Fast Memory, But

A little birdie told me that we should be seeing some new improved memory coming along from at least one of the more established names in memory.

This is good to know because memory is going to be a PITA the rest of this year, pretty much no matter what you do.

On the AMD side, the matter is fairly straightforward. Higher FSBs help Athlon performance more than with the PIV; it enables greater memory bandwidth.

So the strategy for an Athlon is to get the FSB as high up as you can, with a memory ratio of 1:1. Running memory at a higher speed than the FSB does very little good because of the lower bandwidth of the Athlon platform.

PIVs are another story. Matters are starting to get complicated, and will get more so.

Up to now, the general strategy has been to run memory at a higher speed than the FSB, because the bandwidth of the PIV FSB can accommodate the extra memory throughput.

However, this strategy is starting to fall apart now, and will collapse with the 800MHz FSB CPUs and mobos next month.


What is beginning to happen with the latest PIV chips is that memory is becoming the bottleneck.

When people are going over 166MHz with these CPUs, and keeping a 3:4 or 4:5 FSB:Memory ratio, the memory is actually running at well over 200MHz, and it starts crapping out. Push it far enough, and even the best memory available is going to crap out.

What is disturbing about this is that many of those who are running into this bottleneck are either clueless that this is their problem, or think that memory bandwidth is the ultimate performance measurement. I’m seeing people who think that memory bandwidth is even more important than CPU MHz.

It isn’t (outside of extreme situations, which we don’t have here).

In real life, under normal circumstances, increasing the CPU MHzage by X% increases performance far, far more than increasing memory bandwidth by the same X%.

As a very rough guideline, increasing the CPU MHz X% will get you an overall performance benefit yield of about 60%. So a 10% increase in the CPU speed will get you an average 6% performance increase.

In comparison, increasing memory bandwidth X% will get you a overall performance benefit yield of about 10%. So a 10% increase in memory bandwidth will get you an average 1% performance increase.

Big difference, isn’t it?

Increasing CPU speed is a major contributor to performance. Increasing memory bandwidth is a minor contributor to performance.

So if you have to choose between somewhat more memory bandwidth or somewhat more CPU MHzage, choose the latter. Period.

Should you find yourself in this situation and hit a wall while running memory faster than FSB, try running at a 1:1 ratio, and see how you do. You may well be better off taking that route in real life and applications even if your Sandra memory benchmark score suffers for it.

To Come

With April and May comes the 200/800MHz FSB standard, in both the CPUs and the mobos.

There is one unanswered question which will determine how suitable these CPUs will be for overclocking: What memory ratios will be available for the Canterwood/Springfield mobos?

Forget trying to overclock a 200MHz CPU with memory running faster than the FSB. For even a half-decent overclock, that would require memory running at around 300MHz. Current memory can’t do it, and it’s pretty unlikely any DDR-I memory will ever be able to reach those levels.

Running at a 1:1 FSB:Memory ratio isn’t going to be too good, either. At least initially, the somewhat affordable 200MHz FSB PIVs will be the 2.4 and 2.6GHz chip. Even assuming a generous 250MHz FSB/memory speed, that only gets you 3GHz and 3.25GHz respectively, and the latest multvid CPUs look to be a bit more capable than that.

There have been rumors that at least some of the new Intel boards will allow you to run FSB faster than memory, hints are we’ll see at least a 6:5 ratio.

Should this be true, this would be much better for overclocking. You could, for instance, run an FSB of 270MHz, and a memory speed of 225MHz (remember, these are dual DDR boards, so you’ll get a big memory bandwidth boost over current single DDR boards anyway). That should get you decent to good overclocking room for at least the 2.6, plus memory bandwidth to burn.

Of course, we may find out that the boards can’t handle such high FSBs, but if they do, that will become the strategy for such CPUs.

Teaching Old Dogs New Tricks

Even if we do get those ratios in those mobos, I have doubts as to how popular the 200MHz FSB CPUs are going to be.

It may turn out that hyperthreading isn’t terribly worth having, and 133MHzs CPU might offer slightly better performance than the 200s using the old formulas. That’s going to require a little experimenting to find out. Should that prove to be the case, fine.

But if that’s not the case, I wonder whether some old dogs are willing to learn new tricks. There often seems to be a rigidity to a lot of people’s thinking out there. What’s really disturbing is that this website isn’t exactly a senior citizen’s center, a lot of these “old dogs” are chronologically young pups. Geez, if you’re not adaptable and willing to learn at 20 or 25, what will you be like at 60 or 65?

I really wonder if some people will skip the jumps and hoops of overclocking 200MHz processors simply because it’s just too hard for them to figure out.

Take memory benchmarking. In the last few years, this has gone from being a nothing benchmark to one that many consider to be as important or even more important than CPU speed. And it isn’t, but people just won’t believe that.

It’s almost like only one factor, one rule can fit into some people’s heads.

Computer performance, while not horribly complicated at least at the user level, can’t be boiled down to just two or even one rule. Even at the simplest level, there are multiple interrelated factors, and if you choose just one of them, you throw the baby out with the bathwater.

Well, sorry, but the rules are about to change. What worked yesterday may not work today and it won’t work tomorrow.

And tomorrow won’t be as simple (at least not for a while) as yesterday, and if you can’t accept that, you’re old before your time.

No matter what though, a new bleeding-edge system is going to call for very, very fast memory, so you might as well wait until you buy (especially for a PIV system) to get the latest and greatest.

But then, if price takes precedence over performance, giving up a little on the memory isn’t going to kill you in real life.

Email Ed

Be the first to comment

Leave a Reply