AMD's Barton is here (Athlon XP with 512KB cache, 3000+ rating)

[Wraith]

--"The Server Line won't help AMD much, it'll help some."

Sure it will. The margins are much higher in this area. Any significant marketshare gain in this area will help AMD's profits quite a bit, while eroding one of Intel's highest margin lines.

--"But I remember hearing about how much Athlon MP was going to do for the company"

It probably would have done better if AMD had stayed in the chipset business. Still, it gave them a bit of a foothold in the market. Newisys has already announced Opteron systems, and it's not at all unlikely that IBM will be selling them as well. If a Tier 1 vendor like IBM does support the line, AMD will gain marketshare quickly.

--"With a couple of registry entry changes, you can turn off the pagefile and it'll keep all of the OS files in the RAM."

Does that work with 2k as well? I'd love to force the bugger to stop paging stuff to disk when it's got plenty of RAM open.

--"the lack of SSE2 in the Athlon XP isn't really an excuse for the lack of performance."

You're a programmer, you should know there's a whole lot more to the driver optimizations than just SSE2. At present, most Quadros are sold with P4-based workstations, so it's only sense to focus the optimizations on it. This is not true of their GeForce line (which likely does have any SSE2 optimizations they can use in the P4 data path, but also has the appropriate Athlon optimizations).

Wraith
"If it jams, force it. If it breaks, it needed replacing anyway."

[Urban Ranger]

It might also be possible that AMD delays the Clawhammer to squeeze more life for the Athlon.

[Asher]

Originally posted by Wraith
Does that work with 2k as well? I'd love to force the bugger to stop paging stuff to disk when it's got plenty of RAM open.

Yeah, it does. It works on Windows 2000 and XP only.

I've been told by some people that you don't even need to do a registry tweak to do it, just select "No Page File" on the memory management screen and it should handle itself.

You're a programmer, you should know there's a whole lot more to the driver optimizations than just SSE2. At present, most Quadros are sold with P4-based workstations, so it's only sense to focus the optimizations on it. This is not true of their GeForce line (which likely does have any SSE2 optimizations they can use in the P4 data path, but also has the appropriate Athlon optimizations).

Yes, there's a whole ton of stuff they can optimize on the Pentium 4 in addition to SSE2, but the largest benefit to performance in professional 3D applications is made possible with SSE2. Isn't the Specviewperf benchmarks dealing with 64-bit floats? That'd definitely explain why the P4 would do better -- SSE2 supports 2 x 64-bit floats, Athlon doesn't have SSE2 and can only do 1 64-bit float.

[Rasbelin]

Urban,

well, I assume you could be right as they have to delay the Athlon 64, if they want to utilise the new Barton core more effectively, because it lifespan would be quite short if the Athlon 64 would be released soon.

Asher,

I assume you're right about that chipset issue.

[Wraith]

--"I've been told by some people that you don't even need to do a registry tweak to do it"

I've heard a lot about problems with this method, which is why I haven't done it. Oh well, more research I guess. Maybe I'll just pick a sacrificial box and give it a try.

--"Isn't the Specviewperf benchmarks dealing with 64-bit floats?"

I don't know the details of the program, so I'm not going to be able to answer that. However, this is a workstation card. I don't see why their driver optimizations would give up accuracy for speed. That's the kind of thing you do for a gaming card. I'd expect they'd stick with the 80 bit x87 as much as possible.

And that, again, is not all of the story. The P4 requires a completely different set of optimizations. Some of the operations the P3 and the Athlon are good at suck donkey nuts on the P4, and vice versa. If they ran the drivers with all the P4 optimizations and nothing else there'd still be a major speed difference without SSE2.

--"because it lifespan would be quite short if the Athlon 64 would be released soon."

This could be part of it. Intel actually doesn't have a whole lot of speed increases on their immediate roadmap. The P4 is hitting the wall on the .13 process as far as power consumption goes. They might be able to get another speed grade or two out of it, but they've already had to beef up the cooling and power requirements. I can't see them ramping the speed much until they finish their process change.
Given that, Barton's probably good enough for now in the desktop area. Getting the Opterons out is more important.

Also I've seen some rumors flying around suggesting that AMD is having trouble working with the SOI process for speed. Their yields are fine, but the speedbinning isn't what they want. This would make sense, too.

Of course, very few people outside AMD know what's really going on. Doubt we'll ever find out.

Wraith
Everything is going to be alright - for varying values of alright

[Asher]

Originally posted by Wraith
I don't know the details of the program, so I'm not going to be able to answer that. However, this is a workstation card. I don't see why their driver optimizations would give up accuracy for speed. That's the kind of thing you do for a gaming card. I'd expect they'd stick with the 80 bit x87 as much as possible.

I've been told by a couple of people that using 80-bit x87 precision is actually quite rare, especially on crossplatform programs like SPECViewPerf, which would most likely be a standard 64-bit float.

Of course we don't know either way.

And that, again, is not all of the story. The P4 requires a completely different set of optimizations. Some of the operations the P3 and the Athlon are good at suck donkey nuts on the P4, and vice versa. If they ran the drivers with all the P4 optimizations and nothing else there'd still be a major speed difference without SSE2.

That's true, the instructions that took the largest beating in terms of moving from P3 to P4 are the shifting ones, IIRC. Much slower.