by Ilya Gavrichenkov
10/01/2004 | 02:15 PM
The value processor market has lately been a place of much interest for the user community as both suppliers of x86 CPUs, AMD and Intel, reformed their series of products for inexpensive computer systems throughout the last summer, making obsolete our comparative review of value CPUs we published on our site in the middle of May. The report on the Celeron D posted in June doesn’t help much, either.
<%BANNER[article]%>So, we must return to the problem of value processors once again.
The review you’re now reading is dedicated to comparing the performance of new value CPUs from AMD and Intel – the Sempron and the Celeron D, respectively. Intel has equipped its value series with a new and more advanced core achieving a substantial performance improvement. AMD’s new Sempron family has appeared exactly to compete with the rejuvenated Celerons in the bottom tiers of the market. Thus, the situation has changed completely: you had to choose between junior Athlon XP models and Celerons earlier, but now there are other names in the market.
To see the standings of the two manufacturers in this market sector, we carry out this review.
Before offering you the results of the tests, I’d like to describe briefly the basic characteristics of the participants.
Intel Celeron D. The Celeron D CPU owes its origin to Intel’s transition to the 90nm tech process. First appearing in the Pentium 4, the Prescott core soon found an application in the value series. Thanks to the new core, the Celeron D features 256KB L2 cache and SSE2 support. Intel also transferred the new Celeron to the 533MHz front-side bus. All these improvements contribute to a much higher performance of the new processor, making it more competitive from the price/performance point of view, compared to the older Celeron family.
The use of the advanced Prescott core manufactured with 90nm tech process also allows raising the frequency bar of the new Celeron D family. Right now, the senior model of the series is clocked at 2.93GHz; models with 3GHz and higher frequencies are soon to appear, too.
Note that with all the considerable improvements in the Celeron D, Intel doesn’t implement Hyper-Threading technology in it yet. Instead, the new Celeron D models (in the LGA755 variant) feature Execute Disable Bit, a technology for securing the OS against various malicious programs. This technology is supported in Microsoft Windows Server 2003 Service Pack 1, Microsoft Windows XP Service Pack 2, SuSE Linux 9.2 and Red Hat Enterprise Linux 3 Update 3.
AMD Sempron. AMD answered to the launch of the Celeron D family by releasing its own series of value CPUs called Sempron. Unlike Intel, AMD’s engineers didn’t introduce any changes into the characteristics of the processor cores they had at their disposal. The Sempron is in fact a renamed Athlon XP, which became very cheap after the arrival of the Athlon 64 CPU series. Semprons with ratings up to 2800+ are Socket A CPUs on old 0.13-micron Thoroughbred-B and Thorton cores with 333MHz FSB, up to 2GHz clock rates, and 256KB of L2 cache. The Sempron 3000+ is a Barton-core processor with 2GHz frequency, 333MHz FSB, and 512KB L2 cache.
The Sempron family for Socket A systems should oust the Athlon XP processor out of the market: this is an indication of the fact that the time of Socket A processors is over, and they can only serve in low-end computers now.
The Sempron family also includes a Socket 754 model (with 3100+ rating). The Sempron 3100+ differs a lot from its junior mates. It is based on the Paris core, sharing the same K8 architecture with top-end CPUs on ClawHammer and NewCastle cores. However, there’s little left of the power of the top-end models: the Paris core has only 256KB of L2 cache and has the AMD64 technology disabled. In other words, the Sempron 3100+ is a 32-bit processor, in spite of its links to the Athlon 64 family. On the other hand, the Sempron 3100+, just like the Athlon 64, supports Cool’n’Quiet technology, SSE2 instruction set, and can use the NX bit.
The Sempron is a new series of value CPUs from AMD, which is going to develop and extend further. It is clear that we’ll see other Sempron models for Socket 754 and Socket 939 systems soon.
The following table lists the basic characteristics of the models that comprise the AMD Sempron and Intel Celeron D series. Note that the manufacturers both use a rating system for naming their value processors. The correspondence between the rating and the real characteristics is also indicated in this table:
CPU | Clock frequency | Core | L2 cache | Bus frequency | Socket type | SIMD instructions support | Other technologies |
Intel Celeron D | |||||||
Celeron D 340 | 2.93GHz | Prescott | 256KB | 533MHz | LGA775, Socket 478 | SSE, SSE2, SSE3 | Execute Disable Bit* |
Celeron D 335 | 2.8GHz | Prescott | 256KB | 533MHz | LGA775, Socket 478 | SSE, SSE2, SSE3 | Execute Disable Bit* |
Celeron D 330 | 2.66GHz | Prescott | 256KB | 533MHz | LGA775, Socket 478 | SSE, SSE2, SSE3 | Execute Disable Bit* |
Celeron D 325 | 2.53GHz | Prescott | 256KB | 533MHz | LGA775, Socket 478 | SSE, SSE2, SSE3 | Execute Disable Bit* |
Celeron D 320 | 2.4GHz | Prescott | 256KB | 533MHz | LGA775, Socket 478 | SSE, SSE2, SSE3 | Execute Disable Bit* |
Celeron D 315 | 2.26GHz | Prescott | 256KB | 533MHz | LGA775, Socket 478 | SSE, SSE2, SSE3 | Execute Disable Bit* |
AMD Sempron | |||||||
Sempron 2200+ | 1.5GHz | Thoroughbred, Thorton | 256KB | 333MHz | Socket A | SSE, 3DNow! | - |
Sempron 2300+ | 1.58GHz | Thoroughbred | 256KB | 333MHz | Socket A | SSE, 3DNow! | - |
Sempron 2400+ | 1.67GHz | Thoroughbred | 256KB | 333MHz | Socket A | SSE, 3DNow! | - |
Sempron 2500+ | 1.75GHz | Thoroughbred | 256KB | 333MHz | Socket A | SSE, 3DNow! | - |
Sempron 2600+ | 1.83GHz | Thoroughbred | 256KB | 333MHz | Socket A | SSE, 3DNow! | - |
Sempron 2800+ | 2.0GHz | Thoroughbred, Thorton | 256KB | 333MHz | Socket A | SSE, 3DNow! | - |
Sempron 3000+ | 2.0GHz | Barton | 512KB | 333MHz | Socket A | SSE, 3DNow! | - |
Sempron 3100+ | 1.8GHz | Paris | 256KB | 800MHz | Socket 754 | SSE, SSE2, 3DNow! | Cool’n’Quiet, NX-bit |
This table provides all the necessary information about the new value CPU series from AMD and Intel. I’d only like to draw your attention to one thing. The ratings of Semprons are assigned according to different rules than the ratings of the Athlon XP or 64. For example, Athlon XP CPUs of the same ratings as Semprons have a higher performance due to a higher clock rate and a bigger amount of the cache memory. So why does AMD apply a new rating system to the Sempron? The answer is simple as soon as you take a look at the prices the manufacturers set for their products:
The official price lists are compiled in such a way that every Celeron D model has a Sempron counterpart with a rating equal to the frequency of that Celeron D. It is on this basis that AMD compares its Semprons with Celeron D models and claims superiority of its processors, though only in Winstone and SYSmark (AMD says nothing about how the Sempron compares to the Celeron D in games, for example).
The goal of this test session is to give you a most comprehensive view of the current situation in the market of CPUs priced below $120. That’s why I took all the models – from junior up to senior ones – from the two value series, Celeron D and Sempron. To give you a perspective of how the performance of the value products differs from that of top-end processors, I also included two 2.8GHz Pentium 4 CPUs on Northwood and Prescott cores as well as an Athlon XP and Athlon 64 (both with 2800+ rating). Besides that, I will show you the results of the Celeron 2.8GHz CPU, just for you to see the progress in the processor development field.
I tested the Celeron D processors on a Socket 478 platform, because LGA755 systems and appropriate hardware are rather expensive and limited in spread right now. Besides, as we proved it in one of our earlier articles, the use of DDR2 SDRAM and PCI Express graphics cards brings no advantages as yet. Note also that, although we deal with inexpensive processors, the test platforms were assembled of rather costly and high-performance parts (mainboards with a dual-channel memory controller, and a graphics card with the ATI RADEON 9800 XT chip) – we wanted to minimize the influence of other devices of the performance of the CPU proper.
So, the testbed systems were comprised of the following hardware:
I ran my tests in Windows XP with Service Pack 2 and DirectX 9.0c installed. The BIOS Setup of the mainboards was set up for the maximum performance.
The processors of the Celeron D family occupy the top lines of the diagram in this test. The senior models with 2.8 and 2.9GHz frequencies are so confident in PCMark04 that they even outperform the AMD Athlon 64 2800+, which definitely does not belong to the value products category. This is the performance bonus that the Celeron D acquired when transitioning to the Prescott core – 533MHz FSB, dual-channel DDR333 SDRAM and big L2 cache are the factors that contribute to this excellent result. While Prescott-core Pentium 4 CPUs are just slightly faster than their Northwood-core counterparts, the gap between the old and new Celeron families is much wider – the new 90-nm core shows its best sides when employed in low-end processors.
AMD’s Sempron is no match to Intel’s Celeron D in the speed of working with the memory. This comes as the bandwidth of the CPU-memory thoroughfare in Intel’s platforms is higher thanks to the 533MHz Quad Pumped Bus. The theoretical peak bandwidth of this channel is 4.2GB/s with Celeron D CPUs, and only 2.7GB/s with Socket A Semprons (3.2GB/s with the Sempron 3100+).
Note also by how much the Sempron 3100+ is faster than the other members of its family. Based on the K8 architecture, this model features an integrated memory controller that supports single-channel DDR400 SDRAM. That’s why a system with a Sempron 3100+ will have a higher memory subsystem bandwidth than Socket A systems, and a much lower latency, too. Thus, the Sempron 3100+ is capable of providing a much higher performance in real-life applications than other (Socket A) Semprons.
3DMark 2001 SE is very loyal to AMD’s Sempron processors, the result of the Sempron 3100+ for Socket 754 looking especially good. According to 3DMark 2001 SE, this processor is not only much faster than any Celeron D, but even outperforms the 2.8GHz Pentium 4 models. As for the Socket A Semprons, their performance is nearly similar to that of the same-price Celeron D models. However, the senior Sempron for Socket A is slightly faster than the Celeron D 340.
It’s different in the newer 3DMark03 suite: the Sempron family looks worse compared to the Celeron D series in this test. The Semprons for Socket A can only compete with the ordinary Celeron based on the old Northwood core, with 128MB L2 cache and 400MHz FSB. The only Sempron to perform more or less successfully here is the Sempron 3100+ intended for Socket 754 systems. This model leaves all the members of the Celeron D family behind, including the senior 340 model (2.93GHz).
The CPU test from the same benchmarking suite produces somewhat different results. The Semprons don’t look utterly hopeless against the Celeron D gang. The senior Semprons for Socket A are competing with the middle-range Celeron D (320-330), while the Sempron 3100+ for Socket 754 systems is again an unrivalled leader among the value products.
Since testing the Athlon XP in Quake 3 we know that this CPU architecture isn’t the best choice for that shooter. So, we can’t hope for anything exceptional from the Socket A Semprons here, which are clocked at lower frequencies than regular Athlon XPs. And really, the Semprons are all in the lower part of the diagram with a single exception of the Sempron 3100+, which features an integrated memory controller and the K8 architecture. This processor is so fast in Quake 3 that it leaves behind the full-featured Pentium 4 on the Prescott core and clocked at 2.8GHz!
Unreal Tournament 2004 draws quite another performance map: the Socket A Semprons are fast here, outperforming the competing Celeron D models. The Sempron 3100+ is again superior to the rest of the Semprons. This is a characteristic thing, by the way. Although there’s little price difference between the Sempron 3100+ and Sempron 3000+, the Socket 754 processor is more future-proof and is just faster in real applications.
The competing Celeron D and Sempron models of the same price provide nearly the same performance in the semi-synthetic Aquamark3. However, the Socket A Semprons lose the CPU test to their Socket 478 counterparts.
Once again the Sempron 3100+ receives my praises: although its L2 cache is twice smaller than the one of the Athlon 64 2800+, the two processors run Far Cry almost at the same speed. As a result, the Sempron 3100+ outperforms the whole Celeron D family as well as the two Pentium 4 2.8GHz on Prescott and Northwood cores. So, I have to repeat it again that the Sempron 3100+ is a CPU from quite another category than the rest of the value CPUs – both for Socket A and for Socket 478.
The results are typical in the popular Doom 3 game: the performance of the Celeron D and the Sempron differs but slightly, so it’s hard to claim that any of them is a leader here. However, note that the Sempron 3000+, the only value product with 512KB L2 cache, has a significant advantage over the others. It is an indication of the importance of cache memory in this game. Secondly, the Sempron 3100+ is surprisingly good, too, being just a little slower than the 2.8GHz Prescott.
Overall, it’s hard to say that any of the two value CPU families is better in gaming applications. Depending on specific optimizations or algorithms, either the Celeron D or the Sempron may become the winner. Note also that the K8 architecture of the Sempron 3100+ for Socket 754 systems provides the best performance in almost all the gaming applications in comparison to any other value processor.
AMD assigned ratings to the processors of the Sempron family basing on the results these processors had in Winstone and SYSmark benchmarks. It was done for the performance of each Sempron to be higher than that of the Celeron D of the same price. That’s exactly what we see in the diagrams above.
Encoding audio into MP3 format is a specific task. The memory bandwidth and the size of the cache matter little here. It is the raw speed of the central processor that’s important now. Keeping this fact in mind, you can easily explain the results: the Semprons for Socket A are slightly faster than the Celeron D, while the Sempron 3100+ loses to its Socket A mates with a lower rating because they work at higher clock rates. All in all, there are no surprises here.
The cache size and the memory speed don’t play a crucial part when encoding raw video into MPEG-2 format, either. However, unlike in the audio encoding task, Intel’s CPUs are much faster thanks to the efficient support of SIMD instructions (a fast SSE2 unit and support of SSE3).
The Celeron D processors are faster than the Semprons when encoding video into MPEG-4 format with the DivX codec. The Sempron 3100+ surpasses its Socket A mates, but can only compete with the Celeron D 325 that is clocked at 2.53GHz.
It’s practically the same with the XviD codec. Intel’s transitioning the Celeron series to the Prescott core resulted in a considerable performance boost. Older Celeron and Pentium 4 processors (on the Northwood core) used to lose to AMD’s products in this test, but now the Sempron and the Athlon 64 find themselves on the losing side.
It is the memory subsystem that bears the highest load in archiving tasks. Unlike semi-synthetic memory subsystem tests like PCMark04, archiving programs win from both a high CPU-memory bandwidth and a low latency. Besides that, the size of the L2 cache is important here, too.
Top lines of the diagram are populated by the processors with the K8 architecture, including Sempron 3100+, which has outperformed both 2.8GHz Pentium 4. This comes as the memory controller integrated into the K8 core reduces the memory latencies greatly. As for the Celeron D, they are faster than the Socket A Semprons due to a wider memory bus as well as efficient data prefetch algorithms implemented in the Prescott core.
As a kind of experiment, I decided to measure the speed of an antivirus check of a folder with files of various types with a popular antivirus program.
Once again the Semprons for Socket A can’t boast high speeds. The Sempron 3100+, being much faster than the Semprons of smaller ratings, is capable of challenging the senior Celeron D clocked at 2.93GHz.
Although Intel’s top-end CPUs are faster in Photoshop than their rivals from AMD, it’s different with the value products. The senior Sempron models – Socket 754 and Socket A models alike – run Photoshop almost as fast as the senior Celeron D CPUs. Comparing the junior models of the two series I would say that Intel’s ones are somewhat swifter.
Compiling program code is one of the few tasks which AMD’s value CPUs are better in than their competitors from Intel. Note that even the Barton-core Sempron 3000+ is faster than the 2.8GHz Pentium 4 here.
The Sempron family processors are overall better in this type of tasks than Celeron D CPUs. Considering that the Celeron D lacks Hyper-Threading, this result is natural. The Athlon XP used to be superior in such applications too until the Pentium 4 started supporting that technology.
This test session ended with a rather unexpected outcome. Earlier, shopping for a value processor was easy as junior Athlon XP models were much faster than Celerons. The release of the two new CPU families, however, changed the situation in this market sector dramatically.
The performance bar of Intel’s inexpensive products has been raised by the 256-kilobyte L2 cache and the 533MHz FSB of the new Celeron D. AMD, on the contrary, lowered the speed of the Sempron compared to the Athlon XP, clocking them at lower frequencies or giving a smaller L2 cache. The company, however, claims that the Sempron is faster than the Celeron D:
AMD is true: the Sempron is faster in the two benchmarks they used (Winstone and SYSmark). But we have just carried out a more comprehensive testing session, which says that the Celeron D is often better than the Sempron in real-life tasks. We found our own “performance ratings” of the processors by calculating the average of the processor’s relative speeds in the 27 benchmarks we used. The results are normalized to those of the Intel Celeron 2.8:
It is clear that the Semprons with ratings of 3000+ and lower, i.e. intended for Socket A systems, are slower than the Celerons working at the frequency that equals that rating, by about 5-6%.
We shouldn’t forget about the Sempron 3100+ model, though. It must be viewed apart from the rest of the family, since this Paris-core processor for Socket 754 systems with the K8 architecture is incomparably better than its junior mates. According to our averaged rating, the Sempron 3100+ is faster than the Sempron 3000+ by about 15%! That’s why its performance is the highest among all the value processors I have reviewed today, approaching that of the Pentium 4 2.8GHz and Athlon 64 2800+.
Of course, the appeal of a value processor is not only in its performance, but also in price. It is quite unreasonable to talk about advantages of a value CPU series without taking the price factor into account. So the next diagram shows you the price/performance ratios for AMD’s Sempron and Intel’s Celeron D series. The X axis shows price, the Y axis – weighted-mean performance:
This picture is self-explanatory I guess. I only wish to draw your attention to the fact that the “averaged” graph is not true for all possible applications. For example, the Sempron family processors are much faster than the same-price Celeron D CPUs in a number of games.