Core i7 does have competition
NO, IT'S NOT AMD, it's Core 2 Quad. Lotsa website gigglebytes were recently spent comparing the Phenom II Black Edition 955 to the Core i7 965 or the as still yet unannounced 975. (Notice how similar these part numbers are getting, by the way?)
However, aside from extreme overclocking, Core i7 still wins almost all benchmarks. Same for its dual-headed twin, the Nehalem EP Xeon W5580. There is, however, some not so well advertised competition in some areas. Ever looked at the SiSoft Sandra 2009 AES cryptography benchmark? Here it is:
Notice how the Nehalem is throughly overrun - by more than 10 per cent - by its immediate 3.2 GHz Penryn based predecessor despite the newbie running at 3.33 GHz in Turbo mode, while AMD needs to use 4-socket monsters to stay on the charts at least till Istanbul arrives? The same would apply on the desktop side, by extension.
Why? Well, every "Yorkfield" Core 2 Quad or its "Harpertown" Xeon equivalent have two dual core dies inside, each with 6 MB fast (15 to 18 cycle) L2 caches, for 12 MB total. Yes. there is a humongous FSB hop penalty when going between the dies, but it is still smaller than going between two chips across two FSBs and a North Bridge in between when using a DP system.
However, if for some reason your critical code fits nicely in that 12 MB cache, and the threads don't kiss each other too much over that FSB - and yet that same code finds Nehalem's shared 8 MB L3 at 35+ cycles depending on the uncore clock too small and/or slow - then that code may run faster on the older chip.
As mentioned before, I really hope that Westmere will have, besides a 12 MB L3 cache to feed its expected six cores, also an improved latency profile including eliminating the 3-cycle core-to-uncore penalty that exists right now even if both are clocked the same.
A 975X chipset-style PAT mode that cuts that latency for direct in-sync clock situations would be welcome back. In the meantime, if you have similarly clocked and configured "Penryn" and "Nehalem" generation systems, and time at hand, just play with your apps, you might discover some interesting stuff! µ
Share this:
ShareBusiness Intelligence Software
Customer Relationship Management (CRM)
Enterprise Accounting Software
Enterprise Resource Planning (ERP)
Enterprise Systems Management
shouldn't that X7640 rather be called X7460, the old hexacore of the Core2 line?
Core2 has lots of life left in it. I wonder if intel is brave enough to bring core2 onto 32nm and 2xnm. Something tells me intel might have to compete with itself. Low clocked/volted core2 solo and duo would spruce up the netbook market so that they are actually useable for something more then sending an email.
MY Cpu is sooo much better than i7 HAHAHA.
I can't wait to do some cryptography!!!!!!
Core 2 cores are smaller than Nehalem cores too. I would expect a 32 nm Core 2 variant to be pretty small.
2x core 2 quads have 8 PHYSICAL cores.
Nehalem chips have 4 physical and 4 virtual - they will never match the performance of 8 physical cores, and the fact that they perform so well compared to their older siblings is a testament to how well the SMT is performing on core i7. Wait until the 8 core variants of Nehalem arrive, and simply test without SMT enabled, to make it a fair(er) comparison.
When you say the i7 gets beats by more than 10% over the previous generation... I think the blue data point is a dual socket setup with 6 cores per socket (which would not be that surprising against a dual socket **quad** core i7)
I don't think most readers would be surprised to see a dual socket hexacore setup beat a dual core quadcore setup even with the architecture differences.
Could you, or another kind reader, clarify if I'm reading this correctly? (Thanks)
Guys,the benchmark is apples-to-apples, i.e. two 4-core "Penryn" Harpertown Xeon X5482 vs two 4-core "Nehalem-EP" Gainestown Xeon W5580, the red point. So, 8 physical vs 8 physical cores, and you can see 3.2 GHz X5482 outsrips the W5580 pair (once HT is on in W5580 the situation gets even worse as you can see on another point).
Sandra is NOT a good benchmarking tool. I also find it very funny that Intel's only competition is it's self. I love that people hate Intel so much.
Nova - thanks for clearing it up.
Though I still don't see how it is an 'outstrip' by more than 10%, as the author suggests (if looking at the green and red points)... I see ~30/1100 on crypto (or <3 ).
Sandra benchmark, in my opinion, shows nothing. Only real world applications matter.
Sandra isn't even a real benchmark. It's like comparing superpi... pure maths when processors rarely do pure maths. Admittedly the article isn't clear regarding which setups are being compared exactly. However, Core 2 Quad has a lot of life left in it, that's one thing im sure we can all agree on. My next upgrade will be a Q9650 to replace me E8400, and then I'm not upgrading again until Core i7 8-cores are released with pricing that doesn't require one to blackmarket some organs.
Don't the VIA chips get around 800 MB/s? It's not as high as these CPU's, but for a single core it puts them to shame.