by Ilya Gavrichenkov
10/11/2004 | 09:39 AM
AMD’s announcement of the new Sempron processor family has passed without much racket, because the Semprons are mostly well-known Athlon XP models with a new marking and a slightly adjusted performance rating formula. So, the Sempron could hardly hope to become a sensation, although we still dedicated an article to checking out its performance in comparison to its immediate market rival, the Celeron D processor from Intel. Overclockers ignored the new processor from AMD completely since the majority of Semprons are based on long-known and thoroughly-investigated cores Thoroughbred-B, Thorton and Barton. Besides that, it’s not easy to overclock Semprons for the Socket A platform due to their locked multiplier – it is locked to avoid unscrupulous resellers remarking such processors into more expensive Athlon XP CPUs.
The Sempron family, however, isn’t restricted to the now-obsolete Socket A platform – there is a Sempron 3100+ model for Socket 754, the only Sempron to feature the K8 architecture found in Athlon 64 CPUs. Like the Athlon 64, the Sempron 3100+ features a HyperTransport bus and an integrated memory controller with support of DDR400 SDRAM. This makes it an interesting sample to experiment with – if the Sempron 3100+ can overclock well, it can form a good foundation for a low-cost system of some speed. In other words, just the system many overclockers would prefer to use! That’s why we want to dedicate a separate article to the overclocking capabilities of the Sempron 3100+.
By the way, AMD’s policy with respect to the Socket 754 platform results in the company’s pushing it down into the low-end price category. The slowest and low-performance processors with the K8 architecture are released for Socket 754 – besides the Sempron 3100+, they are several Athlon 64 models with ratings of 2800+, 3000+ and so on. But you may remember that the Socket 754 platform used to be the only available for middle-range and top-end Athlon 64 CPUs until the arrival of Socket 939. So now Socket 754 is like an unwanted child for AMD. On the one hand, the company’s trying to sink it down into the bottom of the market as quickly as possible, but on the other hand the current market realities make AMD offer rather fast processors for this platform since there’re quite a lot of computers with Socket 754 mainboards around. Exorcizing the negative aura AMD itself has created around the Socket 754 platform, saying it’s inferior to Socket 939, we’d find that Socket 754 systems are not as hopeless as they seem. Their main disadvantage in comparison to the Socket 939 platform – a single-channel memory controller – is not a very big deal actually. Thanks to the integrated memory controller, the strength of the memory subsystem of AMD’s modern CPUs is in low latencies rather than in high bandwidth. So adding another memory channel does have a positive effect, but the value of this effect is only 3-5% of performance growth. Considering all that, Socket 754 seems to be a most appropriate platform for overclocking. There are inexpensive processors for it, which show good overclockability, while the performance of the platform itself has no artificial ceilings, save for the lack of a second memory channel.
The Sempron 3100+ is the slowest and cheapest embodiment of the K8 architecture in silicon. Targeting low-end systems, it has just some of the characteristics the top-end models (Athlon 64 and Opteron) have. The two main deficiencies:
While it’s all clear with the smaller amount of cache memory, the lack of support of 64-bit extensions requires some comment. Since AMD conceived the Sempron family as a response to Intel’s release of the Celeron D series, the Sempron 3100+, although the major representative of the family, found itself in the value product category. Historically, value processors from both AMD and Intel are made not just slower, but also lacking certain functions – that’s the CPU manufacturers’ way of encouraging us to purchase more expensive processors. For example, the Celeron D series from Intel have a cut-down L2 cache and lack Hyper-Threading technology and use a slower system bus (compared to the Pentium 4). AMD, on its part, disabled the 64-bit extensions in the Sempron 3100+. However, this is not as big a loss as the slower bus coupled with the disabled Hyper-Threading, because even owners of top-end Athlon 64 processors, fully supporting the AMD64 technology, don’t practically use 64-bit modes. The reason is simple – there’re practically no operating systems and applications capable of working in the 64-bit environment. And there will be no such software at least until the release of Windows XP with support of AMD64 and EM64T. Right now Microsoft is only offering a beta version of this OS, regularly postponing its official release. According to the latest information, it is scheduled for the first half of 2005. So, the Sempron 3100+ has no significant disadvantages compared to the Athlon 64 as yet.
As for other progressive technologies, implemented in the Athlon 64, the Sempron 3100+ has them all. It features Enhanced Virus Protection (the so-called NX-bit), which is supported by Windows XP SP2 and helps to protect your computer from malicious program code, and offers Cool’n’Quiet technology that reduces the heat dissipation and power consumption of the processor when it’s idle.
The following table lists the formal characteristics of the Sempron 3100+ in contrast to Athlon 64 CPUs for Socket 754 and Socket 939 systems:
The Paris core the Sempron 3100+ is based on originates from the NewCastle core. As a result, the Socket 754 Sempron is made of CG stepping dies with such consequences as better overclockability, support of the 2T memory timing that improves compatibility with various DDR SDRAM memory modules.
The marking of the Sempron 3100+ looks like that:
And here’s what diagnostic utilities have to say about it:
Based on CG stepping cores, Sempron 3100+ CPUs should be highly overclockable. Why? Top-end Athlon 64 processors of this stepping are rated to work at 2.4GHz, so we should be able to overclock the Sempron 3100+ to this point at least. Or even more, considering that the Sempron 3100+ die is smaller than the Athlon 64 one due to the reduced L2 cache! We hope we’ve intrigued you already, but before getting to the tests, let’s discuss the problem of choosing an overclocking-friendly mainboard.
The standard frequency of the Sempron 3100+ processor is 1.8GHz. Its multiplier is limited from above, like with any Athlon 64-like processor, i.e. you cannot increase it above the nominal value, 9x. The reduction of the multiplier is possible as necessary for Cool’n’Quiet technology, but it can’t help us during overclocking. Thus, the only way of overclocking the Sempron 3100+ is increasing the frequency of the clock generator above the standard 200 megahertz (some mainboard manufacturers still call it FSB frequency in the BIOS Setup). But in the majority of Socket 754 systems this very clock generator also forms other frequencies like those of the HyperTransport, AGP and PCI buses, and overclocking the Sempron 3100+ may lead to instability of the peripheral devices and chipset. For example, to overclock a Sempron 3100+ from 1.8GHz to 2.4GHz, you have to increase the clock generator frequency to 266MHz. In this case, the AGP frequency grows from 66MHz to 89MHz, and none of modern graphics cards would be operational at so high a clock rate. So, overclocking of Athlon 64 CPUs, and also of the Sempron 3100+, can’t be performed fully due to some problems with peripheral buses rather than to any limitations of the CPU itself. This problem was really serious until quite recently: there were no mainboards available to the masses that would clock the CPU and the AGP/PCI buses asynchronously. The situation is different now as NVIDIA has introduced its nForce3 250 chipset, capable of such asynchronous clocking. In other words, nForce3 250-based mainboards can overclock the CPU without raising the AGP/PCI frequencies.
So, we’ve got the answer: a Socket 754 mainboard on the NVIDIA nForce3 250 chipset is the best choice for overclocking the AMD Sempron 3100+ processor. Older mainboards on other chipsets are unlikely to allow getting the most from the processor.
Some things must be said about the memory frequency. The memory controller is integrated into the Sempron 3100+ core (like with Athlon 64 CPUs), and the memory frequency directly depends on the clock rate the processor works at. Higher CPU clock rates automatically lead to a higher memory clock rate. To form the memory frequency, the Sempron 3100+, normally clocked at 1.8GHz, offers the 1:9 divisor for DDR400, 1:11 for DDR333 and 1:13 for DDR266. There are no other divisors possible, so you have to think carefully before setting up the memory frequency at overclocking; otherwise, the memory chips may become unstable. The BIOS Setup of Socket 754 mainboards usually offers you to set up the memory frequency by choosing DDR400/DDR333/DDR266 (or 200/166/133MHz) in an appropriate menu. To understand what effective frequency the memory is working at in different overclocking situations, we offer you the following table:
So, it’s not so hard to set the right frequency for almost any DDR SDRAM modules, although this frequency changes along with the CPU clock rate. What you must do is choose an appropriate divisor in the BIOS Setup.
Now that you are warned about the possible troubles with choosing and configuring the overclocking platform, let’s proceed to the account of our overclocking experiments.
We’d like to put down our goals first. We don’t want to set any overclocking records – our aim is simply to check out the performance of the Sempron 3100+ in real-life conditions – just as you may want to use it in your own system. So we didn’t use any special cooling methods and didn’t modify anything in the mainboards. We just took an ordinary processor sample, a standard cooler and an off-the-shelf mainboard and did our overclocking with simple and easy methods – by changing the BIOS Setup parameters.
First of all we decided to watch the Sempron 3100+ overclock on a rather old, but still popular ABIT KV8-MAX3 mainboard. Against our own recommendations explained above, this mainboard is based on the VIA K8T800 chipset.
This mainboard is widely recognized by PC enthusiasts; it allows increasing the frequency of the clock generator in a rather wide range and also controlling the voltage of the CPU (Vcore can be increased above the nominal by 1 to 350 millivolts), memory slots (from 2.5 to 3.2v), AGP slot (1.5-1.65v) and HyperTransport bus (1.2-1.4v).
Thus, our first testbed was configured like that:
Our testing started out smoothly enough. Without fine-tuning the system or raising the core voltage, we reached 229MHz of the clock generator frequency. Thus, we easily made our Sempron 3100+ speed up from its normal 1.8GHz clock rate to 2.06GHz! The system was absolutely stable at that.
But a further increase of the FSB clock rate – even by a single megahertz – led to the system becoming non-operational. The computer would just hang up at a launch of any 3D application. We tried to persuade with it by raising Vcore and reducing the frequency of the DDR SDRAM modules, but to no effect. It means that our graphics card became unstable at a higher AGP bus frequency, and we didn’t really hit the frequency limit of the Sempron 3100+. Due to the synchronous architecture of the VIA K8T800 chipset, the AGP frequency grew from 66.6MHz to 76.3MHz when the clock generator frequency was set to 229MHz. The graphics card we used definitely didn’t like that.
We had already met such troubles with VIA’s chipsets before. You should also know that the most fastidious part of synchronous chipsets from VIA at overclocking is not the AGP bus even, but the SerialATA controller integrated into the South Bridge. This controller is clocked synchronously with the V-Link bus, which in its turn is clocked relative to the PCI frequency in the K8T800 (and in other VIA’s chipsets, too). Keeping this fact in mind, we refused to use a Serial-ATA hard disk drive in our testbed; otherwise, our experiments would have ended up much sooner – at about 220MHz frequency of the clock generator.
Thus, our theoretical reasoning above is confirmed in practice. To overclock the Sempron 3100+ well, you need a modern mainboard capable of clocking the AGP and PCI buses asynchronously, i.e. a mainboard on the NVIDIA nForce3 250 chipset. So, the second part of our testing was continued on an EPoX EP-8KDA3+ mainboard.
The EPoX EP-8KDA3+, like the ABIT KV8-MAX3, offers widest opportunities to the overclocker: you can change the clock generator frequency from 200 to 400MHz, clock the AGP bus independently at 66 to 100MHz (with a symmetrical change of the PCI bus frequency), control the CPU multiplier (reduce it below the nominal), control the voltages (increase Vcore to 1.8v, Vmem to 2.8v, Vagp to 1.8v, Vchipset to 1.75v; the nominal voltage of the nForce3 250 chipset is 1.6v). Thus, the EPoX EP-8KDA3+ is an excellent overclocking tool (running a little ahead, we should say it did well throughout our tests with the Sempron 3100+, too).
So, we continued investigating the overclocking potential of the Sempron 3100+ on the following testbed:
Overclocking the Sempron 3100+ went more cheerfully on this testbed. Setting the clock generator frequency to 230MHz for a start, we enjoyed an absolute stability of our system. Clearly, the nForce3 250 chipset allows avoiding those overclocking impediments the K8T800 imposes on Socket 754 systems.
Thanks to that, we easily reached 250MHz clock rate of the clock generator – just adjusting the appropriate item in the mainboard’s BIOS Setup. The processor frequency became 2.25GHz and the CPU itself remained perfectly stable. We should note, though, that the EPoX EP-8KDA3+ mainboard thinks that the regular voltage of the Sempron 3100+ is 1.5v rather than 1.4v as it should be. So, the processor voltage was initially higher than normal.
To continue our overclocking further we had to adjust some of the system settings. The memory frequency was growing along with that of the clock generator (1:9 divisor in the BIOS Setup), and it reached DDR500 at 250MHz of the clock gen. That was the guaranteed maximum of the PC4000 modules we used in the test system. We changed the divisor to 1:11, so that our overclocking wouldn’t be limited by the capabilities of the memory chips.
Steadily lifting the clockgen frequency up we reached 270MHz, and the CPU clock rate became 2.43GHz. However surprising it may seem, the Sempron 3100+ was absolutely stable at that, even without our increasing its voltage over 1.5v. Thus, quite effortlessly, we sped the Sempron 3100+ up from 1.8GHz to the frequency the top-end K8 processors are clocked at. I mean the Athlon 64 3700+ and 3800+ and the yet-unannounced 4000+.
Unfortunately, we couldn’t proceed further without adjusting the CPU voltage. The Sempron 3100+ is rated to work at 1.4v, while other processors of the analogous architecture work at 1.5v, so we decided that it would be safe to raise the voltage of our sample to 1.65v.
This Vcore value let us conquer the next frequency peak – 280MHz of the clock generator. The system was actually stable at 281MHz, too, and hang-ups and instability only occurred at 282MHz, but we rolled back to 280MHz for more stability; the effective CPU clock rate was very impressive anyway – 2.52GHz!
So, it is possible to overclock the Sempron 3100+ processor from 1.8GHz to 2.52GHz without special cooling and with a minor voltage increase. The frequency gain is 40%, which is good even compared to legendary processors, “overclocker’s choices” of previous times.
The memory frequency was 229MHz at that (see the table above).
Unfortunately, it is not possible to set a smaller devisor (1:9). In this case, the memory frequency would equal the clock rate as defined by the clock generator, i.e. 280MHz. Only DDR600 SDRAM modules would be capable of working at so high a frequency, and they are not widely available now. However, the system with the overclocked Sempron 3100+ demonstrates good results even in memory subsystem tests.
As for the temperature factor, our Sempron 3100+ remained rather hot at overclocking. We didn’t enable Cool’n’Quiet technology, since this would affect the test results, but you’d better use it if you care about the health of your processor. It works fine with the overclocked Sempron, by the way. When the CPU load is low, the processor’s driver just reduces its multiplier to 5x. So, if the Sempron 3100+ normally drops its frequency to 1GHz, our overclocked sample dropped it to 1.4GHz, which was good, too.
But back to the temperature. The Sempron 3100+ overclocked to 2.52GHz was about 64°C hot when idle, and as high as 79°C hot in test applications (according to the mainboard’s data). We’d like to note that working at its nominal frequency and voltage (1.8GHz and 1.5V) the Sempron 3100+ was 47°C hot in the idle mode and up to 60°C hot under load in our system.
Now we’ve got to the most interesting part of this review – the examination of the performance level of the overclocked Sempron 3100+ processor. If this level is rather low, why bother about overclocking it at all, yes? So, our Sempron 3100+ is overclocked to 2.52GHz, and we’ve got two Intel Pentium 4 3.4GHz CPUs on Prescott and Northwood cores, and another Socket 754 processor, the Athlon 3400+ to serve as reference points.
We used the following hardware in the test platforms:
We performed our 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.
To get a preliminary estimate of the performance level of the Sempron 3100+ overclocked to 2.52GHz we launched the SuperPi test, popular among overclockers:
As you see, this benchmark says the performance of the overclocked CPU is real high. Thanks to its clock rate, which is higher than that of any Athlon 64, the Sempron 3100+ outperforms the Athlon 64 3400+ as well as both Pentium 4 with a frequency of 3.4GHz! Well, if it is the same in other tests, the Sempron 3100+ overclocking should be considered as a highly profitable business! Let’s see…
The Futuremark PCMark04 test uses the Hyper-Threading technology, so the results of the Pentium 4 processors are higher than those of the Athlon 64. The overclocked Sempron 3100+ can’t win this test for the AMD camp, but it is anyway much faster than the Athlon 64 3400+.
The Pentium 4 CPUs are again at the top of the diagram in the memory subsystem test. They work in systems with a dual-channel memory controller, while the Sempron 3100+ and the Athlon 64 model we tested have a single-channel controller. That’s why the K8 architecture doesn’t show its strength here, but let’s see what we have in other tests.
AMD’s CPUs take their revenge in 3DMark2001 SE: the Sempron 3100+ clocked at 2.52GHz looks excellent. It leaves both 3.4GHz Pentium 4 models far behind and is just slightly slower than the Athlon 64 3400+ which has a four times larger L2 cache.
The overclocked Sempron 3100+ wins newer benchmarks – 3DMark03 and 3DMark05!
The CPU test from 3DMark03 says the Sempron 3100+ is slightly slower than the Athlon 64 3400+ clocked at its regular frequency.
The analogous test from the recently-released 3DMark05 puts the overclocked Sempron 3100+ on top, though.
No comments necessary. The gaming tests suggest that the reduced L2 cache doesn’t retard the speed of the Sempron 3100+. The performance of this value CPU, when overclocked to 2.52GHz, is on a higher level than that of performance-mainstream processors. The exceptions are Aquamark3, an application that actively uses Hyper-Threading, and Doom 3 where the L2 cache size is very important. Well, the overclocked Sempron 3100+ is quite fast in these two tests, too.
Encoding audio files into MP3 format is a specific task as it but slightly depends on the memory speed or the size of the L2 cache. That’s why the overclocked Sempron 3100+ is much faster than the Athlon 64 3400+, but not enough fast to outperform the 3.4GHz Pentium on the Northwood core.
The size of the cache and the speed of working with the memory are of little importance at encoding “raw” video into MPEG-2 format, but Intel’s CPUs benefit here from the support of SIMD instructions, including a fast SSE2 unit and SSE3 support. The result of the Sempron 3100+ is anyway good compared to the Athlon 64 3400+ - the high clock rate helps a lot in this test.
The same goes for this test – encoding video into MPEG-4 format with the DivX codec.
When using another codec (XviD), the overclocked Sempron 3100+ only loses to the Prescott-core Pentium 4.
Although the amount of L2 cache memory affects the performance heavily in this type of tasks, the overclocked Sempron 3100+ is faster than the Athlon 64 3400+. As for Intel’s CPUs, they fall behind the K8-architecture processors due to higher latencies of their memory subsystems.
The overclocked Sempron 3100+ is on top again, outperforming both Pentium 4 models as well as the Athlon 64 3400+.
Although Pentium 4 processors are traditional winners in the Photoshop test, the overclocked Sempron 3100+ shows an impressive performance now. Thanks to its high clock rate it is capable of competing with 3.4GHz processors from Intel!
The Sempron 3100+ looks nice in tasks of that type, too. What’s especially remarkable, the Sempron clocked at 2.52GHz does ray-trace rendering faster than the 3.4GHz Prescott-core Pentium 4, whose Hyper-Threading technology should have provided it an additional advantage in this test.
Having thoroughly examined various overclocking-related properties of the new Sempron 3100+ processor from AMD, we can claim that it’s a real overclocker’s choice. Costing about $120 – and this price is going to go down to $100 in a month – this CPU allows clocking itself by 40% higher above the nominal frequency. Such overclocking doesn’t require any special equipment – you only need an ordinary cooler and mainboard. The only thing you should consider carefully is that the mainboard would allow fixing the AGP/PCI bus frequencies at some point, i.e. clock them independently of the clock generator frequency.
We should note that the Sempron 3100+ processor delivers good performance even without overclocking. Having only 256 kilobytes of L2 cache memory, it is just 1-2% slower than the full-featured Athlon 64 2800+. Of course, the lack of AMD64 technology may disappoint some users, but this technology is not very important today – 64-bit operating systems and applications are not widespread, and the competitor desktop processors hasn’t yet acquired a similar technology yet.
As for the gain you receive by overclocking the Sempron 3100+, we enjoyed a performance growth of about 25% by clocking our sample at 2.52GHz (40% frequency boost). Thanks to that, the overclocked Sempron 3100+ could outperform the Pentium 4 3.4GHz as well as the Athlon 64 3400+ by about 5% in average.
Winding up this review, we want to tell you that Sempron 3100+ models on the new 90nm core are coming up to market soon. Their overclockability may be no worse than that of the 0.13-micron Paris-core sample we have tested today, but we can’t claim it of course without running more tests. So, we will surely return once again to investigating the overclocking potential of new-generation inexpensive processors from AMD.