The first generation of desktop AMD64 processors consisted of a high end group, equipped with a dual-channel memory controller and Socket 939, and a waning group of budget processors with a single-channel controller, designed for Socket 754. Having upgraded to Socket AM2, all desktop processors from AMD got an all-purpose socket. It has brought at least two advantages to consumers — a dual channel DDR2 memory controller (up to DDR2-800!) even in budget Sempron processors and an option to buy an inexpensive processor now and upgrade it to a more powerful processor in future.
Let's clear up the issue about DDR2-800 support. The new Semprons officially support 667 MHz maximum. But as most motherboards choose maximum frequency for new Semprons even in automatic mode (provided the system is equipped with high-speed memory), we can say that 800 MHz is the default frequency. Of course, the advantage from increased memory bandwidth is minimal for Semprons - even representatives of the main series do not utilize it to full extent (in some cases, when we managed to reduce frequency together with timings, lower clocks are even better in some tests).
OPN of the standard Sempron models (TDP=65 W) starts with SDA (for example, SDA3000IAA3CN). The name of models with reduced TDP (down to 35 W) start with SDD. You can learn more about OPN in our previous review of budget processors.
Another object of our review (along with Semprons for Socket AM2) is Athlon 64 X2 3600+, the most junior dual-core processor from AMD. It is a sort of a dual-core Sempron, because its cache is not typical of usual dual-core processors - 256 KB per each core. It's only natural to use the next model (3800+) in the series for comparison (and for determining the effect of paired down cache on real performance). This processor has the same clock rate, but twice as large cache. We also added a couple of Celeron D processors, which prices are similar to the top and budget models under review.
Hardware and Software
* — "... x 2" means "... per each core"
Essential foreword to charts
Our test procedure features two peculiarities of data representation: (1) all data types are reduced to one - integer relative score (performance of a given processor relative to that of Pentium D 805, given its performance is 100 points), and (2) detailed results are published in a Microsoft Excel table, while the article contains only summary charts by benchmark classes.
3D Modelling & Rendering
This group of tests makes a good use of the second core. Differences in frequency and cache size in the representatives of the budget series are not critical here. While there are still some differences between the top and the budget Semprons, the effect of 30% difference in frequency and doubled cache in the top Celeron is surprisingly small.
CAD & CAE
The situation is contrary here — Celerons differ greatly, so we can compliment the top model and scold the budget model. And our Semprons are on a par, they outperform competing processors and look good versus "inexpensive" dual-core processors.
Interestingly, clock rates are critical for budget Semprons, while the increased cache size starts making a difference only from the 3600+ model. According to this test, it's no worse than Athlon 64 X2 3600+. Unless you are going to compile code and run some applications in the background, of course :) Celeron does not fare well in compiling. There is nothing surprising about it, it's typical of Netburst processors.
Like any good synthetic test, RightMark ranked processors evenly by their real computational power. Dual-core processors break away radically, but the difference in cache sizes has no effect at all.
Semprons are slightly outperformed by their competitors in this test. However, we cannot ignore the fact that this test measures execution speed of typical Photoshop operations. But as they are executed fast, especially for small files (fractions of a second, several seconds maximum) by any modern processor, they are not necessarily a bottleneck in your case that determines subjective comfort of working with the program. If you are more concerned with the execution speed of the slowest filters, it makes sense to take into account the results of other computational tests.
Semprons demonstrate a small spread in this test, because it equally welcomes increased caches and frequencies. As an increased frequency in this case is usually accompanied by a decreased cache, only a model, which rating is greater by 200, is guaranteed to be more powerful.
AMD processors benefit from large cache, while Celerons again demonstrate extremely low scalability.
Here is a contrary situation. Frequency is of decisive importance. The top Celeron demonstrates results on the level of middle end products from the competing series.
Dual-core processors shoot ahead of the others and are approximately on a par among each other. Higher-clocked Semprons do not show evident advantage. On the contrary, budget models benefit more from a larger cache.
Top budget models compete well with dual-core processors here. Both Celerons look just like their competitors (their results are the best here).
3D Shooter Games
Semprons recoup themselves in games completely. Models with 256KB cache seem most preferable.
Efficiency per GHz
Semprons are nearly aligned on the summary diagram. That is characteristics of models in the series justify their ratings according to iXBT test procedure. In other words, each model is generally as faster than the previous model as it's slower than the next.
The defeat of the competing series is evident. Even though Celerons justified their position in some tests, the top Celeron hasn't risen above Sempron 3200+. You have higher chances to assemble a low-noise computer with Sempron, because the heat release of the top Celeron is noticeably higher than even in budget Athlon 64 X2 (it was noticeable even without measuring it with accurate tools - by the temperature of the coolant inside a water cooling system).
And finally, while all Semprons without exception can be installed on any motherboard with Socket AM2 (even if you didn't update BIOS, your processor will most likely work as an "Unknown Model"; only Cool’n’Quiet may theoretically work incorrectly and you may have to adjust voltage manually for operating stability). The situation with Celerons is much more complex. Out of five LGA775 motherboards on various chipsets that were available in our iXBT test lab, only one model supported the budget Celeron D 326, while we had to order a motherboard for the 356 processor. We chose an Asus mobo. According to the official information, it's compatible only with 2 out of 3 existing revisions of this processor! Fortunately, our Celeron D 356 belongs to one of these two revisions. It should be noted that the real situation with Celeron support is probably not as bad as it seems. Most motherboards we tried were based on expensive chipsets (such models tend to stay in testlabs, while budget models are usually hard to borrow for testing, to say nothing of keeping them for testbeds.) If we had chosen five typical inexpensive motherboards, 3—4 of them would have certainly worked well with any Celeron after a proper BIOS update.
The choice between Athlon 64 X2 3600+ and 3800+ is formally equivalent, the faster processor is correspondingly more expensive. But there is no abyss between them. That is 512 KB is not a borderline value, below which performance might degrade even in "cache-indifferent" applications. So we can give you an evident practical advice — analyze detailed results and find out how much your typical applications benefit from a larger cache (or suffer from a paired down cache).
Memory modules for our testbeds are kindly provided by
Corsair Memory Russia
Dmitry Laptev (firstname.lastname@example.org)
November 7, 2006
Write a comment below. No registration needed!