Unfortunately, we encountered some problems in the process of testing single-core Athlon 64 processors for the old AMD Socket 939 platform. On the other hand, we already have test results of single-core Intel Pentium 4 processors. So we decided to sugar the waiting pill with an article devoted solely to single-core processors from Intel. Single-core processors from AMD for Socket 939 will be analyzed in Part 4 (we hadn't planned to write this part at all).
Hardware and Software
* - "2 x ..." means per core.
Our selection of Pentium 4 processors is very simple: we've covered the bottom and top* borders of the family. This approach to analyzing performance of the old platform seems to be the most logical choice: we can estimate the entire performance range at once. What concerns technical details and properties, they have been discussed so many times that it makes no sense to return to this issue. They are old processors. People buy them to save money, not to make experiments. What concerns the Pentium 4 531, it plays a very small role: to demonstrate scalability of performance as the clock grows.
* - a little note about the top border: yes, we know about Pentium 4 processors with processor number 660, 661 and 670. But we failed to find them: no one imports processors above Pentium 4 651 on a mass scale, you have to place an order if you need such a processor. We cannot imagine a user, who will order a top Pentium 4 processor now (especially after reading this article :). So we've agreed to regard the 651 processor as the top model as decided by the market.
A necessary foreword to the charts
We present the results of our tests in two unique ways due to our testing method. First of all, in the summary diagrams of this article all data types are reduced to one - relative integer points. What we mean by relative integer points is that the performance of all processors is relative to that of one processor, which in this case is the Intel Core 2 Duo E4300. The performance of the Core 2 Duo E4300 is rated at 100 points. Second, detailed results regarding the tests are available in a Microsoft Excel table. In the article itself only summary diagrams are given which are grouped together by benchmark classes. Nevertheless, when required, we will occasionally draw your attention to the expanded results which are located in the Excel file.
3D Modeling and Rendering
On one hand, results of the Pentium 4 651 demonstrate that some dual-core processors can still be outperformed by a single-core Pentium 4 model. On the other hand, in order to outperform the slowest dual-core processor on the market, it needs the advantage of 740 MHz in core clock. It's a Pyrrhic victory. If we also compare the prices...
Strange as it may seem (we've already come across this phenomenon in the previous part of the article), the NetBurst core fares rather well in this sub-test. Now we can add: not only in its dual-core reincarnation.
Digital Photo Processing
However, as soon as we proceed to applications that support multiprocessing, nothing can help the Pentium 4. Three Pentium 4 processors, three lowest places. Adobe Photoshop votes univocally for dual cores.
To all appearances, Microsoft compiler does not have an ideal SMP optimization. We can see it well, if we compare results demonstrated by Pentium 4 651 and Pentium D 940. They have the same architecture and L2 cache size per core. But the clock rate of the former is higher by 200 MHz, and the latter has two cores. So what? Nothing - they have finished the tests practically "head to head". Compare results of these very processors in the previous test...
Apache Benchmark with our settings generates up to 50 simultaneous threads, so dual-core processors get a nice advantage here. Cache size and memory bandwidth are critical for this subtest, but even a faster bus (800 MHz QPB versus 533) didn't help the Pentium 4 651 to outperform the weak Pentium D 805.
Owing to its much higher clock rate, the Pentium 4 651 manages to outperform the Pentium D 805 even in the test optimized well for multiprocessing. If you look at the detailed results, you'll see why: the Solver in CPU RightMark does not run on two cores, SMP is supported only by the rendering module. The Pentium D 805 is even a tad faster in rendering, but it loses in the Solver with the 32:41 score, that is the difference is big.
As we have already noted, our archivers evidently use some SMP optimization, but it's far from perfect. In this case the Pentium D 805 is the slowest model. Results of the Pentium D 915 look much better - owing to the doubled cache size and a faster bus.
It happens! We already wrote that ABBYY FineReader does not support multiprocessing in the batch mode. Besides, we can see well that the program is optimized well for the old architecture from Intel (or vice versa - it's badly optimized for the new one :).
It's an old test with mostly old codecs. Only one of the four supports SMP - results are appropriate: The Pentium 4 processors recalled their good old days, when they were considered fast processors.
The Pentium 4 is a fiasco. As we already wrote above, it was clear even without tests, if we already know that a given class of software can take advantage of multi-core CPUs.
Having the same cache size per core, the same FSB bandwidth, and being faster by 600 MHz, the Pentium 4 651 is still outperformed by the Pentium D 915. It's a brilliant proof that support for multi-core processors is coming into games. But not as fast as we would like.
We'll focus on a single comparison only, because it's the most illustrative one: Pentium 4 651 vs. Pentium D 915. Competing with each other from test to test, they have finished with identical total scores. And now we'll just compare the key properties:
These two processors with the same core architecture demonstrate similar performance (according to our test procedure). We shall not discuss whether the dual-core glass is half full or half empty. It's up to you to decide.
Estimated power consumption
Note that the Pentium 4 521/531 models do not support EIST. We can see the consequences on the idle power consumption diagram (compare it with the Pentium 4 651). On the whole, even the Pentium 4 processor looks good versus dual-core processors based on the NetBurst core. However, it's true only in this case. :)
Do you need it? One of the top single-core Pentium 4 processors (3.4 GHz, 2 MB L2 Cache) is outscored by 16% by the dual-core Pentium E2160, which has twice as small cache, almost twice as low clock (1.89 times, to be more exact), and which maximum power consumption is also almost twice as low. In our opinion, this fact is informative enough a conclusion on the prospects of single-core NetBurst processors.
Memory modules kindly provided by
Corsair Memory Russia
Stanislav Garmatiuk (firstname.lastname@example.org)
August 27, 2007
Write a comment below. No registration needed!