AND uber fsb, will >2meg cache make that much of a difference? At what point do you get diminishing returns?

Its still not as uber as hypertransport. DDR2 offers something like what, 8.5 GB/s? The net available speed for the entire bus will be something like 10 GB/s; AMD will have that entire 8.5 GB/s on the memory controller, or whatever the bandwidth of DDR2 is, and 22.4 GB/s + on the hypertransport link.

AMD has a lot more BUS than Intel; MHz arent everything, neither on die or on the aging FSB of netburst - It can probably fill two processors fine, but any more than that, or a heavy load, and its going to be a bit choked I think.

[Kipman725]

I hear alot of speculation, I will wait untill the actual chips are released before deciding if one is faster than the other :P

[Bindibadgi]

Originally Posted by RotoSequence

Its still not as uber as hypertransport. DDR2 offers something like what, 8.5 GB/s? The net available speed for the entire bus will be something like 10 GB/s; AMD will have that entire 8.5 GB/s on the memory controller, or whatever the bandwidth of DDR2 is, and 22.4 GB/s + on the hypertransport link.

AMD has a lot more BUS than Intel; MHz arent everything, neither on die or on the aging FSB of netburst - It can probably fill two processors fine, but any more than that, or a heavy load, and its going to be a bit choked I think.

There's pro's and con's to everything, Roto. If any of the PCIe/PCI cards (especially gfx) want to access memory through DMA they only have to navigate the chipset, whereas on an AMD system they have to go chipset-cpu-memory-cpu-chipset. Indeed having a direct memory access for the most important component - the cpu - makes for much better performance and less cache need, that's true, but whether 4meg will be useful over 2!? Are the 9xx series really that much faster than the 8xx? And then add in a far shorter pipeilne?

[MiNiMaL_FuSS]

well intel might finally take back the performance crown with this one....but i reckon AMD will still win on the bang-per-buck front, and that gets my vote.

[RotoSequence]

Originally Posted by Bindibadgi

There's pro's and con's to everything, Roto. If any of the PCIe/PCI cards (especially gfx) want to access memory through DMA they only have to navigate the chipset, whereas on an AMD system they have to go chipset-cpu-memory-cpu-chipset. Indeed having a direct memory access for the most important component - the cpu - makes for much better performance and less cache need, that's true, but whether 4meg will be useful over 2!? Are the 9xx series really that much faster than the 8xx? And then add in a far shorter pipeilne?

more like 1 meg over 2; its 4 megs of cache shared between two cores after all. The difference on netburst between 512 and 1 meg was fairly substantial; I suppose it wont make as much a difference, but I still think it could be fairly substantial.

[Bindibadgi]

Well no because the prescott was introduced with 1meg cache which had a completely different pipeline to the northwood with 512k.
And sure, it's shared and better managed but I still think it'll be dimishing returns. It will definately be BETTER, but not by anywhere near twice as much. Look at the difference between pentium M banias versus dothan: 1meg versus 2. You increase the cache size you increase the latency of finding what you need as well, despite the fact it's smaller on a 65nm.

AMD will definately win on bank per buck still. But I still want a Conroe.

[RotoSequence]

Doesnt everybody?

[Tim S]

The problem that Intel currently has isn't related to bandwidth, it's related to latency because the memory controller isn't located on-die. Thus, I think more cache will make a difference, as it's essentially more low-latency memory - the more of that that Intel has, the more likely they are to avoid latency related performance issues.

[Solidus]

Well this has put my building a pc project in turmoil. I had planned to build a socket 939 computer, totally specced out, i was two months away from making it and now....should i just wait that little bit longer? im going to be spending a fortune so might aswell go value for money by waiting? aaaaargh!!

[JADS]

I thought 3GHz and 3.33GHz were being mooted as the top end for the Conroe line with 2.67GHz being a mid range chip?

Well it should be interesting to see some real performance figures.

[Bindibadgi]

Originally Posted by bigz

The problem that Intel currently has isn't related to bandwidth, it's related to latency because the memory controller isn't located on-die. Thus, I think more cache will make a difference, as it's essentially more low-latency memory - the more of that that Intel has, the more likely they are to avoid latency related performance issues.

Well, true, but 4 meg on just 2 cores? It just doesnt seem like it needs huge amounts of cache because despite the latency issue, it has a HUGE amount of bandwidth to play with which is the trade off. Sure, the 4 core jobbies you're gonna need an uber amount of cache because they'll all be sharing a single memory bus! But 2 core, Im not sure itll make that much difference if they were clocked both the same. 5-12% tops

[Cthippo]

Originally Posted by RotoSequence

Doesnt everybody?

No, Intel is going to have to be a LOT better than AMD before I'll switch back. I've had too many problems with stability in Intel based machines to trust them anymore.

That said, i am curious what's coming down the pipeline