Tests
The basic test method is described in this article. However, we modified it for today's review. Firstly, we removed the gaming tests as inappropriate. Besides, poor performance of games in multi-way configurations are widely known. In due time, Intel probably had a right to promote Skulltrail as a gaming platform, but we are not going to trick you like that. Two or three graphics cards are fine for games. Two processors are not.
Secondly, we had to remove two more applications, which wasn't nice. The XviD codec we use, crashed right on detecting 16 threads available. In its turn, SPECjvm2008 did work (at least the runtime looked normal), but couldn't output any results. Sadly, both tests (especially the second) appreciated increasing the amount of cores and threads, and we hoped we could see how they performed on a dual-way machine.
For your convenience, the results are represented in percent, 100% being the result of Intel Core i7-860 in each test. Normally, we use another processor as a reference, but the results of this review are not compatible with the rest anyway. All absolute values are provided in this Microsoft Excel spreadsheet.
3D visualization
The fact that these tests just require a fast dual-core processor, not being able to use more than that, was confirmed yet again. The "burden" of processor interaction and the difference in motherboards must also be considered. Thus, a performance drop.
3D rendering
One of our readers has once opined that when you have two PCs and need to distribute tasks for a 3D modelling suite, it's wiser to use the faster PC for interactive workflow to reduce bothersome delays, and offload final rendering to the slower machine so it's done when it's done. But the diagrams prove this theory wrong. Interactive workflow doens't require a very fast PC, while rendering is a task which can easily consume all resources you provide. Escpecially, when we're talking about a company with multiple developers. They can save on personal workstations (reasonably, of course) and buy a better rendering server. Sounds logical, and reasonable financially. But if you need a single PC for one user, then it's wiser to get a regular desktop processor -- an extreme modification, if you can afford it -- but do not mess with two processors at all. True, a couple of X5570's outperforms a single 975EE by about 25%, but look at the prices. Besides, all that performance will only be needed for rendering, being idle during interactive work. Or you can at least wait a few months for the rollout of Gulftown.
Anyway, we expected more from these tests. The detailed results clearly indicate the bottleneck -- Maya. However, blaming it all on the developers wouldn't be correct. The suite is nicely optimized for 8 threads (a couple of older Xeons or Opterons), it can even handle 12 (two Xeon 7400 or two new hexacore Opteron CPUs). But the rollout of dual-way systems based on Xeon 5500 turned out to be a surprise, and currently the suite cannot provide enough work for 16 threads. This may be fixed in the future, but for now Maya remains another perfect "user" of the future hexacore LGA1366 processors. Or you should again consider a shared rendering server capable of processing multiple tasks, keeping its cores busy.
Scientific calculations
Similarly to the first group of tests, only performance drops here. Still, we can't say this is surprising. We have already confirmed that a couple of cores is quite enough.
Bitmap processing
There is some performance increase here, and noticeable at that. But if you take a look at the detailed results, you'll see the reason is nearly the doubled performance of Paint.NET. But this isn't the application you'll buy a powerful PC for. What's more important, the results of Adobe Photoshop improved as well. Sadly, the improvement doesn't justify the investment. As for the rest, the results only got worse.
The performance boost of Paint.NET has something to do with how .NET Framework works. As we know, this environment is logically similar to the Java Virtual Machine. Not the classic bytecode interpretation, but the modern JIT compilation at that. And it's designed for servers as well. So the great scalability of Paint.NET actually means the great scalability of .NET. Now this can already be useful.
Data compression
No performance drop, some increase even. Is it surprising or not? When we examined the peculiarities of memory controllers in LGA1366 CPUs, the 3 x 1GB configuration performed better than the one with 3 x 2GB. This is your answer why we don't see a performance drop here.
Write a comment below. No registration needed!
