Intel Core 2 Extreme X6800 Preview from Taiwan
by Anand Lal Shimpi & Gary Key on June 6, 2006 7:35 PM EST- Posted in
- CPUs
Memory Latency and Bandwidth
We've never been able to look at some of the low level characteristics of Intel's Core architecture, and although we didn't have enough time to do a thorough run of low level benchmarks we were able to run ScienceMark 2.0 in order to get an idea of how the Core 2 Extreme stacked up against the FX-62 in terms of memory latency and bandwidth.
We had seen Conroe performance results that showed the new architecture being able to offer fairly competitive memory access latencies to AMD's architecture, without the need of an on-die memory controller. Our ScienceMark 2.0 results confirm just that:
While AMD still offers lower memory latency, the Core 2 Extreme X6800 is very close in comparison - especially considering that it has no on-die memory controller. With lower clock speeds than its Pentium D siblings and a faster FSB, memory access latency is reduced tremendously with Conroe. On a larger scale, through a very effective cache subsystem as well as memory disambiguation, Conroe can offer significantly improved memory performance compared to its predecessors, including the Athlon 64 X2/FX.
ScienceMark's memory bandwidth results offer a very telling story, showing us the bandwidth limitations of Intel's FSB architecture. While the FX-62's peak theoretical bandwidth is not achieved in real world, you can see how AMD's Direct Connect architecture offers higher limits for chip-to-chip communication.
Facebook
Twitter
RSS
134 Comments
View All Comments
bob661 - Tuesday, June 6, 2006 - link
He's probably using Opteron vs Zeon benchmarks where the Zeon is held back by its FSB when scaling to more CPU's. I don't see why the same wouldn't apply to Conroe vs FX62's.peternelson - Wednesday, June 7, 2006 - link
Yes that's what I'm talking about. For every opteron you add you get more HT channels which can each connect one (or more using tunnel passthrough) I/O chips.
Whilst not in the opteron league, if 4x4 connects the processors by hypertransport as well as the chipset, it is reasonable to assume it works on either two HT links per socket, or splitting the HT between two peers.
Therefore the second AM2 socket can have it's "empty" HT connection talking to a SECOND I/O chipset. For that reason it has POTENTIALLY say double the I/O performance of a single Intel FSB.
Another point about this comparison generally is that these conroe benchmarks are of the EXTREME edition. That is unusual in the property that it has especially extra fast FSB compared to mainstream conroe. Therefore in benchmark comparisons of the non-extreme conroes, the fsb would be slower and may form a constraint which would alter (probably reduce) the relative performance difference between AMD and Intel seen here. The benchmarks will reveal.
zsdersw - Wednesday, June 7, 2006 - link
The EE used was on a 1066 fsb.. same as mainstream Conroe.goz314 - Tuesday, June 6, 2006 - link
Apples and Oranges... It will also cost vastly more than a Conroe-based system. Even if the performance of an AMD 4x4 platform is significantly higher, Intel would still have a significant edge in a performance/dollar analysis.
4x4 is currently vaporware and it's at least a quarter behind Intel's comparable quad-core processor roadmap. Conroe, on the other hand, is launching next month and will be available for a year before a 4x4 platform even sees the light of day.
Real hardware will always beat an marketing engineer's 'idea' no matter how good that idea looks on paper.
bob661 - Tuesday, June 6, 2006 - link
Can I quote you on this and use it against the hypocrites? BTW, you got a link on where I can buy a Conroe?Josh7289 - Tuesday, June 6, 2006 - link
Why do you have to buy the processor yourself to know how it performs? Real hardware was used in these benchmarks. AMD's 4x4 does not have any real hardware available yet, so it is vaporware as of now.bob661 - Tuesday, June 6, 2006 - link
I don't have to buy it but it does need to be available for purchase.Conroe isn't available either. So the vaporware argument could be used here as well.
zsdersw - Wednesday, June 7, 2006 - link
AMD's 4x4 is much more vaporware at this point than Conroe.peternelson - Wednesday, June 7, 2006 - link
However 4x4 is NOT a year away it will be here in 2006 from at least three motherboard vendors.
As for pricing criticisms, the AMD processor prices will be lowered WHEN conroe actually ships ;-)
I'm just pointing out that with Intel conroe, you have nowhere to go to scale up without switching to Xeon platforms and they're not particularly cheap.
What a 4x4 AM2 system will let you do is obtain a reasonably fast system initially using ONE fx2, and LATER spend a bit more and you have a MONSTER.
This phased upgrade may suit people who get paid monthly and want to upgrade gradually. AMD 4x4 will provide that upgrade path.
fitten - Wednesday, June 7, 2006 - link
You do realize that you could have been doing this for ages... just buy any of the dual socket S940 motherboards, put one dual core Opteron in it today, then add another at some later time? This 4x4 flailing is just marketing. The only real gain is that it doesn't require registered/ecc memory.The FX line is still going to be expensive AND it's not like there's a ton of software (especially games) that use dual cores... much less quad cores (although this may change with time... but by that time, there'll be even better CPUs out, probably even *real* quad core chips).
Even being an AMD supporter, the 4x4 is not much more than a panic reaction by AMD in response to Core (and its variants). I, for one, will NOT be throwing money into a 4x4 system. At best, it seems like a one-off and a dead-end path before it is even released.