by Ilya Gavrichenkov
10/12/2005 | 09:43 AM
A possible release of Sempron processors in the Socket 939 form-factor was rumored since the very moment when AMD decided to transfer its value processor series to the progressive K8 architecture. On the roadmaps AMD showed to its partners in the summer of 2004 there were Socket 939 Semprons beside the Socket 754 models. However, Semprons for Socket 939 didn’t appear on the market at that time despite all the preparations. Why? AMD’s partners were just afraid the Socket 754 platform they had put so much money and effort into would die prematurely. So AMD had to release the Sempron only in the Socket 754 form-factor in order to support the whole Socket 754 infrastructure.
The lifecycle of the Socket 754 platform became longer as a consequence, while AMD separated its platforms even more by stopping to release new Athlon 64 CPUs for Socket 754. The Socket 754 platform has eventually come to be viewed as a foundation of an inexpensive computer, with limited upgrade opportunities.
The potential buyers were not, however, worried much about the deficiencies of Socket 754 – Sempron CPUs and appropriate mainboards have been selling very well due to a number of factors. First, Sempron processors based on the K8 architecture are highly competitive against Intel’s low-end Celeron D series. Second, Sempron CPUs have a high overclocking potential, so they can be made to perform like mainstream Athlon 64 models. And third, CPUs of this series have steadily acquired all the basic features of Athlon 64 processors, including the 64-bit AMD64 extensions in the first place. As a result, Socket 754 Semprons currently account for about 50% of AMD’s total shipments which is an outstanding achievement.
Semprons based on the K8 architecture didn’t let the Socket 754 platform die too soon, yet its lifecycle is drawing to the end. This is partially due to the transition of the Athlon 64 series to the new socket (Socket M2), scheduled for 2006, after which AMD’s processors will be able to work with DDR2 SDRAM. Following this transition, the Socket 939 platform will move down from the high-performance class into the entry-level sector, filling in the niche that is currently occupied by Socket 754. Another argument against Socket 754 is that some PC suppliers aren’t happy with a separate, low-end systems-oriented platform as they have to diversify their product range especially for it. These factors make AMD revise its vision of the relations between the two sockets and push the company to mass production of affordable Socket 939 processors.
As a matter of fact, AMD has long been making Semprons for the Socket 939 platform, but only for computer manufacturers rather than for retail sales. For example, Fujitsu-Siemens and HP have been offering Socket 939 Sempron-based PCs almost since the beginning of this year. About 10% of all Semprons based on the K8 architecture were made in Socket 939 packaging in the first half of 2005 to meet the demand of leading PC suppliers and this percentage has grown up more by today, of course.
And since there are so many Socket 939 Sempron processors produced, we could expect them to emerge in the retail market, too, sooner or later. And last week we really spotted such processors in a few retail shops here, in Moscow. And since the Socket 939 Sempron may be like a completely new processor for many users, we decided to buy one and test it our labs.
The sample we bought had a rating of 3000+. It looks exactly like an Athlon 64, if you don’t read the marking. But the marking does tell us that it is a Sempron for Socket 939 rather than an Athlon 64.
AMD’s website is as yet silent about the existence of any Sempron for Socket 939 processors, so we have to decipher the marking string of our sample of the CPU (SDA3000DIO2BP) by ourselves:
AMD Sempron 3000+
939-pin organic micro-PGA
Supported memory types
HyperTransport bus frequency
Typical heat dissipation
Max case temperature
AMD64 technology support
NX- bit support
The characteristics listed in the table above tell you quite clearly that the Socket 939 Sempron is nothing else but a Venice-core Athlon 64 with a cut-down L2 cache. This comes from the fact, for example, that having an E3 stepping core, our Sempron supports AMD64 technology which has just recently appeared in Socket 754 processors (Socket 939 Semprons, on the contrary, always supported this technology). The reviewed processor is also close to the Athlon 64 in having a dual-channel memory controller and a 1GHz HyperTransport bus.
The CPU-Z version 1.30 utility doesn’t give us full info about the processor, obviously because there are no official data about Socket 939 Semprons from their manufacturer:
However, CPU-Z correctly detects a 128KB L2 cache as well as SSE3 and AMD64 instructions. The Socket 939 Sempron 3000+ also supports AMD’s Cool’n’Quiet technology, dropping its frequency to 1GHz and voltage to 1.1V in the power-saving mode.
The following table compares the characteristics of two existing Sempron 3000+ (for Socket 754 and for Socket 939) and shows how they differ from the Athlon 64 3000+ model:
Athlon 64 3000+
L2 cache size
The two Sempron 3000+ CPUs differ a lot. The Socket 939 model has the same frequency and the same amount of cache memory as its Socket 754 analog, but features a faster HyperTransport bus and a dual-channel instead of a single-channel memory controller. These factors should obviously make the Socket 939 Sempron 3000+ the faster of the two. As for the price factor, both Sempron 3000+ versions have the same official price and do not differ much in their retail prices, and so the Socket 939 version seems a much more appealing buy. In fact, the Socket 939 Sempron 3000+ is inferior to the full-fledged Athlon 64 3000+ in one thing only – its L2 cache is reduced to 128 kilobytes.
The Socket 939 Sempron having been a very rare product in the retail market until recently, you may wonder about its compatibility with existing mainboards. Well, you shouldn’t worry about that. AMD has been supplying such processors since the beginning of this year, even though without much fuss, so all the leading mainboard makers have long introduced the necessary changes into the BIOS code of their products. We installed our sample of the processor into several popular mainboard models and they all worked normally and correctly identified the CPU as Sempron 3000+ for Socket 939.
As said above, we tested a single sample of the Sempron for Socket 939 processor, with a rating of 3000+, but the series is not limited to this single product. Today three Socket 939 Sempron models, with ratings of 3000+, 3200+ and 3400+, are being distributed among the OEMs. All these processors are going to emerge in retail stores sooner or later, so you may want to have a general idea about each of them, considering that AMD hasn’t yet published their specifications on its official website.
Click to enlarge
As you see, the Socket 939 Sempron 3000+ is the only model in the series whose frequency and cache memory amount coincide with the corresponding characteristics of its Socket 754 counterpart. For example, the 3400+ model in Socket 939 packaging has a smaller L2 cache, while the Sempron 3200+ doesn’t have a Socket 754 analog at all.
More Socket 939 Semprons are going to appear in the future – AMD is expected to simultaneously introduce new value CPUs for both sockets. It is important to note that all Socket 939 Semprons, without an exception, feature the full set of technologies implemented in Athlon 64 series CPUs, including 64-bit AMD64 extensions, SSE3 instructions, NX-bit and Cool’n’Quiet. Newer Sempron for Socket 754 models also have the same capabilities, but Socket 939 CPUs have always had them because AMD64 technology was never disabled in them and because older, revision D (and earlier) cores have almost never been used in Socket 939 Semprons.
The technological advantages of the Sempron 3000+ for Socket 939 over its Socket 754 analog put it into an intermediate position between the Socket 754 Sempron 3000+ and the Socket 939 Athlon 64 3000+. Considering that both Sempron 3000+ models are going to cost you about the same money, it will be quite interesting to check the performance level of the Socket 939 one.
So, we performed our tests to compare the two versions of Sempron 3000+ between each other and with the Athlon 64 3000+. To have a bigger picture, we also admitted a competing product to the race – an Intel Celeron D 351 processor with a default clock rate of 3.2GHz.
Here’s a full list of hardware parts we made use of during this test session:
We performed the tests in Windows XP SP2; the mainboards’ BIOSes were tuned for the maximum performance.
The Socket 939 Sempron 3000+ is between the classic Sempron 3000+ and the Athlon 64 3000+ in typical office and multimedia applications.
The popular synthetic benchmarks also react positively to the architectural advantages of the Socket 939 Sempron over its Socket 754 analog (particularly, a higher memory bandwidth).
The memory subsystem performance influences the speed of the reviewed processor the most in ScienceMark 2.0. The new Sempron 3000+ is very close to the Athlon 64 3000+ which enjoys a single architectural advantage over it – a larger L2 cache.
WinRAR yields an opposite picture: the single-channel memory controller of the Socket 754 Sempron has a lower latency compared with the dual-channel controller of its Socket 939 counterpart. That’s why the transition of the Sempron from Socket 754 to Socket 939 brings about a certain performance hit in this particular test.
The memory bandwidth is most important for video encoding tasks and the Sempron 3000+ with a dual-channel memory controller is much faster in applications of that type than its Socket 754 analog.
The Socket 939 Sempron 3000+ behaves quite predictably in Adobe Photoshop CS2, taking an intermediate position between the classic Sempron 3000+ and the Athlon 64 3000+.
The memory subsystem performance and the amount of L2 cache memory have just a small influence on the system’s overall speed in final rendering tasks. But the size of the cache becomes a crucial performance-defining factor when you work in the viewports.
It is also the amount of cache memory rather than the memory bandwidth that determines the speed of the system in popular games. The more progressive Sempron 3000+ enjoys an advantage of 5-7% over its Socket 754 analog.
Summarizing the results of the tests we should acknowledge that the transition of the Sempron processor to Socket 939 is going to make it a little faster. However, this architectural change doesn’t affect the overall situation in this market sector much. Intel’s entry-level Celerons D are still superior in some media content processing tasks, while the Semprons are faster in the rest of applications. CPUs of the K8 architecture are especially better than their competitors in modern games (the same situation as in the high-performance CPU sector).
Besides measuring the speed of the Sempron 3000+ for Socket 939 in our benchmarks, we also checked its power consumption level (we hadn’t measured the power consumption of value processors from AMD or Intel before, by the way).
Our method remains unchanged. We measure the consumed current (that goes along the 12V CPU power circuit) with a clip-on ammeter. The real voltage in the circuit is verified with an ordinary multimeter. Based on the fact that the CPU is fed by the PSU through a separate circuit, this method is not very accurate since the CPU power converter is not 100% efficient, yet its accuracy is satisfactory for our purpose, especially since the power consumption of processors may vary a little from sample to sample.
There were two test modes:
The results are presented below:
The diagrams show that the power consumption of different Semprons is roughly the same or a little lower than that of Athlon 64 processors that have the same frequency and voltage but four times the Semprons’ L2 cache memory amount. Intel’s value processor, Celeron D, looks a real voracious beast against AMD’s CPUs, consuming much more power than Sempron or even Athlon 64.
So, if you want to have an inexpensive and also economical system, you may want to consider value CPUs from AMD first since they will permit you to save some money on your electricity bills.
We have repeatedly compared the Socket 939 Sempron with the Athlon 64 in this review. Our Socket 939 Sempron 3000+, based on a Palermo core, has much in common with the Athlon 64 3000+ on the E3 stepping Venice core. And we would really want the Socket 939 Sempron 3000+ to have the same consumer qualities as Venice-core Athlon 64 CPUs, particularly in terms of overclockability.
Athlon 64 CPUs with an analogous core of the E3 stepping have a good overclocking potential in general, so overclockers may be interested in appropriate tests of the reviewed Sempron. So, how does it do at overclocking? To answer this question we assembled this testbed:
We mounted a powerful CPU cooler to get better results, but this measure proved to be superfluous. Having low power consumption and heat dissipation, the reviewed Sempron 3000+ for Socket 939 was not too hot even when we increased its voltage and used a standard aluminum cooler. So, you don’t have to spend your money for a top-end cooler if you’ve got a Sempron on an E revision core. This will bring just a minor overclocking gain, if any. The Sempron 3000+ has a default frequency of 1.8GHz; its frequency multiplier is 9x. Since this CPU supports Cool’n’Quiet technology, its multiplier can be decreased, but cannot be increased. So, we will have to raise the clock-gen frequency very high, if our CPU has a good overclocking potential. And it means we have to use a high-quality mainboard.
Frankly speaking, our overclocking experience with Palermo-core Socket 754 Semprons has been rather ambiguous. There are samples that can work at frequencies up to 2.8GHz without any special cooling, but there are also samples that find it difficult to start up even at 2.4GHz. AMD’s policy with respect to Semprons obviously implies using in them cores that do not suit for more expensive CPUs, for example for Athlon 64. But it seems that sometimes these “defective” cores are so few that AMD has to use “normal” cores (i.e. with a high frequency potential) for the Sempron, too. As a consequence, overclocking a Sempron is like a lottery the outcome of which you can hardly predict. That’s why we shouldn’t extrapolate the overclocking performance of one Sempron sample to another. The result may differ dramatically in each particular case.
As for our sample of the Socket 939 Sempron 3000+, we first checked it at its default voltage. Without losing stability we increased the frequency of the clock generator from 200 to 255MHz. The resulting CPU frequency was 2295MHz. This was not a satisfactory result, so we continued our experiments at a higher CPU voltage.
Having set it at 1.55V, we found the system to behave much better. For example, the OS could boot up at clock-gen frequencies up to 283MHz, i.e. up to 2547MHz CPU clock rate. The system was far from stable at that, however, so we began to reduce the clock-gen frequency until the computer could pass some basic tests (S&M, SuperPi and 3DMark 2001SE). Unfortunately, we had to go down for quite long – the system only became absolutely stable at 270MHz clock-gen frequency. So, we managed to overclock the CPU to 2.43GHz at 1.55V voltage.
This is not impressive, either, since the CPU frequency grew up by 35% only. The resulting frequency was just a little higher than the default frequency of the Athlon 64 3800+ processor that is based on an analogous core of the same E3 stepping. We must have been not very lucky with the sample of the processor.
Yet you should not think it was just a rare accident. We took two more Socket 939 Sempron 3000+ samples from the same batch and found their frequency potential to be much alike.
Of course, this frequency growth cannot boost the performance of the CPU much. To estimate the performance gain, we compared our Socket 939 Sempron 3000+ overclocked to 2.43GHz with an Athlon 64 3800+ (2.4GHz default frequency) using the same computer as in the tests above.
As you can see, the L2 cache reduced to 128KB does limit the performance of the Sempron 3000+ considerably, even though this CPU is equipped with a dual-channel memory controller and a 1GHz HyperTransport bus. The Socket 939 Sempron 3000+ clocked at 2.43GHz was about 5-10% slower than the Athlon 64 3800+.
But well, this is a matter of perspective. Our overclocking results may look quite bright if viewed from the price factor standpoint. In fact, our overclocked Sempron 3000+ closely approached the level of performance of a CPU model thrice its cost!
The first value processor for the Socket 939 platform is, without a doubt, a success. And we call it a success not only because the Socket 939 Sempron has all the advantages of AMD’s senior CPUs like 64-bit AMD64 extensions, a dual-channel memory controller and a 1GHz HyperTransport, its single disadvantage being a smaller L2 cache. And not only because it is faster than its Socket 754 counterpart which costs the same money. The main success of this product is in bringing the Socket 939 platform closer to more users.
The budget Socket 939 Sempron 3000+ is about $55 cheaper than the cheapest Socket 939 Athlon 64, so a Socket 939 mainboard plus CPU kit will cost you a mere $150-200. It means that economical users will have an opportunity to invest into a modern and future-proof rather than into an obsolescent platform.
We could find some faults with the Socket 939 Sempron, if were asked to. The processor we tested was not very good at overclocking, for example. It is possible we just had a not very overclockable sample, yet we don’t want you to foster any hopes to increase the performance of a Socket 939 Sempron through overclocking.
A second and much more distressing fact about the reviewed processor is its limited market availability. As far as we know, AMD doesn’t plan to ship value Socket 939 processors in mass quantities until the end of this year. So, if you want to have such a CPU for yourself, you should keep a watchful eye on your nearby shops and get one before others do!