some thoughts (blame NFS)

[BenSkywalker]

"Sigh.. We can go back and forth with this all day and night."

Yes, and so far every time anyone has entered into this conversation they have yet to disprove the fact that T&L, in its' current form, is not an advantage. All the benchmarks back my side of the discussion, none back yours.

"Ben I could easily answer everything you are brining up, without question.

I'm not the first one to bring many of these up, and they have yet to be answered in a definitive way by anyone, here or elsewhere.

"My point was to explain mis-conceptions about the V5. I had to do it by comparing it to something, and the GTS was the logical choice."

But you picked T&L to do it with. You knew that that would enter into a lengthy discussion(as you mentioned after my first post).

"But to say the V5 has all the disadvantages that I explained it doesn't (meaning it doesn't have the disadantages that I talked about) is not true."

T&L and Dot3 are disadvantages for the V5(along with trilinear and anisotropic). FSAA and 2D on Trinitron monitors are disadvantages for the GTS. I haven't argued about any sort of fillrate bound performance as I agree with you on that point, nor with any assumption that the V5 has an edge in FSAA(which I also think is true). You are arguing that the V5 is at least equal to the GF2 in every way and better in others, that is not the case. None of the current boards are better in everyway then any of the other current boards.

I have not stated the V5 svcks or any other such foolishness, you are just trying to remove very real advantages from the GF2. The V5, Radeon and GF2 all have their particular strengths, I am simply defending the ones that are real for the GF2.

Robo-

&quot;1) Ben likes 320x240 for his resolution, because he wants to play Toy Story with a gazillion polys <g>&quot;

I would take Toy Story graphics at 320x240 over Quake3@ 1600x1200, without hesitation.

&quot;2) That T&amp;L list is a joke. I'm sure 3dMark2001 and PacMan3d run much better with T&amp;L enabled <rolls eyes>&quot;

Tell that to someone with a PII 300

&quot;3) Where did you get the idea that Evolva runs much faster on a Radeon and/or a GTS?&quot;

Benchmarks.

&quot;4) The peaks can be truly &quot;discovered&quot; in a game like Q3 if you use a demo that is more &quot;crusher-like&quot;. i.e. remember crusher.dm2? It was intense. Here's a little trick. Run the Q3:TA timedemo. The GTS and 5500 are surprisingly close. With depth precision=faster on the 5500 in 32-bit, the 5500 is faster. D'OH!&quot;

Crusher.dm2 was a great CPU test, currently we have fillrate tests almost exclusively. I'm not sure on the bench that you are talking about, when do FPS fall off?

 

[DaveB3D]

lol.. Ben. If you seriously think the benchmarks back your side, you are soooooo wrong. I've run all the benchmarks countless times. They completely back me. I KNOW you can't say you've tested this as much as me. I never would have said anything if this wasn't the case.

And the REALLY funny part about your comment on the crusher demo being a CPU test (he didn't say that, he said LIKE that). if this is the case, the T&amp;L engine should be BEATING the V5. But it isn't. That speaks for itself.

 

[BenSkywalker]

&quot;If you seriously think the benchmarks back your side, you are soooooo wrong. I've run all the benchmarks countless times. They completely back me. I KNOW you can't say you've tested this as much as me. I never would have said anything if this wasn't the case.&quot;

MDK2, Evolva, TD6 and Quake3 in non fillrate bound situations back you? Could you please post some numbers then. No posted results, on any site anywhere on the net, have backed what you are claiming. I stated-

I haven't argued about any sort of fillrate bound performance

Please show numbers in non fillrate bound situations in the above games where the V5 is the equal of the GTS.

 

[BFG10K]

Dave:

I can't explain them. Not allowed.

You're no making this easy for me, are you?

It can cause some issues in 16-bit. However you won't find them in 32-bit. Well I've looked all over for them and I can't find a single one.

But surely the V4 is memory bandwith limited at 32 bit colour?

The GTS gets a higher frame-rate at 640x480. Why? Because of T&amp;L.

If you look at Anandtech's V4 4500 review, at 640 x 480 x 32 you will see the minimum framerate of the V4 is 48 FPS and for the MX it's 55 FPS. Now there is no memory bandwidth limitation or T&amp;L to affect the results and given how heavily CPU dependant UT is, it's amazing such a difference occurs.

The V4 is clocked at 166 MHz and the GF2 MX is at 175 MHz. Anybody can see the MX is more powerful than the V4, even without T&amp;L. Unless you consider video card speed how fast the RAM is.

Not more, just different operations.

What do you mean &quot;not more&quot;? If T&amp;L is being done by the CPU you are using cycles which would otherwise be available for other things.

The Ultra will be better simply because it has more fill-rate

Exactly. Just like the GF2 has a higher fillrate than the V5 and the GF2 MX has a higher fillrate than the V4. Assuming infinite memory bandwidth, nVidia's solutions will always be faster, with or without T&amp;L.

Ok Ben, I hope you don't seriously believe that increasing polygon counts reduces aliasing.

I just had to answer thise one. Of course it reduces aliasing. When you reduce the geometric detail in Quake 3 it reduces the polygon counts on the curves and makes them look rougher (ie aliased).

Also Quake 1 had a very low polygon counts so a lot of the curved surfaces (ie enemy faces) where basically just large squares/triangles. When Quake 2 came out the polygon count rates went up to much higher levels so the curves were much smoother.

And take a look at what they can do with Unreal 2 engine. They have a monstrous amount of polygons (with the help of T&amp;L) to use and the images look stunning.

No comments on SGI? I didn't think so. SGI would never be where they are today if they relied on raw clockspeed CPUs to do all of their tasks.

 

[BenSkywalker]

&quot;And the REALLY funny part about your comment on the crusher demo being a CPU test (he didn't say that, he said LIKE that). if this is the case, the T&amp;L engine should be BEATING the V5. But it isn't.&quot;

Huh?

Robo-

Can you run Crusher.dm2 and post your numbers- 640x480 16bit, you have nearly double the CPU power I have so you should blow me away(you have the V5 in now don't you?). I will log off and try it myself(though I need to install Quake2 first so it will take me a few).

 

[DaveB3D]

No Ben, I'm talking about when you are using fill-rate. Why? Because that is what situation 99% of the people are going to be in.

BFG, nope. Not going to make it easy.

As for those benchmarks. Again, not the latest drivers (I don't think) and not using the optimizations I told you about.

I misunderstood what you were talking about with T&amp;L on the CPU. Granted, yes you are. However when you are fill limited you have CPU cycles to burn so it really doesn't matter.

As for having higher fill-rate, yeah they do. However the fill-rate is pointless because they don't have the bandwidth to support it. Why do you think NVIDIA doesn't allow 128-bit DDR on the MX? Because it would perform the same as the GTS because the GTS is so bandwidth limited. The GTS and V5 have the exact same bandwidth.

As for more triangles reducing aliasing, it again comes to signal theory and just plain old what is happening. You have the SAME aliasing. The difference is the case state. Consider how different angles look different for aliasing. How the V5 takes care of worse case angles better. Now does this mean that the worse case angles have more aliasing? No. The same thing applies with triangles. You have JUST as much aliasing, however the angles change and pixel alignment changes. The aliasing is still there though, the case state is simply different.

As for the SGI part, I'm not sure what you are referring to.

 

[DaveB3D]

Ben.. My point is that he didn't say he WAS running crusher he said he was running a demo LIKE crusher.

As for my point, it was that if this demo was more CPU intensive then the GTS should have no trouble beating out the V5... but the thing is, it doesn't. As he stated, he is finding the V5 is beating the GTS. That speaks for itself, whatever the resolution is.

 

[BFG10K]

Dave:

As for having higher fill-rate, yeah they do. However the fill-rate is pointless because they don't have the bandwidth to support it.

But you can drop resolutions. I would rather much rather play 100 FPS @ 640 x 480 than 50 FPS @ 1024 x 768. While you are enjoying your high res I'll be running circles around you in multiplayer games.

As for more triangles reducing aliasing, it again comes to signal theory and just plain old what is happening.

I don't get what you are saying at all. All I know is that when you increase the gemetric detail in Quake 3 the curves are *much* smoother and look nicer. All I care about is how it looks on the screen.

Because T&amp;L enables you to push more polys, games like Unreal 2 are becoming possible and will soon become the norm.

As for the SGI part, I'm not sure what you are referring to.

Alright! I know something about graphics that Dave doesn't.

 

[RoboTECH]

Ben:

Where did you get the idea that Evolva runs much faster on a Radeon and/or a GTS?&quot;

Benchmarks.



uh...that's my point. The benchmarks DISPROVE what you say, not &quot;prove&quot;

Crusher.dm2 was a great CPU test, currently we have fillrate tests almost exclusively.

it wasn't a CPU test above 1024. 10 bots on Q3tourney4, with all those shaders, brought both cards to the 30 fps area.

BTW, if it was a 'CPU-limited test', then wouldn't a MASSIVE advantage go to the GTS and it's awesome T&amp;L?

MDK2, Evolva, TD6 and Quake3 in non fillrate bound situations back you?

we've had this discussion. you have a low-end CPU. It'll help you. T&amp;L does nothing for me.

BFG:

But surely the V4 is memory bandwith limited at 32 bit colour?

fillrate limited more than likely

Anybody can see the MX is more powerful than the V4, even without T&amp;L.

no doubt about that

 

[DaveB3D]

sigh..

Well if I play at 1024x768 with 16-bit I can hit 100 fps. If I go to 32-bit and still at maximum detail I can hit around 85. This is the average mind you.

I don't think I can explain the aliasing part any better. Not without more typing than I'm willing to do.

As for the new Unreal stuff, yeah on advanced T&amp;L hardware, something that the GTS isn't.

As for the SGI stuff, I'm just not sure what you are refering to.


Robo,

the MX is not faster than the V4 in fill-rate limited situations... here are a couple of scores:


1600x1200x32 High Quality
MX 7.17 - 12.6
V4 latest - 13.4

 

[lsd]

<< the MX is not faster than the V4 in fill-rate limited situations... here are a couple of scores: >>

go here...
Maybe you misplaced the 2..

 

[DaveB3D]

nope, those are the scores.... I'd suspect it has something to do with the detail settings used.

 

[lsd]

Try using the offical 6.31 drivers. 7.17's generally slowed down performace on all gf's.

 

[DaveB3D]

I will tomorrow. I didn't run these ones myself. I don't have any V4 scores with me, so I just asked somebody for them.

 

[RoboTECH]

could be a settings issue

Anand uses &quot;normal&quot;, then sets the color and texture depth manually

that would explain why the MX does better. it's using bilinear, just like the V4.

of cousre, that is with TC enabled on the MX, which is ass-ugly.

 

[lsd]

Either way the mx does have a higher fillrate than the v4.

 

[DaveB3D]

Yeah, this was max detail AFAIK. I'll run some tests tomorrow if I have time. I've got a LOAD to do tomorrow though, so no promises.

 

[BenSkywalker]

For performance then, we were not arguing the same thing. If you are saying when fill limited the V5 will be nearly equal with the GF2 in terms of performance then I agree. My point is that if you have a situation that is T&amp;L/CPU limited the GF2 will be faster.

&quot;Ben.. My point is that he didn't say he WAS running crusher he said he was running a demo LIKE crusher.

As for my point, it was that if this demo was more CPU intensive then the GTS should have no trouble beating out the V5... but the thing is, it doesn't. As he stated, he is finding the V5 is beating the GTS. That speaks for itself, whatever the resolution is.&quot;

And that is why I asked when the FPS dropped off. IF the drop with resolution increases then it isn't a very good CPU test. If they stay roughly flat between 640x480 16bit and 800x600 32 bit or roughly there abouts, then it would be a good point. Without seeing when the FPS dop off neither of us can say if it IS a good CPU bench or not. There was a Quake3 crush demo that was made, but even with my 550 I had to crank the res, detail, and color depth up to get any drop in FPS.

 

[ElFenix]

the SGI thing was that SGI used lots of parallel processing to get things done using a lot of specialized processors. in comparison to the t&amp;l part being the specialized processor running in parallel with the CPU. of course, you can also look at it saying that the v5 with SLI is using the SGI route of parallel processing rather than nvidia's single processor. its all brute force though. v5 and GTS have 25 million transistors. imagine what a tiler with 25 million transistors and 5.2 GB/sec of bandwidth would accomplish.

oh, and anand's review was using the 1.03 voodoo drivers, probably with no driver tweaking if that is even possible. i consider this one of the better review sites but there is still no full disclosure to the methodology/settings.

 

[NFS4]

<< Robo,

the MX is not faster than the V4 in fill-rate limited situations... here are a couple of scores:


1600x1200x32 High Quality
MX 7.17 - 12.6
V4 latest - 13.4 >>


Ummm, yeah....



http://www.anandtech.com/showdoc.html?i=1340&amp;p=6
QIII Arena 1600x1200x32 High Quality

Voodoo 4 4500 AGP - 15
nVidia GeForce MX - 21

MDK2 1024x768x32

Voodoo 4 4500 AGP - 40
nVidia GeForce MX - 69

UT 1024x768x32

Voodoo 4 4500 AGP - 12
nVidia GeForce MX - 24

UT 1024x768x32 &quot;minimum frame rate&quot;

Voodoo 4 4500 AGP - 12
nVidia GeForce MX - 24

Seems to debunk your theory of nVidia cards bouncing all around - doesn't bounce any more so than the Voodoo 4.

UT 1024x768x32 &quot;average frame rate&quot;

Voodoo 4 4500 AGP - 34
nVidia GeForce MX - 53

UT 1600x1200x32 &quot;minimum frame rate&quot;

Voodoo 4 4500 AGP - 9
nVidia GeForce MX - 13

UT 1600x1200x32 &quot;average frame rate&quot;

Voodoo 4 4500 AGP - 27
nVidia GeForce MX - 36

Not ONCE does the V4 top the MX. Not even in UT.


And why the heck are you using the 7.17 drivers for the MX? Those are leaked. They aren't even official. You should be using the 6.31's? I don't see you using leaked alpha/beta build of 3dfx drivers that haven't been officially released by 3dfx to do benchmarks.

 

[DaveB3D]

Well as I said, I didn't run those numbers.. and also, as I said, maximum detail. High quality is lower than that... and your V4 numbers aren't using the driver optimizations either..

 

[NFS4]

But how do you justify using the 7.17 drivers?

 

[DaveB3D]

As I SAID... I DIDN'T RUN THE TEST!

 

[OneOfTheseDays]

I know, what kind of idiot benchmarks with the 7.17 detonators. Everyone knows they suck and that they offer crappy performance and that they are betas. Robo's just trying to stack the deck against nvidia. Please if your going to benchmark cards at least do it fairly. Oh and Dave i borrowed a v5500 from a friend this morning and had all day to try it out. Let me say this, it is not worth the trouble of switching cards over. I run my desktop at 1280 and the text was only a little bit sharper, but it is already sharp enough on my geforce 2. As for quake 3, it IS smoother on my geforce 2. Unreal tournament, using loki, looks better on my geforce but plays a bit smoother on the voodoo 5500. One thing i can say about the voodoo was that it had good direct 3d perf., better than my geforce i must say. Everything was very smooth,and the FSAA is good, but not that much better than my geforce. I managed to play around with the HSR and let me say this, it is REALLY BUGGY. Sure, the benchmarks give higher fps, but when you actually start playing you get weird graphical glitches all over the place. I had my fsaa at X2 and i set the hsr setting to 2. Anyways, both cards have their own faults, but i am happy with my geforce 2 because i believe it will outlast the voodoo in the long run because of t&amp;l. Hopefully game companies will implement t&amp;l better so we can see some real advantages.

 

[DaveB3D]

sigh... well the last time I used a V5 in a system that had a GTS installed on it, I lost 9 fps off the top.. then factor in that I know you didn't use the optimization features I've talked about. I've never talking about the HSR...


If you are running games in any kind of decent resolution (10x7x32 and up) T&amp;L isn't going to make any difference for you.