Original Link: http://www.anandtech.com/show/1241
PCI Speed and Overclocking: A Closer Look at A64 and P4 Chipsetsby Wesley Fink on February 16, 2004 9:42 AM EST
- Posted in
Sometimes things are not as they appear. Nowhere is this more true than the Computer Industry. Technology should be very straightforward, but all too often, marketing concerns get thrown into the mix and we end up with specifications with a marketing slant. It has been observed in reviews of the nForce3-150 boards that they often overclock better than VIA K8T800 boards. The nF3 sports a PCI/AGP adjustment in BIOS and it is usually stated that the nF3 should overclock better because of the fixed AGP/PCI bus.
Generally, we have also seen that Intel chipsets with Northwood processors reach incredible overclock levels. Intel has a widely published PCI/AGP lock in 865/875 chipsets, and it is clear that speeds like DDR550 (275x4 FSB) could never be reached without the fixed or locked AGP/PCI speeds. Most peripherals simply can't handle a PCI speed of 45.8 (275/6) or AGP of 91.7 (275/3) when specification is 33/66. Our Radeon 9800 PRO cards can rarely handle anything above about 70-72 on the AGP bus; this is typical of most current AGP cards.
Many tests have demonstrated the ability of Intel chipsets to lock the PCI/AGP bus, but questions are being raised again with the introduction of VIA's new PT880 chipset for the Pentium 4. VIA has been historically the only chipset maker not to utilize a PCI/AGP lock, but in the new PT880, VIA claims that they have implemented an asynchronous PCI/AGP bus. This is another way of saying that the processor and memory bus run at one speed and the PCI/AGP bus is not run at fixed divisors of that speed. This keeps the sensitive PCI and AGP cards in specification and allows a potentially higher overclock of the CPU and memory. Since this is VIA's first AGP/PCI, many are asking whether VIA really pulled this off - it has been such an issue with their chipsets in the recent past.
The most effective way to measure PCI/AGP bus speed is probably oscilloscope testing of bus input frequencies. To see an example of how this is done, you may want to check out tests on the nForce3-150 PCI bus at Lost Circuits. There is also a simpler tool available with the feature of measuring and displaying PCI bus speed, and that is PC Geiger. This device is a card that is mounted in a PCI slot and a digital display that provides all kinds of information - post codes, bus utilization, and PCI speed. The Taiwanese company, IOSS, markets PC Geiger in a joint venture with VICS in Japan. While PC Geiger provides many types of information, we will only be looking at PCI speed for these tests.
PCI Speed and Overclocking: How We TestedAt stocks speeds and a x1 setting on the Geiger, the PCI bus is reported as 33.3MHz:
The shot above was actually taken on an Asus P4C800-E running at a 250 (1000FSB) setting. When PCI lock is working, the PCI bus stays around the default of 33.3. AGP speed is a 2X multiple of PCI in the most common setup.
When the PCI bus floats, or is unlocked, the PCI speed floats with FSB settings. You can see this here where the base FSB is set at 210:
The PCI is 1/6 the base setting of 210, or 35. This level of overclock is generally not a problem even when the PCI bus floats.
Raising the base FSB to 220, PCI increases to 36.6, or an unlocked AGP of 73.
This is often a problem for current AGP cards. Our ATI Radeon 9800 PRO tests cards generally fail in intensive benchmarking above an unlocked base setting of around 218-220.
PCI Speed and Overclocking: Test ConfigurationOne of the nice things about the PC Geiger is that PCI speed is displayed as soon as boot begins. To check the PCI speed, any PCI card polling and disabling was turned off in BIOS. We monitored the PC Geiger reported PCI speed at the beginning of boot and as we entered the BIOS screen. We did not boot into an OS.
|Athlon64 FX51 Performance Test Configuration|
|Processor(s):||Intel Pentium 4 3.2C
AMD Athlon64 3200+
|RAM:||2 x 512Mb OCZ 3500 Platinum Ltd
2 x 512Mb Mushkin PC3500 Level II
|Hard Drive(s):||Seagate 120GB 7200 RPM (8MB Buffer)|
|Video AGP & IDE Bus Master Drivers:||N/A|
|Video Card(s):||ATI Radeon 9800 PRO 128MB (AGP 8X)|
|Motherboards:||Asus P4C800-E (Intel 875p)
Asus P4S800D-E (SiS 655TX)
Soltek SL-PT880PRO (VIA PT880)
AOpen AK86-L (VIA K8T800)
ECS 755A2 (SiS 755)
Soyo CK8 Dragon Plus (nForce3-150)
The motherboards tested were those available in our lab, but they were also selected to test the availability of PCI lock on various chipsets. The new VIA PT880 chipset claims a PCI lock and the Soltek SL-PT880PRO is the first production PT880 board that we have received. We were very impressed with the AOpen AK86-L for A64, which is the first VIA K8T800 board to show a working PCI/AGP lock in BIOS. The SiS 755 also showed a working PCI/AGP lock on the Reference Board, and the ECS is the first production SiS 755 board that we have received. The Soyo CK8 Dragon is a nForce3-150 board, and all the nF3 boards have claimed working PCI/AGP locks in their BIOS'.
PCI Speed and Overclocking: Test Results
|PCI Bus Overclocking Tests|
|Chipset||PCI Speed at 220 Base FSB Setting||Functional PCI Lock|
Pentium 4 3.2C
Pentium 4 3.2C
Pentium 4 3.2C
Athlon 64 3200+
Athlon 64 3200+
|Soyo CK8 Dragon Plus
Athlon 64 3200+
As you will see in results with the new VIA PT880 board for Intel, early testing raised questions about whether the PCI/AGP lock was working. However, tests with PC Geiger showed the new VIA PT880, as well as the Intel 875P and SiS 655TX, all have a working PCI lock. We have seen outstanding overclocking results with the Intel chipsets and SiS 655TX chipsets; the working PCI lock certainly contributes to that outstanding overclocking performance.
The big surprise here is that none of the current chipsets for Athlon 64 have a working PCI lock, whether they claim one or not. This was a complete surprise and our first inclination was to throw out the results, since speeds on the Athlon 64 are derived in a very different way with Hypertransport. However, oscilloscope results and tests elsewhere pretty much support our findings. There does not appear to be a working PCI lock on any current chipsets for Athlon 64. Certainly, this contributes to the poorer overclocking that we have seen on A64 inasmuch as on-board component frequencies and AGP bus are related to PCI bus.
Please keep in mind that we have only measured PCI bus speed with PC Geiger. While we expect on-board component speeds and AGP to be affected also, we cannot say that with certainty on Athlon 64 systems. Some reports on the web claim that AGP frequency is fixed on both VIA and nVidia chipsets on the Athlon 64. We can neither confirm nor deny these claims with the simple tests that we have performed here. Also keep in mind that results are for 6 individual motherboards. We can generally assume these results apply to the chipset used, but in some cases, this may be a stretch. The ECS 755A2 is designed to be a value board, for example, and 'PCI lock' may simply not have been implemented on this board. On VIA and nVidia chipsets for Athlon 64, our results do mirror those measured objectively at other sites using other VIA and nVidia A64 boards and we are more confident applying these results to chipset assumptions.
We will continue to look at PCI speeds in future tests of new boards and chipsets, particularly those designed for Athlon 64 processors. For now, it certainly appears that a floating PCI bus is one of the factors holding back overclocking on all current Chipsets for the Athlon 64.