By: Mark Roulo (nothanks.delete@this.xxx.com), January 16, 2011 11:37 am
Room: Moderated Discussions
Sigh ...
Lets assume that the claim is correct, and lets also assume that the claim is for integer throughput type loads. If it isn't, then the core count and Hyperthreading are irrelevant.
Lets also assume that the claim is *GENERAL* ... that we'll get similar results for a broad range of loads.
Lots of assuming going on here.
Finally, lets figure that HT gives Intel cores a 20% boost at throughput loads, and that the Bulldozer design allows the 2nd 'core' in a module to effectively run at 80% of the performance of the first core for throughput loads (I think AMD has made claims that this is what is to be expected).
So a given AMD Bulldozer module has 1.5x the performance of an Intel Nehalem core with HT turned on.
1.2 x 1.5 = 1.8
We can conclude that the single threaded AMD Bulldozer performance is roughly equivalent to an Intel Nehalem when comparing cores versus modules.
But SandyBridge is faster than Nehalem for single threaded loads.
Bulldozer isn't going to blow SandyBridge away for any single-threaded loads ... which is a lot of desktop/laptop loads.
We also might look at transistor counts. Intel says that the Core i7-950 comes in at 731M transistors.
Wiki says that a Bulldozer module comes in at 213M transistors. 4 of these would be 852M transistors (about 16% more than 731M).
More transistors for the AMD chip, better throughput performance, worse single threaded performance. This seems like a competitive chip, but not one that has a crushing advantage over the *current* Intel chips.
Additionally, Intel has had a recent (and maybe longer than that) advantage in transistor density. For the same die size at the same process node, Intel can get more transistors on a chip. A lot of this is cache/SRAM design. But, the key upshot is that it suggests that Intel could, if necessary, compete with the 4-module/8-core Bulldozer with a 6-core/HT-enabled SandyBridge. Not Intel's first choice, but clearly something they could do.
We really won't know until Bulldozer ships and we get a full set of benchmarks, but until then this one sketchy benchmark suggests to me that AMD might be competitive again, not that the game is over.
-Mark Roulo
http://ark.intel.com/Compare.aspx?ids=46473,48496,37150,47932,
Lets assume that the claim is correct, and lets also assume that the claim is for integer throughput type loads. If it isn't, then the core count and Hyperthreading are irrelevant.
Lets also assume that the claim is *GENERAL* ... that we'll get similar results for a broad range of loads.
Lots of assuming going on here.
Finally, lets figure that HT gives Intel cores a 20% boost at throughput loads, and that the Bulldozer design allows the 2nd 'core' in a module to effectively run at 80% of the performance of the first core for throughput loads (I think AMD has made claims that this is what is to be expected).
So a given AMD Bulldozer module has 1.5x the performance of an Intel Nehalem core with HT turned on.
1.2 x 1.5 = 1.8
We can conclude that the single threaded AMD Bulldozer performance is roughly equivalent to an Intel Nehalem when comparing cores versus modules.
But SandyBridge is faster than Nehalem for single threaded loads.
Bulldozer isn't going to blow SandyBridge away for any single-threaded loads ... which is a lot of desktop/laptop loads.
We also might look at transistor counts. Intel says that the Core i7-950 comes in at 731M transistors.
Wiki says that a Bulldozer module comes in at 213M transistors. 4 of these would be 852M transistors (about 16% more than 731M).
More transistors for the AMD chip, better throughput performance, worse single threaded performance. This seems like a competitive chip, but not one that has a crushing advantage over the *current* Intel chips.
Additionally, Intel has had a recent (and maybe longer than that) advantage in transistor density. For the same die size at the same process node, Intel can get more transistors on a chip. A lot of this is cache/SRAM design. But, the key upshot is that it suggests that Intel could, if necessary, compete with the 4-module/8-core Bulldozer with a 6-core/HT-enabled SandyBridge. Not Intel's first choice, but clearly something they could do.
We really won't know until Bulldozer ships and we get a full set of benchmarks, but until then this one sketchy benchmark suggests to me that AMD might be competitive again, not that the game is over.
-Mark Roulo
http://ark.intel.com/Compare.aspx?ids=46473,48496,37150,47932,
