The Promised Land?

The Claim

The whole PR campaign for ATI is meant to get you to believe the following:

“[N]obody should have to run any game with any of the features turned off . . . .”

For gaming, that is the Promised Land. It means having your cake and eat it, too. No more compromises.

That’s the notion ATI is trying to put into your heads with all these games and manipulations and suborning of websites, and that the claim that needs to be tested.

The problem is the Radeon 9700 looks more like Moses than Joshua.

Moses got the Israelites out of the Egypt and eventually got them near the Promised Land. Near, not in. His successor Joshua did.

From the numbers we’ve seen so far, the Radeon 9700 gets close to the Promised Land, but not into it. All the fudging is meant to make it look like it does.

The Undeniable ATI Advantage

Like the Parhelia, the Radeon 9700 packs much higher memory bandwidth than previous video cards (including the GF4s). Unlike the Parhelia, the Radeon 9700 uses it effectively.

What this does is essentially remove the memory bandwidth bottleneck that in the past have crippled video cards when using high resolution/antialiasing/anisotropic filtering.

What we saw in the ATI PR exercise was that the vast majority of benchmarks were run at 1600X1200, usually with features like AA and/or anisotropic filtering also turned on.

How Much Does It Help, And When?

This raises two questions:

1) Is the Radeon 9700 so powerful that it can handle any current gaming burden (including items like AA and aniso) at 1600X1200? To be more concrete, can it run most anything at, say, 60-80 fps? If the answer
to that question is “Yes,” then (presuming no massive finagling by ATI) what it does with a lighter load matters little. The previewers were right, and Ed really has been the village idiot about all this.

2) If the answer to that question is “No,” as I’ve said before, do you get the same kind of improvement at lower resolutions that were reported for 1600X1200 at lower resolutions? Again, if the answer to the second question is
“Yes,” then again, Ed’s been stupid.

But if the answer to both question is “No,” then there’s a problem here, and it’s not Ed.

This doesn’t mean if the answer to both questions is “No,” that the card sucks and that no one should buy it.

It does mean this card won’t bring you into the Promised Land, and if that’s what you want and expect, this card won’t do that.

If you’d be happier with somewhat less, you’ll need to find out just what the card can do across a spectrum of resolutions, and then you’ll have to think a bit about whether or not this will do what you want (and it may well).

This involves much more thought than repeating, “This will get me to the Promised Land,” and no doubt will reduce the number of pilgrims.

1600X1200

I think some people believe I have something against 1600X1200 benchmarks. No. I don’t have a problem with seeing 1600X1200 benchmarks. I have a problem with seeing only 1600X1200 benchmarks, especially when I have good reason to believe they’re not representative.

Don’t you find it a little odd that a dozen or so reviewers who up to now have always benchmarked at lower resolutions (and rarely with AA and aniso) all of a sudden change their whole routine and almost unanimously decide that the only benchmarks all you need are 1600X1200 benchmarks with AA and/or aniso? Doesn’t that strike you as being at least a little strange?

Again, to rehash what I’ve said before; I’m estimating that the actual frame rates at 1600X1200 with at least some goodies turned on will often be in the 35-50 fps range.

There will be a few that will do better than that (there’s some indication the Quake engine likes this). If that is all you do is that, then it will be the Promised Land for you. But that’s not everyone.

A Few Good Facts

I looked at the Extremetech review. I found it the best of this bad lot, for a couple reasons.

First, while pretty positive, they at least occasionallly showed some skepticism about some of the items they were being fed.

Second, they did come up with a few benchmarks at 1024X768 with 4X FSAA turned on:

Oh, my. Not exactly a blowout all the time, is it?

Extremetech ran two of these at 1600X1200 with 4X FSAA, too. Here’s where the increased bandwidth comes into play:

Falling Short

Let’s look a little more closely at Comanche 4. What do these percentages get you in actual frame rate?

You can find some Comanche 4 benchmarks using FSAA here.
You’ll find that using an Athlon 2000+, a Ti4600 handles 4X FSAA pretty well at 1024X768X32, only losing about 10% performance.

It comes in at about 37 fps. Improve that 15%, and you get an estimated 43 fps from the Radeon 9700. Better, yes. Enough to replace your GF4? Enough for you even to use that configuration? I doubt it.

Double Nothing Is Still Nothing

Let’s now look at what happens at 1600X1200X32. The Ti4600 doesn’t handle 4X FSAA well at all at that resolution, losing over 50% performance.

At 1600X1200X32, it comes in at only 17 fps. The Radeon does 86% better than that, but 86% better than 17 fps comes out to only 32 fps. Is that good enough for you to get one, or play at that setting? I very much doubt it.

This is the pattern you’re going to see a lot when final benchmarks get run. When FSAA gets turned on, you’ll find the Ti4600 getting low but playable FPS at 1024X768, and the R9700 not doing a whole lot better. Throw it up to 1600X1200, and the Ti4600 goes into the toilet while the R9700 hangs on the rim.

What About UT and Serious Sam and Jedi Knight II?

These do indeed fall into a much different pattern, so much so that Extremetech openly wondered about the huge differences. I’m wondering more than a bit about that, too, and so should you.

Conclusive? Not at all. Worth further investigation? Absolutely.

Actions have consequences. This is the consequence of Quack 3. It’s the old “fool me once, shame on you, fool me twice, shame on me.”

Personally, if there is something bogus here, my bet is that you’re going to find it somewhere in the FSAA/anisotropic filtering section. Perhaps, just perhaps, this card doesn’t FSAA or aniso quite as much as it claims in some of these benchmarks. That’s where I’d look.

It could well be that the Radeon 9700 might just happen to do very well at some things, and if you do those things, fine. Saying that, though, is not the same as saying it’s good at everything.

It does seem like the more memory-bandwidth sensitive a game has demonstrated to be in the past, the more likely it is the Radeon 9700 will do very well against the Ti4600.

Better Doesn’t Mean Good

That’s not the point, though.

It’s not sufficient that the R9700 does a lot better than the Ti4600. It has to do well, period.

It’s like having two cars. One can’t go more than 15 mph, the other can do a max 35 or 40. The second one is much better than the first, but it’s still not good enough for highway driving.

The reality is this card isn’t meaningfully better in many of the situations measured so far. When 25% or 80% or 100% or 150% better than lousy or terrible is still inadequate; then “better” is worthless. A lot of the presented benchmarks are essentially the Ti4600 going at 15 mph, while the R9700 is doing 35 to 40. That doesn’t do you much good on the highway.

Now let’s suppose anybody was trying to sell you the second car, and they gave you a lot of speed comparisons between Car Number One and Car Number Two. Never any MPHs, just percentages. And this person told you that for all different kinds of highway driving, Car Number Two was twice as fast as Car Number One.

What sort of impression would you be likely to get?

If you just went and bought the car, and found yourself doing 35-40 on the highway, what would you think of those who gave you all of those percentages? What would you think of those who essentially agreed with him at the time and said, “He’s absolutely right, isn’t it great?”

If you haven’t seen this before, do you see why this tees me off so?

Better Sometimes Hardly Matters

The converse also applies. If a Radeon 9000 Pro gets 65 fps in a certain situation, the Ti4200 gets 75 fps, a Ti4600 gets 95 fps, and a Radeon 9700 gets 130 fps, which is better? Again, turn the fps into mph and you can see it’s at least very arguable that you don’t need a Ferrari capable of doing 130 mph all day for highway commuting, especially when you look at the price tags.

This is not to say the R9700 may not prove to be a very good idea sometimes for some people. My whole point is that sometimes for some people is not the same as all the time for everyone.

How To Judge When Better Matters

Figure out the configuration you want for the games you play the most. Set the minimum fps you’ll accept for that configuration (60 fps seems to be a good all-purpose figure; adjust to your own personal predilections).

Judge video cards by whether or not they meet your criteria. If the candidates all meet it by a good margin, consider upping the configuration. If that doesn’t work for you, consider price.

If none of the candidates meet it, the test is meaningless so far as you’re concerned, so ignore it. Lower the configuration until some card does meet it.

The only tests that are truly meaningful to you are those where some candidates make it, and some don’t.

What About The Future?

You should judge video cards by what they can do for you now, not down the road. These cards just depreciate too quickly to pay premium prices for something in the future.

Even when “down the road” is pretty precisely defined and not too far away, buying in advance is rarely a good idea.

Let’s say you want the best possible card for Doom III. You have no idea what the best possible card for Doom III is going to be when Doom III comes out, and John Carmack hasn’t told you yet, either.

Sure, Carmack’s said nice things about the R300. He’s said
nice things about the NV30, too. Forget about nVidia, if the Radeon 9700 is so wonderful at .15 micron, what will it be like at .13 micron?

There are other options besides laying out $380 now. You might just wait until the blessed event occurs, and decide then. If this is a matter of a new box, you might want to buy a $100 card now, and spend the rest of the $380 later.

What’s The Matchup?

The way the review sites have presented it, the previews presume that most people out there have $350+ burning a hole in their pockets and they’re demanding to know which is better, the Ti4600 or the Radeon 9700.

Problem is: that’s less than 10% of at least our audience, based on the most recent surveying.

It’s not the situation the vast majority of people find themselves in, and that’s not how they approach it. They don’t just consider framerates; they also money, and they consider time.

If you make people believe the Radeon 9700 is truly special, that it really reaches the Promised Land, far more people will dig deep to get there than if it is just a somewhat better video card. Good for ATI, not so good for those digging deep.

If it’s just a somewhat better video card, then most will come back to their senses and go back to the old mental calculus as to what is best for them.

They’ll go back to thinking like “Now nVidia will really drop the price on GF4s” or, as one person just wrote me “The Ti4200 is better than the 8500, but not $50 better.”

In the case of this card, the thinking will often be “I’ll wait until NV30 comes out and the R9700 drops in price a lot.” However, the competitor in these folks’ minds won’t be the Ti4600; it will be the low-end GF5 card.

Or finally (and I suspect best), “Maybe I ought to wait until this card goes to .13 micron (or even better, get a .13 micron Radeon 9500).”

Why Don’t You Trash nVidia Like This?

Actually, ATI should be honored I’m giving this so much attention.:)

When the Ti4600 came out, the most positive thing I said about it was that it wasn’t a blatant ripoff. My recommendation at the time was to tell people to wait for beefed-up cheap Ti4200s.

At least I can see why the R9700 could be a good idea for some people. I really couldn’t see any good reason why anybody with a healthy regard for money would buy a Ti4600.

Not The Big One For Everyone

This card is not the be-all and end-all for the next few months. It’s not a card everybody should go run out and buy, or even one worth a big stretch of the budget for everyone.

In short, this is not The Big One for everyone. It’s just a somewhat better video card that may or may not be worth it to you.

I wouldn’t be surprised if this proves to be The Big One for some people doing some things, but we don’t know enough to say for certain what they’ll be (though there are certainly a few strong hints).

Just don’t get fooled by anybody who says this is the Promised Land for everyone. Wait and watch and read and think about what is best for you.

What ATI Really Needs Is A $150 Radeon 9500, Now

If I were ATI, this is what I would have done:

Clock the GPU down to 250-275MHz. Put it 3.6-3.8ns dual-bank RAM. Sell it for $150 street price. It probably would outperform any nVidia card short of a Ti4600 and maybe even nudge it out of the way.

That would get OEMs flocking to ATI for Christmas. That would scare the hell out of nVidia and rather wreck their rather comfortable pricing structure.

Most amazing, that might even get me to recommend it, even without more bull, uhhh, marketing.

I’m not really against this card. What I’m really against is the BS surrounding it. Clean it off, and (presuming there isn’t a ton of turd lurking inside the BIOS and drivers), it looks like a fine card. It’s just not the earth, moon and stars, not even the Promised Land.

The tragedy is that ATI doesn’t think good isn’t good enough, better isn’t good enough. Only BS is best.

Ed