- Our test method is described here.
- All data types are reduced to one relative integer score: performance of a reviewed processor compared to performance of Intel Core 2 Quad Q6600, which is considered 100 points.
- Detailed results are available in this cumulative Excel spreadsheet. The article contains only summaries of each benchmark class.
- If detailed results are worthy of your attention, we mention them.
||Core i7 920
||Core i7 Extreme 965
|Process technology, nm
|Core clock, GHz
|Number of cores
|L1 cache, I/D, KB
|L2 cache, KB
||4 x 256
|L3 cache, KB
||3 x DDR3-1066
(*) When Turbo Boost is enabled (by default), the real clock rate of separate cores grows by 133-266MHz depending on the current load.
(**) The maximum frequency supported by CPU memory controller; frequency and memory type of LGA775 processors are detected by chipset.
(***) Unlocked for overclocking needs.
|Core i7 920
||ECS X58B-A (X58)
|Core i7 Extreme 965
||Intel DX58SO (X58)
||Intel DX58SO (X58)
There is one interesting nuance in our tests. In terms of performance, the X5560 must be closer to the 920 than to the 965: it differs from the former only by 133MHz, while the latter differs by 400MHz. So we can assume that its test results will fall within these limits, 5% off one limit and 15% off the other. But there are some reservations. For one, the 5500 series uses the new stepping (otherwise, engineers couldn't have abated appetites to the level of the i7 and 3500 at the same frequency), which may affect performance. For two, we have tested the 920 on a different motherboard. It's even not about models, as even Intel has been releasing two BIOS versions a month for DX58SO. That is, the new platform was not hatching so easy, if it took so much effort to debug. We tested Core i7 920 two times, and our performance results differed significantly. Three more months have passed since that time, two months since our tests of the i7 EE 965. Let's see how both factors (the new stepping and new BIOS versions) affect CPU performance.
3D modelling and rendering
No miracles here, of course -- in this and following tests we'll see the same layout of forces, which could have been predicted beforehand judging by the clock rates. But what concerns absolute values, deviations from our assumptions are quite significant: if not for the optimizations of the last months, this result should have been demonstrated by at least X5570, or even the hypothetical Xeon 3.06GHz.
Just like in the previous case, only more pronounced.
Practical results match the theory here -- just a 5% advantage over Core i7 920.
Professional photo processing
It's something in between: the results of the X5560 are better than planned, but worse than in the first two groups of tests.
Total professional score
The clock rate pulls it to the 920, the other factors -- to the EE 965. So the result is right in between.
As with the compiling test, results are fully compliant with the theory. Such groups of tests are in minority today.
It's all clear: the effect of processors is getting less noticeable. The more powerful is a processor, the lower is its effect. So it's apparently time to change our test procedure. In fact, this may be the last review written according to the old procedure.
Non-professional photo processing
Total non-professional score
If we retested the 920 on the motherboard from Intel, there would be certainly nothing to comment. As it is, all test results are as we expected -- our 2.8GHz processor is faster than the same processor operating at 2.66GHz, and slower than the one at 3.2GHz. We've just confirmed it one more time.
Most individual users will ignore the rollout of Xeon 5500. Although some of them will certainly benefit from it. Just take a look at selling prices for some "E" models. For example, the E5506 is less than $300, so a couple of such processors will not be much more expensive than a single Core i7 940. Six DDR3 modules will have the same price tag in both cases, as they are the same. Prices for motherboards with two LGA1366 sockets are not available yet. However, many products presented at the last CEBIT are apparently designed for the mass market (for example, the standard ATX motherboard ASUS Z8NA-D6), so they must have affordable prices. Thus, it will be possible to assemble a relatively inexpensive dual-CPU system with eight "true" cores. So if you need high performance of tasks well-optimized for multiprocessing, you can build such a computer. "People's SMP", in a sense implied ten years ago, is out of the question, of course. But it's not necessary -- installing two processors was the only way at that time to get more than one computing core in a system. Now you can do it, if eight simultaneous threads are not enough, and you need 16 of them (a half will be actual, not virtual). There are much fewer users who really need it.
So, Nehalem processors have entered the market of multiprocessor systems. That's good for those who use (or is going to use) such systems, but it doesn't affect the mass market. Everything will be as it has used to be. And it won't change until the long-awaited rollout of Core i5.
We express gratitude to Corsair Memory for contribution to our testbeds.
Write a comment below. No registration needed!