Beruflich Dokumente
Kultur Dokumente
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
The arrival of low-cost Athlon 64 processor models will help AMD 64bit processor family to become really popular, which is a very important aspect of the cut-throat competition with Intel, which is actively preparing for the launch of the new Prescott based CPU family. Lets try to analyze the situation in order to find out what the advantages of the new AMD offer are for us, the users, and what benefits the new AMD Athlon 64 3000+ processor will bring us.
1 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
While everybody was waiting for the new Athlon 64 3000+ processor, they were absolutely sure that it will differ from the top Athlon64 3200+ model only by the core clock frequency. This confidence was based on the fact that Desktop Replacement (DTR) AMD Athlon 64 3000+ processor for the mobile solutions available since the end of September 2003 works at 1.8GHz core frequency and has no other differences from the DTR Athlon 64 3200+. However, AMD decided to do a completely different thing this time: the new Athlon 64 3000+ works at the same 2GHz clock frequency as the elder Athlon 64 3200+ model. The difference between these two solutions is the size of L2 cache memory. L2 cache of the new Athlon 64 3000+ is cut down to 512KB compared with the 1MB L2 cache of Athlon 64 3200+ CPU. You can check this with the help of any diagnostics utility revealing the processor specifications:
2 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
As you can see, the differences in the specs of Athlon 64 3200+ and Athlon 64 3000+ are minimal. Besides the smaller L2 cache, the new processor is just the same as the 3200+ model announced on September 23, 2003. Even the core stepping is the same. All this means that Athlon 64 3000+ is based on the same semiconductor die as the top Athlon 64 processor models. AMD simply disables half of the L2 cache memory in Athlon 64 3000+ just like they did in Athlon XP CPUs based on Thorton core. As a result, you can come across three really different AMD Athlon 64 processor models in the todays CPU market (I mean three models, which differ by more parameters than just the core clock frequency): Athlon 64 FX-51 Socket 940 2.2GHz Athlon 64 3200+ Socket 754 2.0GHz 0.13micron, SOI 105.9 mln 193 sq.mm 1.5V Single-channel, 64-bit DDR400/ DDR333/ DDR266 SDRAM + Athlon 64 3000+ Socket 754 2.0GHz 0.13micron, SOI 105.9 mln 193 sq.mm 1.5V Single-channel, 64-bit DDR400/ DDR333/ DDR266 SDRAM +
Packaging Frequency Manufacturing 0.13micron, SOI technology 105.9 mln Number of transistors 193 sq.mm Die size 1.5V Nominal Vcore Integrated memory Dual-channel, 128-bit controller Registered DDR400/ DDR333/ Supported memory types DDR266 SDRAM + ECC support
3 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
L1 cache L2 cache CoolnQuiet technology SIMD instructions support AMD64 technology support
128KB (64KB for instructions and 64KB for data) 1024KB (exclusive) SSE2/SSE/3DNow! +
128KB (64KB for instructions and 64KB for data) 1024KB (exclusive) + SSE2/SSE/3DNow! +
128KB (64KB for instructions and 64KB for data) 512KB (exclusive) + SSE2/SSE/3DNow! +
Athlon 64 3000+ features smaller 512KB L2 cache, which is also indicated by the CPU marking:
The number 4 in the third position from the end of the marking implies that the L2 cache size of the CPU with 3000+ performance rating is equal to 512KB. The Athlon 64 3200+ processor is marked with number 5 in this position.
The appearance of Athlon 64 processors with smaller amount of L2 cache memory is quite understandable. No
4 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
doubt that AMD did need less expensive CPU models to be able to win a market share. And then AMD faced a question: how should they arrange the production of less expensive Athlon 64 processors at the minimal expenses? Of course, the production cost of these processors is pretty high because they are based on a rather big core. However, reducing the die size will require additional investments into the R&D (Research and Development), even if they will only have to reduce the size of the on-die L2 cache memory. Therefore it will make sense only in case the production volumes will be extremely high. At the same time large die size leads to pretty high share of defective dies, which cannot be used in Athlon 64 3200+ CPUs. Since L2 cache memory of Athlon 64 3200+ occupies more than 50% of the die, most defective dies are discarded because of the problems with the on-die cache memory. It is quite logical that disabling half of the L2 cache memory can help revive these dies. Therefore the launching of Athlon 64 3000+ with 512KB L2 cache is also a good way to get rid of some defective dies. Especially since the yields are evidently pretty low, because AMD64 architecture is still rather new. As a result, AMD Athlon 64 3000+ is an attempt to kill two birds with one stone: the users will get low-cost processors based on AMD64 architecture, and the manufacturer will have a great opportunity to get rid of some defective dies, which cannot be used for more expensive processors, and win some additional revenue from that. Since Athlon 64 3000+ is based on the same dies as Athlon 64 3200+, all its specifications are just the same as those of its elder brother. It is also true for the support of CoolnQuiet technology, which we have already told you about in great detail in our AMD Athlon 64 3200+ CPU Review, and for all thermal parameters: Athlon 64 FX-51 Athlon 64 3200+ Athlon 64 3000+ 1.5V 1.5V 1.5V 89W 89W 89W
According to AMDs current plans, Athlon 64 3000+ processors will remain the slowest model in the family for the entire family existence. In other words, AMD is not going to release any processors with lower performance rating within this family. As for the life cycle of the new Athlon 64 3000+ processor, AMD is planning to continue making them for at least Q304. This way, these processors will stay in the market for quite a while. However, Athlon 64 3000+ will still be unable to last longer than Socket A processors, which should be manufactured until mid 2005. Speaking about the future of CPUs based on AMD64 architecture and featuring 512KB L2 cache, we should definitely take into account the companys official roadmap:
5 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
As you can see, there is one more core there aka Newcastle, which distinguishing feature is also smaller 512KB L2 cache. Therefore, many of you could have considered Athlon 64 3000+ to be the first representative of the Newcastle family, but this is not quite correct. Newcastle core will be used for the upcoming Socket939 Athlon 64 processors, which will also feature a dual-channel memory controller besides the 512KB L2 cache. The idea behind Newcastle core has to do with the intention to reduce the production costs for the new AMD64 processors by reducing the die size. The freshly released Athlon 64 3000+ is designed for Socket754 mainboards and features a single-channel memory controller. It is based on Clawhammer core and features the entire L2 cache memory of the Clawhammer, one half of which is disabled. This way it would be completely incorrect to say that the new Athlon 64 3000+ is the first processor on the new core. It looks as if Athlon 64 3000+ were about to stay the only processor with 512KB L2 cache for Socket754, at least for the next half a year.
6 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
1024MB Registered DDR400 SDRAM (Mushkin High Performance ECC Registered 2 x 512MB, 2-3-2-6). Graphics card: ASUS RADEON 9800XT (Catalyst 3.10). Storage subsystem: 2 x Western Digital Raptor WD360GD HDDs in RAID 0 array. Notes: The memory (registered and unbuffered) was working in the same mode in all cases, namely with the timings set to 2-3-2-6; We ran all tests in Windows XP SP1 with installed DirectX 9.0b pack.
Overclocking
Before we pass over to the actual benchmarks results, we decided to undertake a few overclocking attempts to figure out the overclocking potential of the new AMD Athlon 64 3000+ processors. The thing is that the CPUs of this price range are usually considered a good buy for overclocking purposes. That is why it would make a lot of sense to add the performance rates for the overclocked AMD Athlon 64 3000+ processor to the benchmark results. I would like to stress right away that we carried out all our overclocking experiments without any extreme cooling systems involved. We used the regular cooler shipped with the boxed Athlon 64 3000+. At first I would like to say a few words about the overclocking friendly features of this CPU. Since Athlon 64 3000+ is based on the same core as Athlon 64 3200+, it doesnt allow increasing the clock frequency multiplier beyond the nominal 10x, just like its elder brother. However, you can set the clock frequency multiplier to a lower value, though it hardly makes much sense for overclockers. This way, we will have to overclock our Athlon 64 3000+ by increasing the FSB frequency. By the way, AMD has been very specific about it: they gave to understand very clearly that there wouldnt be any Athlon 64 processors in the market, which would allow increasing the clock frequency multiplier beyond the nominal value. This option will exist only in more expensive Athlon 64 FX processors targeted at dedicated hardware enthusiasts. Here I would also like to point out that we overclocked the CPU on the same mainboard we used for the entire benchmarking session: it was ABIT KV8-MAX3. Since this mainboard is based on VIA K8T800 chipset, the AGP and PCI bus frequency is increased simultaneously with the FSB frequency during overclocking. However, despite this fact we had to give up the idea of using a more advanced NVIDIA nForce3 150 chipset this time. NVIDIAs chipset doesnt support SerialATA and uses slow HyperTransport bus, which negatively tells on the performance in some contemporary games and professional applications. That is why it doesnt make much sense today to build Socket754 systems with an NVIDIA based mainboard. However, VIA K8T800 is not a bad choice for overclocking needs. Having tested 13 mainboards based on NVIDIA nForce3 150 and VIA K8T800 chipset we didnt notice any significant differences in the actual overclocking performance (see our Socket754 Platform: 13 Mainboards Roundup). Maybe there could appear some problems with the AGP and PCI devices on VIA K8T800 based mainboards when the FSB frequency is increased too much. However, when we overclock todays Athlon 64 processors with C0 core stepping and air cooling solution, these problems are very unlikely to occur. We reached the maximum of the processor potential (2.3-2.4GHz) much sooner, then the problems with external devices. Since Athlon 64 3000+ processors are based on absolutely identical cores as Athlon 64 3200+, we expected them to show similar overclockability. If you read our AMD Athlon 64 3200+ CPU Review, you should remember that we managed to overclock our processor up to 2.34GHz core frequency. As a result, we expected our Athlon 64 3000+ to be able to notch something close to that. However, our practical tests showed a slightly different result. Before will actually tell you about our overclocking achievements, I would like to mention that we used different memory for overclocking. The Corsair CMX512-3200LLPRO memory, which we use for regular testing works perfectly well at the nominal
7 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
frequency and with low memory timings. However, as soon as we increase the memory bus working frequency, the modules get significantly less stable. That is why I decided to use special overclockers memory for our overclocking experiments. It was OCZ PC4000 Dual Channel Gold Edition, which is guaranteed to work fine at the frequencies up to 500MHz. This memory allowed us to overclock Athlon 64 3000+ processor without increasing the memory frequency divider, which was set at 1/10 of the CPU frequency through the entire test session (DDR400, if we use the terms of BIOS Setup). To ensure that we achieve better results during overclocking, we increased the processor Vcore by 10%, namely up to 1.65V, and then started increasing the FSB frequency very smoothly. The first problems turned up when we reached 222MHz FSB frequency. To be more exact, the RAID array simply refused to work properly. It turned out that the SerialATA RAID controller built into the VIA VT8237 South Bridge is very sensitive to the PCI bus frequency increase. When the FSB frequency reached 222MHz, the PCI frequency was only 37MHz, however, it was more than enough for the SerialATA RAID controller to lose its stability. That is why I carried out all ongoing overclocking experiments with a Parallel ATA hard disk drive Western Digital Caviar WD400JB. Luckily, Parallel ATA controller of the VIA VT8237 South Bridge behaves much better during overclocking. However, the change of the disk subsystem didnt improve the situation that much. Having reached 226MHz FSB the system grew unstable again. This time it was the because of the CPU. To make sure that this not an accidental result, we went 1MHz back and ran the entire set of stability tests. We discovered no problems whatsoever, that is why I have every right to claim 2250MHz to be the maximum result our CPU managed to achieve during overclocking.
As you remember, the processor Vcore during overclocking was set to 1.65V, while the increase of the FSB frequency to 225MHz pushed the other system buses to 450MHz for the memory bus, 37.5MHz for the PCI bus,
8 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
75MHz for the AGP bus and 900MHz for HyperTransport bus. I would hardly call this result a good one. We managed to speed up the CPU by only 12% above the nominal value. Moreover, we failed to reach the frequency our Athlon 64 3200+ achieved during overclocking, even though the two processors are based on the same core. It probably has to do with the fact that for Athlon 64 3000+ processors AMD uses defective dies, which cannot be used for more expensive Athlon 64 3200+ CPUs any more. However, even a 10% frequency growth is a good result, which shouldnt be neglected. Among the benchmark results later in this article you will also find the numbers indicating the performance of our system working with an AMD Athlon 64 3000+ processor overclocked to 2.2GHz by raising the FSB frequency to 220MHz. I considered it fair to test the system with 220MHz FSB because in this case all subsystems including SerialATA did work fine. The memory in this case worked at 440MHz and used 2.5-3-3-6 timings.
9 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
10 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
11 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
Athlon 64 3000+ demonstrates pretty good results in gaming applications. It is only 3-5% behind the Athlon 64 3200+ with twice as large L2 cache. It also looks quite attractive against the background of Intel Pentium 4 processor. In almost 50% of these tests, the new AMD CPU manages to defeat even Pentium 4 3.2GHz. And in the games, which usually favor Athlon processors, such as Unreal Tournament 2003 or Tomb Raider, Athlon 64 3000+ outperforms even Intel Pentium 4 Extreme Edition.
12 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
Overclocking Athlon 64 3000+ to 2.2GHz appears quite advantageous and improves the performance by another 7-8%. It is not only higher CPU clock frequency that contributes to the performance improvement here, but also faster memory, higher working frequencies of HyperTransport and AGP buses. Overclocking makes our hero run as fast as Intel Pentium 4 Extreme Edition, although it still fails to catch up with the Athlon 64 FX-51 working at 2.2GHz and featuring 1MB L2 cache and dual-channel memory controller. As for the performance ratio between the overclocked Athlon 64 working at 2.2GHz and featuring twice as small L2 cache, and the Athlon 64 3200+ with the full 1MB L2 cache and 2.0GHz core clock, we can see that the additional 200MHz do matter much more for the performance than another 512KB of the L2 cache memory. As a result, AMD Athlon 64 3000+ overclocked to 2.2GHz outperforms Athlon 64 3200+ with the nominal frequency of 2.0GHz in all gaming tests.
13 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
In Winstone tests Athlon 64 processors performed impressively great. For instance, Athlon 64 3000+ demonstrated the performance as high as that of Pentium 4 Extreme Edition 3.2GHz. One more pleasing thing is the performance difference between Athlon 64 3000+ and Athlon 64 3200+, which was equal to 2% at the most.
14 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
PCMark04, on the contrary, indicates dramatic lag of the AMD processors behind the Pentium 4 rivals. Frankly speaking, I even suspected this test to be optimized for Pentium 4 architecture, because the lag of all Athlon 64 processors even behind Pentium 4 3.0GHz was too evident. However, the obtained results can be easily explained. Have a look at a little bit more detailed PCMark04 performance table for Pentium 4 3.2GHz and Athlon 64 2300+: Athlon 64 3200+ Pentium 4 3.2 Multithreaded test 1 2.7197 5.3799 File Compression 31.269 49.375 File Encryption Multithreaded test 2 24.287 35.7 File Decompression 12.32 13.79 Image Processing Multithreaded test 3 2002.6 2619.9 Virus Scanning 3.128 1.9596 Grammar Check Singlethreaded tests 62.056 81.82 File Decryption
15 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
Audio Conversion Web Page Rendering WMV Video Compression DivX Video Compression Physics Calculation and 3D Graphics Memory - 64 Lines
Athlon 64 falls behind the Pentium processor in the first 6 subtests, because of their multi-threaded nature. Here PCMark04 starts two computational threads synchronously. In this case Pentium 4 is evidently in a better situation due to Hyper-Threading technology, which allows optimal processing of two data streams simultaneously.
As for the second part of the subtests, they mostly deal with streaming data encoding, and this is exactly the type of applications where Pentium 4 is initially much faster. The advantages of Athlon 64 during physical modeling and spell-checking are definitely not enough to make up for the victory of Pentium 4 processor in other subtests. This way, PCMark04 is a pretty fair test, since it doesnt use any specific optimizations for Pentium 4 architecture. However, all in all, its structure will always favor Pentium 4, so that it becomes an indisputable leader anyway. This is how the list of selected applications and their order affect the performance of our testing participants in PCMark04.
16 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
17 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
18 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
You can notice the influence of the L2 cache size on the performance in the tasks of the kind only during data compression in archiving utilities we used. In all other cases Athlon 64 3200+ and Athlon 64 3000+ perform very close to one another: the performance difference lies within 1%. As for the general performance of Athlon 64 3000+ in this type of tasks, it is pretty high during data archiving and MPEG2 data encoding. As for all other applications, such as MP3 encoding, WME and MPEG4 data encoding, and the like, Pentium 4 processors have always been much faster there. Even overclocking Athlon 64 3000+ up to 2.2GHz wont help, as well as the use of the dual-channel memory controller in Athlon 64 FX-51.
19 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
20 of 21
9/13/2012 10:17 PM
http://www.xbitlabs.com/articles/cpu/print/athlon64-3000.html
Final rendering in 3D modeling packages is not the best task for Athlon 64 processors. Pentium 4 is much more efficient here due to Hyper-Threading technology support. As for OpenGL tests, AMDs new solution is quite strong here and doesnt yield even a tiny bit to Pentium 4 3.2GHz.
Conclusion
Well, lets cast a glance at the price-list. The official price of AMD Athlon 64 3000+ processors is $218. It means that this CPU is positioned as a competitor to Pentium 4 2.8GHz, which is selling for the same money, according to Intels official price-list. However, as we saw in the tests, Athlon 64 3000+ can outperform even faster Intel CPUs in most benchmarks. We can also state that Athlon 64 3000+ is faster than Athlon XP 3200+. This way, it definitely means that this processor is one of the best buys in its price category from priceto-performance point of view. Of course, we could complain about low overclockability of the new Athlon 64 3000+ processor. However, these CPUs are pretty fast even at nominal frequencies. Besides, the possible 10% frequency growth could anyway ensure 7-8% extra performance. Moreover, there are some questions about the upgradeability of the Socket754 processors, where Athlon 64 3000+ actually belongs. Athlon 64 3700+ will be the top model for this type of processor socket and it may cause some concerns about Athlon 64 3000+ being the best buy in its price range. However, if you take a look at any other processor sockets you will see that all of them have pretty limited prospects for future upgrades. Socket478 will be soon replaced with Socket T. Socket A CPUs will be simply discontinued after a while, and Socket940 will be replaced with Socket939 in the nearest future. This way the concern is absolutely unjustified. As a result, there should be no doubts about AMD Athlon 64 3000+ being the best processor choice in the $200 price group today. And taking into account that this price category is the most popular among the DIY users (see our poll results here), we dare conclude that this product release will have a great influence on the wide-spreading and popularity of the AMD64 architecture in general. The major problem AMD might face now is the necessity to supply enough processors to satisfy the growing demand for these CPUs. And in conclusion I would like to point out that the reduced L2 cache of the Athlon 64 processor doesnt lead to any dramatic performance drops. The performance differences between Athlon 64 3000+ and Athlon 64 3200+ working at the same clock frequency but differing in L2 cache size do not exceed 5% in most cases. It means that the upcoming processors on the Newcastle core will be able to cope with their task OK. If they manage to retain this expected Athlon 64 performance level, the new AMD core will allow the company to reduce production costs and increase the revenues.
21 of 21
9/13/2012 10:17 PM