Overclocking Buyer's Guide - August 2004
by Wesley Fink on July 27, 2004 11:24 AM EST- Posted in
- Guides
CPU and Motherboard: PERFORMANCE OC Alternatives
CPU: Intel Pentium 4 540 (3.2Ghz)Motherboard: Asus P5AD2 Premium (Intel 925X)
Price: CPU - $288 shipped (Retail with HSF). Motherboard - $279 shipped
It was difficult to decide between the Single-Channel Socket 754 Athlon 64 and the new Intel Socket 775 for the Performance Overclocking alternative. Despite the massive increase in memory bandwidth that comes with the Dual-Channel 939, the fact remains that the Athlon 64 is not particularly bandwidth-starved compared to recent deep-pipe Intel designs. As you can see in the launch article on Socket 939, this translates into performance of the single-channel 754 Athlon 64 that is only 2% to 9% slower than the new 939 at the same speed. In real terms, this means that a Socket 754 combined with a second-generation Socket 754 motherboard having AGP/PCI locks is still outperforming Intel's latest and greatest Pentium 4 560 (3.6GHz). The new pricing of the 2.4GHz 3700+ at about $525 also makes this tempting for the Performance overclocker. This is still a very good alternative, but in the end, the new Socket 775 on one of the motherboards that solves the Intel OC lock riddle won out as the OC alternative for features and future.
Intel produces processors with locked multipliers, so there is no real advantage for an overclocker to choose a top Intel chip other than the higher stock speed. Since most Socket 775 processors that we have tested tend to top out in the 3.8GHz to 4.0GHz range with modest voltage increases, the task of choosing a 775 CPU is one of balancing ultimate overclock with board Clock frequency capabilities. In this case, the Asus P5AD2 tops out at 278 with ATI PCIe video or about 260 with nVidia PCIe. We also showed, in our DDR2 memory roundup, that all current DDR2 memory could run at DDR2 667, which is achieved at a Clock frequency of 250. If, for margin, we choose 250 as our clock goal, we need a CPU with a multiplier of 16 to reach 4.0Ghz. This makes the 3.2E Socket 775 an ideal match to a Performance overclocking system.
If you need to save a bit of money, the 3.0E or 2.8E 775 is also a great choice for overclocking. Keep in mind that a 14 multiplier will require an unreachable 285 for 4.0GHz, making the 3.0E a better choice of the two for performance overclocking. If you could reach a CPU clock of 285, then the memory becomes an issue since DDR2 becomes "if-y" above DDR2 700. At 285 the DDR2 memory speed at stock 4:3 ratio is DDR2 760.
When Socket 775 was launched a few weeks ago, it did not appear that any of the new 925X/915 motherboards would ever become a recommended Overclockers board. The 925x/915 boards appeared limited to about a 10% overclock by design. However, several manufacturers found effective ways around the overclocking issues, and the 925X boards began to open up.
The best overclocker that we have seen so far among the 925X motherboards is the Asus P5AD2 Premium, which has already shown the ability to reach 278 FSB with the difficult combination of a top-end ATI X800 XT PCIe video card and a SATA hard drive. The P5AD2 was designed from the start to be a dream board for Performance overclocking. It is the only board that we have tested with a DDR2 memory option of 600MHz at stock speed in addition to the common 533 and 400 options. Asus also provides a full range of AI overclock options for the more timid, with auto presets from 5% to 30% overclock, which represent a range of auto overclocks to FSB1066/DDR2-710.
Serious overclockers will want to dial their own, and the Asus offers a full range of both familiar and new manual settings. PCI Express can be adjusted from 90 to 133, but we have found that the best performance is with Auto, which adjusts PCI dynamically at boot for best overclock. There are also adjustments to fix or float the PCI clock. CPU frequency is adjustable from 100 to 400, CPU voltages can be tweaked from default (1.3875V) to 1.7000V in very fine .0125V increments, and memory voltage is adjustable from 1.8V to 2.1V. There are even very useful adjustments for Chipset Core voltage and FSB Termination voltage.
While we could wish a wider range of Memory voltage adjustments, the Asus P5AD2 delivers on overclocking when many other 925X boards struggle with PCIe and SATA overclocking issues. The P5AD2 is an expensive motherboard, but it will take 775 overclocking to places that are difficult to reach with other 925X motherboards.
Listed below is part of our RealTime pricing engine, which lists the lowest prices available on Intel CPUs from many different reputable vendors:
If you cannot find the lowest prices on the products that we've recommended on this page, it's because we don't list some of them in our RealTime pricing engine. Until we do, we suggest that you do an independent search online at the various vendors' web sites. Just pick and choose where you want to buy your products by looking for a vendor located under the "Vendor" heading.
Facebook
Twitter
RSS
40 Comments
View All Comments
vaeren - Thursday, July 29, 2004 - link
Well I decided to try the EVGA card for 389 from www.buyxg.com Went to the site, lo and behold the card is stated there at 389. I go to buy it and it rings up 409. Ok, so I see it's backordered and I send them an e-mail basically asking them to honor the homepage. They essentially tell me that I don't know what I'm talking about and the page is now fixed. I don't think I'd recommend them in the future as an accurate price guideline. Granted Anandtech doesn't guarentee prices, but I think poor business practices should be addressed.PrinceGaz - Thursday, July 29, 2004 - link
Just a slight error in the article regarding the Mobile XP 2500+"However, the FSB speeds and multiplier are also a fortunate accident on the 2500+, since 11X is also the same multiplier as the top 3200+ Athlon XP. Since the 2500+ runs at 166 FSB, 3200+ performance is often as easy as setting the FSB to 200 from the default 166. Most, but not all, 2500+ can easily reach 3200+ speeds. The 2600+ mobile also looks even more promising as an overclocker, though we have not yet tested it. With specifications of the same 45 watt power consumption, a 12 multiplier, and 166FSB, it is hard not to be tempted when it is less than $10 more than a 2500+."
The desktop version of the 2500+ does run at 166*11 (1833 MHz), but the Mobile version has a 133FSB default. Thats not a problem because as you say in the article, they are unlocked so you can set the FSB and multiplier to whatever you like. But you will need to lower the multiplier with the Mobile 2500+ to run it as a 3200+.
The Mobile 2600+ is probably a much better choice because it has a much higher default speed than you'd expect,
Desktop 2500+ 166*11 = 1833 MHz
Desktop 2600+ 166*11.5 = 1917 MHz
Mobile 2500+ 133*14 = 1867 MHz (1.45V)
Mobile 2600+ 133*15 = 2000 MHz (1.45V)
the Mobile versions of both chips have a slightly higher clock speed than the desktop versions to compensate for the lower default FSB. In particular the Mobile 2600+ is a whole 133MHz faster than the Mobile 2500+ to compensate for the ever diminishing returns of higher multipliers.
For an overclocker that means the Mobile 2600+ is an unlocked chip that is guaranteed to run easily at 2GHz while still at 1.45V. That pretty much guarantees it will reach 2.3GHz, and maybe as much as 2.5GHz at the desktop 1.65V. And you can always give it a little extra juice if that isn't enough :)
roostercrows - Thursday, July 29, 2004 - link
#2 bluedarti did some research about a month ago on the synthetic diamond heatsinks and it seems they have been used for many decades (40+ years)."Swans" research was extremely helpful. i would like to know what material you are coating the diamonds on and also if you are infiltrating the diamonds with copper etc,(if you are makeing one you know what i mean). i have one ordered using tungsten and 6%cobalt as the base metal, 1mm to 1.5mm thick. i'm glad to hear someone else is playing around with this. btw, the specs that i looked at were 7x better than silver, using single crystal. also, what size are you using and are you using an existing product or are you starting from scratch? have fun.
Pathogen03 - Thursday, July 29, 2004 - link
I like the writting style and thought processes behind the review.. It definetly was the most fun, and most detailed one ive seen.BUT,
Your Recommendations are shaky at best. When you argued the 2500+ over the 2600+ for the mobile chip, it made sense, but the inclusion of the FX-53 over an Athlon64 just completely eludes me.. Id suggest you talk to some people on the forums before you do your next one, to doublecheck you have all of the current overclocking trends down. Oh, and a true overclocker would never even CONSIDER a Speeze heatsink.. If you want to talk about the article I can try to give you my opinion, you have my name in the forums just PM me.
TrogdorJW - Wednesday, July 28, 2004 - link
I have to say that this is by far the best of ANY of the guides that you and Evan have put together for Anandtech, Wesley. Perhaps it's the inclusion of four system options (with prices for all four, even!), or maybe it's just that I'm an overclocker at heart? Anyway, I'm not too sure about the FX-53 on the performance end, as it's just way too expensive, but that's about my only complaint. You mention the 3500+, and for the price, I would think that's the way to go.I have to agree with the suggestion of just making suggestions for the five basic platforms, though. I mean, you already have done that, with the exception of the socket 754 system. Add in a motherboard (MSI K8N Neo Platinum?) and CPU (3200+ 1 MB?), and you're pretty much done.
Anyway, great job! I'm seriously looking at a socket 754 overclocking setup in the near future, unless someone can convince me that the extra $120 for the 3500+ over the 3200+ (not to mention the motherboard probably costing ~$40 more for S939) is worthwhile. Any takers, or should I just go with a 3200+? I won't be buying for at least another month, Wesley, so you can address my needs in your next OC guide! :)
Wesley Fink - Wednesday, July 28, 2004 - link
MSI just sent the following reply to our request for information about when the K8N Neo2 would be available for purchase by our readers:"I am sorry for this kind of situation and inconvenience. My first shipment of K8N Neo2 will be this coming Tue. (Aug 3rd). Most on-line etailers (newegg, ziproomfly, .......) will post the board on Wed. or Thu to allow customers to buy it.
We will also modify the product page from "available soon" to "first week of AUG." http://www.msicomputer.com/product/p_spec.asp?mode..."
Parc - Wednesday, July 28, 2004 - link
Great article as always. These are my favorite because it is exactly what I look to your website for the best of the hardware out. Key word out. The MSI board is not out and now had been delayed to somewhere around August 9. By August 9 the board will probably be delayed again. Why do you keep putting a board up that no one can get? The 6800s are hard to get but can be gotten . This just a suggestion. Still great article except board.Wesley Fink - Wednesday, July 28, 2004 - link
The Coolermaster CM stacker is not an all-Aluminum case, but a mixture of 1.0 mm steel plate for strength and Aluminum alloy. The description has been corrected in the article.Anemone - Wednesday, July 28, 2004 - link
Agreed Shinei...I'm looking at the 3700 EB (3-3-2-8) and then at the new 4000 Rev 2 (2.5-3-3-?) and curious how they compare. Had a chance to look over the new 4000 Rev 2 yet Wesley?
Loved this article totally - as well as the 939 reviews. OC'ing on the new 925X boards has a host of connected issues with pci-e and sata, so bye bye Intel...
Another great AT article for sure
Shinei - Wednesday, July 28, 2004 - link
T8000 has apparently never read any of the Socket 939 benchmarks AT did, or he'd know that the AFX's stomp out Intel at media encoding (along with just about everything else). Anyway, moving on from Trollville.I agree about the 3500+ being the better price/performance part over the FX53. Of course, I don't own one, so I can't comment on performance, but based on AT's results, it looks like a promising component...