Intel 925X Roundup: Creative Engineering 101
by Wesley Fink on August 12, 2004 12:05 AM EST- Posted in
- Motherboards
Foxconn 925A01: Overclocking and Stress Testing
FSB Overclocking Results
| Front Side Bus Overclocking Testbed | ||
| Processor: | Pentium 4 Prescott LGA 775 2.8GHz |
|
| CPU Voltage: | 1.3875V (default) | |
| Cooling: | Thermaltake Jungle 502 | |
| Power Supply: | HiPro 470W | |
| Maximum OC: | 218FSB (+9%) - 100 PCIe or Auto 248FSB (+24%) - 114 PCIe |
|
The LGA 775 processors are multiplier locked, so the only way to overclock the CPU is to increase the FSB. Most of the manufacturers in this roundup float the PCI Express bus and manipulate ratios to achieve higher overclocks. Foxconn takes a different approach by providing a wide adjustment range of fixed PCIe frequencies only. There is no provision to float the PCIe bus.
We were able to confirm the limitations of this approach by reaching a rather dismal 9% overclock with the PCIe fixed at 100. However, we had learned on the DFI and Abit that the nVidia could tolerate a PCIe frequency of at least 114 with simple algorithms or perhaps as high as 120 with more exotic manipulations. So, we set the PCIe to a fixed 114 and started increasing the CPU Clock. With a 114 PCIe, we reached a 248 overclock or 24%. We could not reach any higher, however, with either the ATI X800 XT or the nVidia. Apparently, there is more to the overclocks that others are achieving than just a tales of forced PCIe:CPU ratios. Considering the simplicity of the Foxconn design, however, this is a really excellent overclock that will satisfy many potential 925X shoppers. Just keep in mind that you really have to play with the very sensitive ratios between CPU and PCIe to achieve this kind of overclock on the Foxconn 925x-A01.
Memory Stress Test Results:
The memory stress test measures the ability of the Foxconn 925A01 to operate at its officially supported memory frequency (533MHz DDR2), at the best performing memory timings that Crucial/Micron PC2-4300U will support. Memory stress testing was conducted by running DDR2 at 533MHz (stock 3:4 ratio) with 2 DIMM slots operating in Dual-Channel mode.| Stable DDR533 Timings - 2 DIMMs (2/4 DIMMs - 1 Dual-Channel Bank) |
|
| Clock Speed: | 266MHz |
| Timing Mode: | 3:4 (200:266 - Default) |
| CAS Latency: | 3.0 |
| Bank Interleave: | Auto |
| RAS to CAS Delay: | 3 |
| RAS Precharge: | 3 |
| Cycle Time (tRAS): | 10* |
| Command Rate: | N/A |
The Foxconn matched the other boards in the 925X roundup in being able to run with complete stability at 3-3-3-10 timings with two DDR2 DIMMs. The 925A01 was completely stable at these timings at the default 1.8V.
Filling all four available memory slots is more strenuous on the memory sub-system than testing 2 DDR2 modules on a motherboard.
| Stable DDR533 Timings - 4 DIMMs (4/4 DIMMs - 2 Dual-Channel Banks) |
|
| Clock Speed: | 266MHz |
| Timing Mode: | 3:4 (200:266 - Default) |
| CAS Latency: | 4.0 |
| Bank Interleave: | Auto |
| RAS to CAS Delay: | 4 |
| RAS Precharge: | 3 |
| Cycle Time (tRAS): | 10 |
| Command Rate: | N/A |
When all 4 DDR2 slots are filled, the Foxconn required slightly slower timings than the top boards in the roundup, requiring 4-4-3 timings for complete stability instead of the 4-3-3, which worked well on the Abit, Asus and DFI boards. This is only very slightly slower than the timings on the top 3 boards, and could just as well represent variations in board samples as much as any differences in memory performance with 4 DIMMs.
Facebook
Twitter
RSS
30 Comments
View All Comments
johnsonx - Thursday, August 12, 2004 - link
What is it with you people griping about CPU choices? This is a review of current top-end 925X boards, not a CPU review! The FX-53 scores are there only for a point of reference. Added to that, Wesley's point is VERY valid: the 560 and FX-53 ARE the top CPU's from each camp.If you really want to know how a 3800+ would perform, refer to past Socket-939 reviews, or just mentally subtract about 3% or so.
STOP WHINING!
Wesley Fink - Thursday, August 12, 2004 - link
#17 - Since we were trying to determine the maximum overclocking ability of the boards tested, we used a 3.6 ES LGA 775 Prescott at a 14 multiplier (2.8Ghz). The 14x280 is close to 3.9GHz speed. We also checked with a retail 540 (3.2GHz) and reached 250FSB (4.0GHz) at 1.45V.These results lead us to believe that many 775 Prescotts will top out at 3.9 to 4.0GHz on boards that will support those overclock levels. That means that there are likely some 2.8 Prescotts out there that can reach 280FSB.
As always, overclocking is variable, and you need a really great power supply and decent cooling to support the power requirements at these kinds of overclocks.
Carfax - Thursday, August 12, 2004 - link
Wesley, is it possible to do a review of Prescott which focuses on the upcoming 1ghz FSB? I've heard that Prescott scales better than N.W with a higher FSB and greater clockspeed..To do the review correctly, you'd need an engineering sample with an unlocked multiplier, so you can see the benefit of the increased FSB, without raising the clockspeed.
I think Prescott would do pretty well on 1066FSB and with fast DDR2 memory..
danidentity - Thursday, August 12, 2004 - link
Wes,When you say you hit 280 FSB with the Asus P5AD2, was that with a retail chip, multiplier locked? Or were you using an ES chip. If you were using a retail, that is an absolutely insane overclock.
danidentity - Thursday, August 12, 2004 - link
>> Better than comparing a 3500+ to a 3.6F anyway :PHow would a 3500+ compare with a Intel 3.6? Could it hang? :)
RyanVM - Thursday, August 12, 2004 - link
I have no problem with the 3.6E and FX53 being shown together since both platforms will end up costing about the same (factoring in CPU, mobo, and memory costs). Prices fluctuate, yes, but both companies (OK, mainly AMD) tend to adjust prices to stay in line with performance levels (if Intel drops the 3.6E price, I'd put money on AMD dropping prices at the high end within a day or two).Better than comparing a 3500+ to a 3.6F anyway :P
Creig - Thursday, August 12, 2004 - link
#12/#13 Given the way pricing can fluctuate, it would be futile to compare Intel $$$ to AMD $$$. A couple of days after the article was published, pricing could change to make the monetary comparison useless and therefore misleading.I think they're doing it the correct way. It's up to the end user to find his/her best balance between performance and price.
mjz5 - Thursday, August 12, 2004 - link
man, i should of read #12 first before posting it.. why not have an edit button?anyhow, u all know what i'm saying!!!
mjz5 - Thursday, August 12, 2004 - link
the way i see it is that CPUs should be compared by price. If an AMD FX-53 cost as much as a Celeron 2.4 GHz, why not compare the two? If someone is going to looking at these products because they cost X dollars, they aren't interested in seeing that an Intel CPU that cost (X*2) may or not surpass it the competitor at only X dollars.Wesley Fink - Thursday, August 12, 2004 - link
#9 & #10 - Corrected