Not Quite That Good

We now have some gaming benchmarks on the Opteron.

For the moment, doing so is not an easy task. The video card used had to be a low-end PCI card, and one that AthlonXPs had particular problems with in certain games.

This means you can’t take the results as literally as you might like.

If you walked away from that review thinking that the Opteron is a God-almighty processor and you need an Athlon64 right away, you walked away with the wrong impression.

And we’ll explain why.

What You’ll Get Will Be Different

The Opteron tested ran at 1.8GHz. The chip had 1Mb of cache, and was running dual-channel DDR.

This is not what the average gamer is going to end up with. It’s just too expensive, especially for AMDers’ tastes, and will remain so throughout 2003. Putting 1Mb of cache on a CPU will keep any Hammer-based CPU pretty expensive, and Opteron mobos won’t be cheap, either.

In 2003, what the average Hammer-hellbent gamer will end up with is a “baby” Hammer. This will be an Athlon64 with just 256K cache, probably running at 2GHz. This will run on a socket 754 board, which will be single-channel DDR.

What the difference?

Cache Counts

How much of a difference does a bigger cache make?

Let’s take Unreal Tournament. If you go here, and compare a Thoroughbred running at the same speed as a Barton, you’ll see about an 8% difference in the benchmark.

That’s just going from 256K to 512K. In this situation, it’s a matter of 256K to 1Mb.

Based on earlier data from the old K6 era, you get less of an improvment going from 512 to 1Mb than you do from 256 to 512. Let’s be conservative and assume half.

8% + 4% = 12%. That’s a substantial difference.

Unreal Tournament does appear to be more cache sensitive than other games. If you do the same thing with the other games benchmarked in the same article, you find about a 5% difference. Do the same math that was done with Unreal Tournament, and you get about an 8% performance difference due solely to cache.

Single Vs Dual

We don’t yet have a comparison between Hammer single-channel vs. Hammer dual-channel, but we now have that statistic for at least Unreal Tournament for PIV systems. It comes in at about 4%.

Add 4% to the totals, and we find that a small cache and a single-channel DDR platform with a Hammer will cut gaming performance 12-16% from a Hammer with a big cache and dual-channels.

But 2GHz Is More Than 1.8GHz!

This is true. We need to factor that in, too, before we’re finished.

If we go back to the benchmarks mentioned above, and this time look at the increase in performance you get going from 1.8 to 2.0 GHz, you see that it’s about 5-6%.

Subtract that from our earlier estimate of 12-16%, and you get a figure of 6-11%. Take a figure inbetween, and that’s a little less than the gap shown between the Opteron and the PIV.

But What About Etc, Etc, Etc . . .

There are a lot of unknowns at this point. This is just a rough approximation. Due to the roughness of the approximation, the most one can safely say is that a baby Athlon64 is likely to do a bit worse than this 1.8GHz Opteron.

Then again, one shouldn’t take video tests run with a PCI card too literally, either.

Not Great Opteron, Bad PCI Results

There’s an unfortunate chart which has a few numbers which will be used as factoids by idoitoids.

The numbers for Quake III Arena and Jedi Knight II were included even though even the review called at least the Jedi Knight numbers garbage.

Actually, both sets of numbers are.

While we don’t know how the Opteron performs with a real video card, we do know how a Barton and PIV 3.0C measure up with one. Let’s look at the performance gap bwtween a Barton and a PIV 3.0C, first with a PCI video card, then with a PCI video card, then with a serious video card (also go to the next page) under more typical conditions. Let’s see if that kind of gap exists between the two.

Yes, the Quake test is somewhat different, but run the same test and I’d bet dollars to doughnuts the difference will be more like 15% than 44%

Big difference, isn’t it, and the only difference is the video card.

So most (in the case of Quake) to almost all (in the case of Jedi Knight) of the difference between Opteron and AthlonXP reported is due to a PCI card doing badly with an Athlon, not any huge Opteron superiority

Conclusion

The Athlon64 early adapters are likely to buy next fall will be relatively more competitive in gaming against PIVs than they are in workstation apps. They are likely to be roughly equivalent to 3.0C PIVs.

However, they are unlikely to have the kind of significant lead over overclocked PIVs that these benchmarks imply if taken literally.

Presuming some degree of overclocking 2GHz Athlon64s with 1Mb caches might well be able to have a consistent lead against overclocked PIVs, but the cost of such a system will be high and comparable to PIV systems.

A House of Cards

This review is basically a house of cards, and this analysis is building on top of it. 🙂 That’s primarily due to the circumstances at this point. It’s better than nothing, but . . . .

There’s two intelligent things you can do with these two pieces: