Looking At Socket 754

For the past few weeks, I’ve been working with a watercooled socket 754 system consisting of an Athlon 64 3200 (2GHz, 1Mb cache) and an Asus K8V Deluxe motherboard. It was coupled with two 256Mb sticks of Corsair XMS 3500 RAM and a Radeon 9700 Pro.

In general, the impressions were positive.

Motherboard Impressions

There has been a problem with inferior capacitors with some shipped boards. We’ll talk more about that on page three of the article, but the board reviewed did not have any of the capacitors in questions.

If you want to see a list of what the mobo contains, go here. It has all the usual frills boards have these days, including a gigabit Ethernet connection. It also has two RAID controllers, which isn’t.

Per motherboard design, the floppy connector is located towards the bottom of the motherboard, which might make for a stretched connection. What may be more inconvenient for some is the location of the Clear CMOS connector; it’s a bit to the right and below the second PCI slot. It’s not very hard to get to, but if you’re working in a cramped case with a big blower on your video card, it could be awkward to get to.

Otherwise, mobo design is pretty good. There is good clearance for big coolers around the CPU, no awkward capacitors in awkward places.

BIOS controls were OK, though often worded rather clumsily.

For instance, “CPU Speed” really means “Multiplier.” (You can set the multiplier downward, but not above the default multiplier.) Instead of seeing “4X, 5X, 6X,” you see “800MHz, 1000MHz, 1200MHz.”

Setting memory speed can be a bit confusing, also. If you want your RAM to run at the same speed as your FSB, you need to set your memory to a 2:1 (DDR 300) setting, not 1:1.

Operational Impressions

Installation was routine.

After using the system for a few weeks, I have just one thing to say about it. It just works. I liked that very much, and so should you. I don’t recall a single functional problem or even hiccup during that time. Since this might come as a surprise to those with earlier experiences with Via, I’d thought I’d mention it. I would not hesitate to install this system for some Joe Sixpack.

Overclocking Impressions

The big theoretical overclocking issue with this or any other Via-based board (it’s also the case with nForce 150 boards) is the PCI divisor issue. Via continues to provide no means to lock the PCI/AGP speeds at 33/66MHz.

That means when you increase the FSB to overclock the processor, you end up also overclocking the PCI and AGP bus. If you run this system at 240MHz, for instance, you also end up increasing the speed of the PCI bus from 33MHz to 40MHz, and the AGP bus from 66MHz to 80MHz. This can cause video cards and (more often) PCI-based components (including hard drives) to stop functioning and thus limit your overclock.

Given that, when we went to overclock, we tried to minimize the likelihood that this issue would come up by using a parallel rather than SATA hard drive, and not using any PCI/AGP devices other than the (unoverclocked) video card.

Since the Corsair RAM used was only stable up to 220MHz in PIV systems, we set the FSB:memory ratio to 5:4 to keep memory from becoming a bottleneck.

We used the latest version of Prime95 to test for stability.

The best we could do and still retain solid stability was this:

We could not get Prime to run stably at 2300MHz even with 1.675V, while 2250MHz ran fine with 1.575V. However, for this particular setup, PCI/AGP was not an issue, there was no problem booting into Windows and accessing files from the hard drive at 230MHz.

Unfortunately, given the multiplier lock of the Athlon 64, increasing FSB is going to be the only way to overclock this CPU, and you really can’t expect to count on boosting the FSB to 240MHz and more without running into problems, so this does effectively limit the overclocking of this processor. Probably not a big deal for the current crop of A64s, but it could be for later ones.

Benchmarks?

There’s no point in reinventing the wheel. This CPU has been benchmarked to death, and our results (for those benchmarks not affected by video card) are just a tiny bit better than those recorded for an Athlon 3400+. A very good benchmarking summary of the leading-edge (or in the case of Prescott, heating-edge) processors can be found here, and the comparisons made there among the front-runners are a good deal more comprehensive than any we could make.

Besides, I like Xbit Labs. I think they’re the best general hardware site on the Web. 🙂

Besides, the most important for socket 754 is not whether it does 3% better than a Northwood or 2% worse than a FX-51.

Ed

Technologically Incorrect

When you work with this board, you can see why AMD thought it could get away with just one memory channel for its next-generation Hammers. The Athlon64 delivers as much punch (or a bit more) than an overclocked Northwood system for about the same price, and unlike the Northwood, has at least the potential of punching rather harder with x86-64.

But then again, this system as configured costs as much if not a bit more than that Northwood system. It ought to be this good.

In the long run, though, especially for gamers, you’d have to expect single-channel memory to become a bigger anchor on performance.

In the long run, though, if AMD is going to essentially abandon leading edge technological progress on socket 754, that’s another big negative.

That’s what makes socket 754 technologically incorrect.

Now under normal circumstances, the answer to this problem is simple. Wait for socket 939.

However, if AMD holds to its current intentions, a socket 939 system is likely to cost $200 more than a socket 754 system. Even assuming socket 939 will squeeze out 5-10% more performance than the current socket 940 (and 10% is probably too high), $200 for what will probably be about 10% extra performance hardly qualifies as great bang for the buck.

That would make socket 939 technologically correct, but financially very incorrect.

The biggest underlying factor, though, especially for those who can’t upgrade at the drop of a hat, is that the perfomance difference between Athlon 64 and XP systems is still generally in the 20-30% range, and people with fairly current XP systems generally aren’t hurting by any means. If an expensive socket 754 system today is good, a (probably) less expensive faster 90nm socket 939 next year will be much better.

And they’re in no rush.

So What To Do?

A long, long time ago, the comedian Jack Benny (who, if you don’t know, generally played a very cheap person) had a routine in which a robber came up to him and said, “Your money or your life!” and he would just ponder the question.

That’s sort of how I feel about choosing a system now. If you put a gun to my head and said, “Your recommendation or your life!” I’d probably do some Jack Benny-like pondering and when the trigger started getting squeezed, say, “Go with the socket 754.”

What you should do really depends on whether you’re under the gun or not.

If you need a new system NOW, the best choice you can make is this. Socket 754 will generally let you slightly match or beat overclocked Intel Northwood systems, will certainly give you less grief and concern that any Prescott system, and eventually will give you the chance of running x86-64.

If you can wait a few months, you may end up with a few improvements . . .