Dear forum members,
We are delighted to inform you that our forums are back online. Every single topic and post are now on their places and everything works just as before, only better. Welcome back!
Discussion on Article:
Sometimes They Come Back: AMD Launching Phenom II X4
this test is not correct,with every cpu you use faster memory with amd at the lowest speed. if you use the same speeed memory for every cpu the results will be diffrent!!!
Either X-bit labs staff do not beleive the tests they make or they don't bother to read their own tests. There are 7 test on the same page (page5: PC Mark Vantage) they make these comments and X4 940 soundly beats Core i7 920 in 2 tests (communications and productivity), 3 of the tests are such close calls and Core i7 920 clearly beats only in 2 tests. Read their comment now...and say crap!
Why do they test Phenom II with DDR2 800 and Core 2 Quads with DDR2 1066? There is just one explanation: they don't want fair competition. By doing this they hold back Phenom II by about 3-5%, which is such a critical trick, because Phenom II 940 is having such a close competition with Q9550. So they force it one step back.
X-bit labs you suck!
This struck me as odd as well. However, I think, in this case, it is correct to be doubtful of the PCMark tests since they are known to be extremely favourable (or even deliberately biased) toward Core 2; indeed, this outcome is not mirrored in any of the application tests later in the review. If anything, this comment should have been left until later, but I believe that it is factually correct provided that one doesn't put too much faith in PCMark as a representative application.
However, I think the power consumption tests in this review were done a little less ideally than I'm accustomed to from X-bit. For example, the Cool'n'Quiet vs. SpeedStep test made no comment on the nature of the workload, but looking at the scaling of the curve, I'd say it was probably something that depends significantly on interprocessor communication. Also, the comments toward the end of the article describe Phenom II as being more "energy efficient" than some other processors, but surely efficiency must be defined as proportional to the amount of work done per unit of energy consumed, and not the peak power consumption. Integrating power with respect to time while performing a given task yields a much more sensible estimate of true efficiency.
Finally, the description of the Mathematica 7 benchmark is incorrect. Mathematica 7 provides multi-processor support in exactly the same way as previous versions: via the Parallel Computing Toolkit, and by distributing work to slave kernels. The only difference is that this is now a standard package rather than a paid-for addon, so you don't need to use the script I sent previously in order to do multi-core benchmarks (although you still can if you want, and the results will be the same either way). Moreover, the actual tests performed by MathematicaMark7 are identical to those in version 6, so the scores obtained should still be around 8-10, rather than 3-5 as reported. Judging by the erroneous comment about multi-processor support using a single kernel, the problem was that the command LaunchKernels@Min[$ProcessorCount,$MaxLicenseProcesses] was not run before starting the benchmark, so it executed on only a single core. Also, for tests on Core i7 using HT, make sure that you have at least 8 kernel licences available--only four are provided by default with a standard installation of Mathematica 7. You can see how many you have on the System Information panel which is available via Help|About.
Remember when Intel screwed up with the P4, and how long it took for them to recover? It seems as if people think that it's easy to overhaul a core in a year or so and make things competitive again.
I'm actually surprised that AMD managed to work in some branch-prediction improvements into the Phenom II. Historically, Intel's king on the branch-prediction, and it's actually a good thing to see that AMD's trying to do what it can in that front.
The K10 architecture's flawed. I'm no expert, but from the docs and diagrams I've seen, they've 3 Integer execution units, 3 floating point execution units, but the dispatcher could only dispatch 3 uops. Compare that to the Core 2 architecture, they have 3 int/fp execution unit, and a load and store, and dispatch 4+1 uops IIRC.
I'm willing to bet that things aren't a drop in changes. There might be an off chance that there's more core improvements, but even I'm not holding out on that. I'm just hoping that AMD's going to hold out long enough for their next architecture.
I agree - they should have been tested on an equal footing. But from all the benchies I've seen memory bandwidth has virtually no (<1%) impact on system performance, especially on processors with an IMC. Even on the C2Q chips it's a marginal effect, they have enough L2 cache to buffer out the infrequent trips into slow system memory.
There are two things I would have liked to see in this article.
1) High resolution game benchmarks. Show the low res stuff if you like, but include 19x12 4AA also. I know CPU has far less impact at this resolution than GPU but I think it's important to see if you can get playable fps with these different processors with a good GPU.
2) Include a fast dual core processor. I know you guys have an e8600, use it. At some point duals are going to become obsolete & I want to know when they start showing signs of falling behind so I can upgrade.
Add your Comment
Enter your username and e-mail address. Password will be sent to you.