Back in the bad old days, say, nine months ago, people used to have to judge a bit before buying a CPU. After all, they weren’t
disposable diapers, this was something you had to live with for a while.
Now we have a CPU of the Month Club. You don’t have relationships with your processor anymore, you have one-night stands.
I got an Athlon MP 1.2Ghz yesterday. I’m in danger of becoming CPU-infested just trying to keep up with the club. If all the little CPU legs I now have were roach legs, I’d have to call in the exterminator.:)
Let’s see how it does.
We put in a watercooling system into the Master so no CPU would have any heat prostration excuses in any test. So far, quite good; it’s keeping CPU temperature not much warmer than I am.
Will all this support, first thing I did was to go to 1.5Ghz. Bad move.
To make a long story short, this chip doesn’t want to hear 1.5Ghz. Not talking, “unstable in Prime95;” the only booting it did at 1.5Ghz was aimed at my rear end.
The best I could do was 9.5X150, or 1425Mhz. Even at that relatively low speed, I need 1.93V to get into Win98 (Win2K does OK at 1.83V).
In contrast, same setup, no problem at 1.575Ghz with an AYHJA.
You might say I got a runt, but taking a look around the forums, I seem to have a decent amount of company. Between 1.4 and low 1.5s seem to be the range for most of these.
As Joe reported, it does a few C better. Every little bit helps, but it’s not going to turn a lousy cooler into a good one.
1) The myth of memory
With one exception, there’s practically no difference in component measurements. The one glaring difference is in part of Sandra Memory.
In contrast, running this at the same speed with an AYHJA gives you about a 720/920. So there’s a bit more than a 10% improvement in ALU memory bandwidth.
But who cares what Sandra memory says? Do you spend most of your work or play time running Sandra memory? If you do, you really need to get a life. 🙂
A benchmark is only important when it is a useful predictor of what real programs do.
So what I did was slow down the memory as much as I reasonably could. Then I got this.
That’s about an 8% difference in both measurements. It’s enough of a gap to show if programs actually benefit from increased bandwidth or not.
So I ran SysMark 2000 twice, once with “fast RAM” and the other with “slow RAM.” I eliminated any reading that showed a difference of
less than 2%.
1.2Ghz Athlon [email protected]/150Mhz FSB
512Mb Crucial PC2100
MSI K7Master 1.1, bios 1.2b2
45Gb IBM 75GXP
Matrox G450 32Mb Dualhead video card
1024X768, 16-bit Color
What kind of difference did I get?
Only one program significantly benefited by increased memory bandwidth. The majority of programs showed negligible
This type of pattern is usually what you get when you make a system change. One program tends to be helped a lot, a few a little,
most not at all. For instance, if you go from 256Mb to 512Mb, Corel Draw is again the big winner from that change.
That’s the problem with the one-stop shopping benchmark combos. They aren’t just useless, they’re misleading. If you heavily used
Corel Draw, you’d never know from the averages how much improvement you’d get from increasing RAM or increasing bandwidth.
On the other hand, if you use
Photoshop instead, you might think you’d get some improvement, and get diddly-squat instead. You can’t even assume that all ‘
programs of a certain type will behave more or less the same after a system change.
(Granted, Sysmark2000 uses relatively
small graphics file, if you work with 200Mb files all the time, any program would do a lot better with more RAM).
You might say, “but games benefit more.” Some do, some don’t. If you want to look at some gaming benchmarks, go here (no point reinventing the wheel).
2. SSE For You and Me
How does the MP fare against a TBird? Take a look yourself (same setup and rules as above)
Photoshop is a huge winner here, with Premiere and Windows Media Encoder racking up solid gains (but notice Corel doesn’t benefit at all). Again, most apps got little to no improvement; PowerPoint actually went backwards a bit.
The MP supports SSE optimization. The TBird does not. Photoshop is heavily SSE-optimized (and TBirds up to now have handled Photoshop relatively poorly).
So if you use applications and games with heavy SSE optimizations, this Palomino stuff is going to be a great bonus to you. If you hear about a program doing a lot better with an Athlon MP than a TBird, this will probably be the reason why.
Of course, your motherboard has to know that. With the latest beta bios, the Master does, but most others don’t at the moment. If your mobo doesn’t know it can do SSE, it won’t, and you won’t.
Why Buy Now?
Those who belong to the CPU of the Month Club are going to buy anyway.
The only other group with a good reason to buy now are those who use heavily optimized SSE applications are understandably tempted. However, if SSE does such a great job at a little above 1.4Ghz, it can only get better at a higher speed. It looks like we’re only talking about
a few weeks before we see a 1.4Ghz MP out.
Outside of those two groups, I don’t see a good reason for anybody else buying what’s available now.
Yes, it’s capable of SMP, but unless you want to spend over $400 for the Tyan, most of you aren’t going to be doing any SMP for a while.
The limited overclockability reported so far is a big negative. The desktop Palominos running at stock will be as fast or faster than what most people are getting now.
A 1.4Ghz MP should be out in a few weeks time. If I had to bet right now, I would bet that it will do better than what we have right now. So should the desktop Palominos.