"Current" Games and replacing the benches.....

[BFG10K]

BenSkywalker:

60FPS for single player is what I look for, and a Celeron 300MHZ could deliver that with my gfx card at 1024x768 32bit UHQ

But a Celeron 300 couldn't deliver anywhere near 60 FPS in Quake 3. My point exactly.

I have tried it. Going from 320x240 all the way up to 1280x960 results in a whopping 12FPS drop, about the same that UT does(actualy, UT scales quite a bit more using Rev's bench).

That's not what I'm seeing.

640 x 480 x 16....85.74 FPS
800 x 600 x 16....79.01 FFS
1024 x 768 x 16....66.47 FPS

640 x 480 x 32....75.71 FPS
800 x 600 x 32....64.32 FPS
1024 x 768 x 32....47.34 FPS

Also with my old Voodoo 3 it used to get quite jerky when going close to walls (640 x 480 x 16). A CPU upgrade had no effect on it, but when I upgraded my video card the problem disappeared. If Unreal was CPU dependant like UT is as you say, a CPU upgrade would have fixed the problem.

Also getting close to walls has a high amount of overdraw. That explains why the Voodoo couldn't handle it but the GF2 MX could. If the CPU was removing overdraw like you say it is, it wouldn't have made a difference what video card I used.

 

[jpprod]

Also with my old Voodoo 3 it used to get quite jerky when going close to walls (640 x 480 x 16). A CPU upgrade had no effect on it, but when I upgraded my video card the problem disappeared. If Unreal was CPU dependant like UT is as you say, a CPU upgrade would have fixed the problem.

Just a brief comment. Do you play with detail textures enabled? Unreal as well as UT apply multiple layers of detail textures on surfaces as their primary color texture gets overfiltered. I don't know the Unreal engine technology on code level, but I've seen easily three to four layers of detail textures appear on walls when approached. Voodoo3 dies in this situation simply because of fillrate. Additional layers of textures which cannot be handled by multitexturing also need surface's polygons re-transformed and lit, so browsing detail textures close to walls requires a little extra CPU power too.

But Unreal engine is a CPU dependant one majorly because it incorporates a software driven overdraw elimination/surface visibility determination algorithm. It actually seems to work very well and effectively eliminating much of the overdraw (plus it no doubt justifies fillrate-hogging surface effects such as multilayered detail textures), but on current systems with plenty of fillrate it's just a big CPU time hog. Unreal was originally designed to be an engine with software rendering, and in that world filling polygons with perspective-corrected textures is easily more CPU-intensive than a heavy per-pixel/per-vertex visibility determination algorithm.

While impressive and innovative on many areas, a video card's performance in Unreal engine is by no means a good representation of generic 3D-performance. There are just too many quirks and oddities in the engine. Quake III is a much more streamlined benchmark, but it's really getting old - who cares whether the game runs at 120FPS or 180FPS?

Out of current games, Giants would be the one from which I'd first want to see benchmark results taken from. It's still pretty unclear what limits performance of this game, but being a good game with a bunch of DX8 features make it a viable next-generation benchmark in my book.

 

[BenSkywalker]

BFG-

But a Celeron 300 couldn't deliver anywhere near 60 FPS in Quake 3. My point exactly.

A Celeron 300 DOES push ~60FPS in Quake3

Jukka already covered Unreal, have you force enabled detail texutres on your GF2MX under Unreal yet?

Jukka-

Have you downloaded the two Giants demos yet? They just posted the second one(Delphi demo) a few days ago.

 

[jpprod]

Have you downloaded the two Giants demos yet? They just posted the second one(Delphi demo) a few days ago.

Played through both aldeady, a manificent game! I especially like the huge size of environments, excellent music and cutscenes with twisted humour. I'm torn whether I should pick this one up, since there won't be a patch (Planetmoon has said one would be too large, around 300 megabytes) which'll upgrade the game to "GF3 special version" utilizing advanced DX8 features. Still hoping that the price of Elsa GeForce III (salty $660 will be the retail price) would come down to a more resonable level so I could get the enhanced version of Giants

 

[BenSkywalker]

Jukka-

I see your situation like this. If you go out and buy the game, you could get your hands on a burned copy of the updated version and replace your own. I'm firmly against piracy, but if you already own the game, and are just burning an update to it, I can honestly say that I wouldn't have the slightest moral issue with it.

The game fuggin rocks, you gotta pick this one up

BTW- On the Elsa price, is it that expensive over there???? I've seen pre orders stateside going for less then $500.

 

[RoboTECH]

60 fps average in q3, even in botmatch, is kinda crappy.

 

[BFG10K]

Jpprod:

Do you play with detail textures enabled?

Yes. With both cards. Even with 32 bit colour (vs my Voodoo's 16 bit colour) everything stays smooth.

Voodoo3 dies in this situation simply because of fillrate.

Which verifies my original claim that Unreal is video card dependent. I'll have to test whether UT has this wall problem on my friend's system because he still has a TNT2.

BenSkywalker:

A Celeron 300 DOES push ~60FPS in Quake3

Sorry, my mistake. I forgot about T&L. I was using the scores from a TNT2.

Jukka already covered Unreal, have you force enabled detail texutres on your GF2MX under Unreal yet?

I am using the same settings that my Voodoo 3 was using. Except for 32 bit colour, of course.

 

[jpprod]

BFG: Voodoo3 dies in this situation simply because of fillrate.

Which verifies my original claim that Unreal is video card dependent. I'll have to test whether UT has this wall problem on my friend's system because he still has a TNT2.

Well yes and no. My point was that Unreal engine's performance depends from different factors at a different situation. Whereas Quake III is clearly fillrate limited at high and CPU limited at low resolution, limiting factor in Unreal engine varies during gameplay. On the average my estimate would be that CPU limits Unreal more than video card - player doesn't usually spend long times staring at multilayered wall detail texture at a very close distance. Even though I am myself an exception to this rule

Ben: I see your situation like this. If you go out and buy the game, you could get your hands on a burned copy of the updated version and replace your own. I'm firmly against piracy, but if you already own the game, and are just burning an update to it, I can honestly say that I wouldn't have the slightest moral issue with it.

I agree, and in my opinion the patch should be made publicly downloadable no matter what size it is. But I'll have to observe the situation - I'd love to get a GeForce III card anyway to get into some DX8 3D programming, and Elsa would be my choise of brand because of their high quality and 6 year warranty. But the price has to come down a little.

BTW- On the Elsa price, is it that expensive over there???? I've seen pre orders stateside going for less then $500.

It's the grim truth - retail price, directly from importer's mouth, is 4300 finnish marks which translates to chest-pain-inflicting $660. Whitebox OEM version (Giants still bundled) will sell for 4100 finnish marks, "just" $630.

Perhaps I should participate into nVidia donor program

 

[Vrangel]

Guys, Serious Sam will be out in a few days.
Serious candidate for benchmarking IMO.

 

[oldfart]

I agree that SS will ROCK!! It should be a pretty high stress OpenGL benchmark. Huge open areas, high detail textures, lots of lighting effects, S3TC. Bye-Bye Q3, Hello Serious Sam.

 

[sharkeeper]

Get Serious!

Cheers!

 

[Vrangel]

Yeah, hardware T&L too !

And huge number of enemies.

 

[BFG10K]

Jpprod, I understand your comments but I'm not convinced Unreal is using exactly the same engine as UT. I'm more inclined to say that UT is more CPU dependent than Unreal.

Also I've seen numerous comments on the web that say that UT is using an enhanced version of the Unreal engine. To add some more food for thought, here are my 32 bit UT scores compared to Unreal:

UT
640 x 480 x 32....75.84 FPS
800 x 600 x 32....74.49 FPS
1024 x 768 x 32....62.13 FPS

Unreal
640 x 480 x 32....75.71 FPS
800 x 600 x 32....64.32 FPS
1024 x 768 x 32....47.34 FPS

As you can see, my system is doing much better in video card limited situations in UT. Granted they're different benchmarks, but it still shows interesting results. Unreal is scaling linearly downward with resolution increases but UT isn't.

 

[jpprod]

Here's the formula

UT engine = Unreal + character LOD + slightly less horrible non-glide renderers

Engine architecture's exactly the same, UT codebase is a small upgrade. Some licensees - like 3DRealms - have upgraded their in-developement games to use UT codebase; if it was radically different this would not have been a sensible transition. On a positive side note, 3Drealms has also confirmed that they removed engine's heavy software HSR algorithm, which was one of the reasons why Unreal/UT engine is so CPU-bound.

I have another theory which your benchmark results support: UT may seem more CPU-bound because a bigger bottleneck than CPU or fillrate has been removed. According to Tim Sweeney's updates at Unreal tech site (at least some version of) Unreal's D3D renderer allocated each texture up to three times (non-local copy, non-local format-converted copy, local copy). The horrible renderer may be dying in the high-resolution benchmark runs because there's less local video card memory for textures. UT is CPU-bound because it's improved D3D/OpenGL renderer handles texture memory in a less-horrific way, and simply because there's likely less distinct textures in that benchmark run. Try decreasing texture detail keeping "detail textures" setting the same as in those runs so fillrate requirements don't change. I'm willing to bet that with texture memory shackles released somewhat, Unreal runs smooth even at very high resolutions.

 

[oldfart]

I have to agree with BFG. Unreal does not run as smoothly on my system as UT. Also notice that UT has in the menu system, 32 bit color while Unreal does not. I can enable Trilinear in UT with no frame rate hit. In Unreal, it causes a large hit. They are different. I once tried to transplant the D3DDrv.dll and D3DDrv.int from UT to Unreal. Lets just say it didn't work.

 

[BFG10K]

Oldfart:

are different. I once tried to transplant the D3DDrv.dll and D3DDrv.int from UT to Unreal. Lets just say it didn't work

Hehehe! That's exactly what I tried to do as well.
Great minds think alike.

Try decreasing texture detail keeping "detail textures" setting the same as in those runs so fillrate requirements don't change.

Jpprod, I don't remember there being a texture detail slider in Unreal. The only thing I remember is the "Detail Textures" setting, which is either yes or no.

I'll have a look though.

 

[BFG10K]

Jpprod I checked but there's no texture slider in Unreal. The only texture controller is "Detail Textures". I turned it off and got a small boost (~3 FPS) across most resolutions.

 

[BenSkywalker]

Serious Sam is definately the bench that I would like to see for OpenGL, at least if they have to stick to one bench per API

UT/Unreal and switching dlls-

You can't switch them between different versions of UT, from one build to another, there is no way you would be able to swap them to the considerably older Unreal level support. The engine is the same, it has been modified but it is still the same engine. For quite some time after Unreal shipped it was unplayable under D3D at all(in terms of FPS), now it runs pretty smooth. UT when it launhced was simply OK, that has also gotten quite a bit better with time.

Half-Life runs on a modified Quake1/2 engine, but it is using the same engine. UT is built on the Unreal engine. Unreal2 will be the second engine for the series(it isn't a coincidence it has a "2" in the title).

BFG- Which benches were you using for each game? Was it "Flyby" for Unreal and Rev's Thunder for UT? Or was it "Flyby" for both?

 

[RoboTECH]

BFG:

I am using the same settings that my Voodoo 3 was using. Except for 32 bit colour, of course

please tell me you weren't using D3d for UT with the V3. Please....

. I'm more inclined to say that UT is more CPU dependent than Unreal.

the benchmark may have effects which are more CPU dependent. Remember, it depends on what happens in the game.

To add some more food for thought, here are my 32 bit UT scores compared to Unreal

again, it depends on the benchmark. Different maps may have different effects which affect the game differently. A perfect example is Q3 and Q3:TA. The TA maps are far more CPU-dependent, despite the fact the same game engine is used.

remember dude, Alice and Q3 have the same game engine, but that doesn't mean that a recorded demo will benchmark the same.

 

[JellyBaby]

Whatever happened to 3D Gamegauge? Seemed like a reasonable approach for "real world" benching. You take several fps samples from a handful of games from a variety of genres using a mix of apis, then average them. Of course video card makers probably don't care for it as new products don't always "shine" under a bench like this. However, consumers can get an accurate idea as to how much better their games will run after an upgrade.

I'd also like to see a decent fps counter in more games, one that tells me the average, lowest and highest fps reading.

 

[PeAK]

Ben, you seem to like MTBR a lot (so does Tom) to reveal the capabilities of the latest generation hardware is made of as you can see from your response in this post

1. 
   
   << &quot;Oh yeah, my vote for the current Direct 3D game would have to be Mercedes Beinz Truck Racing. That game will bring any system to it's knees.&quot;
   
   Now that's what I'm talking about That is the type of benching I want to see >>
   

Tom made his name originally with gathering benchmarks on something called &quot;Monster Truck Madness&quot; and now Mercedes Benz Truck Racing...the more things change...

You also stated the advantages following from this post

1. 
   
   << MBTR utilizes features and reflects performance common amongst current games, not just MBTR...MBTR, OTOH, gives a much better indicator....MBTR is an example of an already shipping game that uses features that most new 3D games are using.
   >>
   

to which Hans ask the following question about Radeon's
performance in the following post

1. 
   
   << ...even if the radeon is better at mbtf (i wouldnt know since i havent really researched this) >>
   

I ran across this MBTR benchmark on the low cost version of the Radeon called the VE underclocked by 10% at 183/183 beating out its direct competition in the form of the MX by about 30% in both 16 bit and 32 bit benchmarks in Mercedes Benz Truck Racing. The other Radeon cards should do much better especially since recent 7089 driver releases have seemingly removed all game glitches. An earlier driver used on the Radeon 32MB DDR review at Tom's showed the same performance gains on early generation drivers but also rendering errors that seemed to have been corrected.

 

[BenSkywalker]

PeAK

I think the more demanding and feature rich a game is, the better the Radeon will look in terms of performance. Outside of people using 3D visualization along with gaming, I have been reccomending the Radeon for some time now and unless you are looking for absolute performance(pre-GF3) where the Pro or Ultra would be the best bet, the Radeon is the overall better card for gaming when compared to the GF2 GTS and the budget DDR Radeon clearly blows away the GF2 MX.

 

[BFG10K]

RoboTECH:

please tell me you weren't using D3d for UT with the V3. Please....

Yes I was. I've already said multiple times in the past that it was because Glide was too dark.

As for your comments about different maps, I understand what you're saying. It's just that Unreal Tournament feels faster and seems less affected by video card capabilities than Unreal is.

Ben, I was using the castle flybuy for Unreal and the Cityintro for UT. Also I was comparing the gameplay between the games in various levels.

 

[Rellik]

BEN: About your earlier question: Yes, the databes I am currently develloping is for a website. It is still in it´s early alpha stage(sadly) and is aimed to be a sort of a portal which also featrues hardware. To give you an idea, here is the site´s hardware testing missin statement:

...Large Hardware sites such as Anandtech and Tom´s are great places
for detailed tests, but their testbeds don´t reflect a typical gamers system. While I know why they benchmark this way(it certainly has it´s merrits) we will take REAL gamers systems. This means old installs, tons of apps(or less) win98/se/me/2k and different drivers. All systems will have detailed descriptions and their scores will be comparable in a pull down database. Result differentials are shown in percent.

So the databese will have a clean install 700 P3 SCSI system as well as a 464Mhz P2 with winME.

If you like more info, pm me.

As for the general discussion:

When I read 60fps at 1024 UHQ on a 300 celeron with a MX I care to ask
HOW? I get under 60 fps at 1024 HQ (everything max and radeon tweaked to quality) on a 933 P3. And the rendering of the GF is not at the same level as the Radeon(please, no flames) So if one was to TRULY COMPARE the cards, we should tune the radeon to speed and the gf to quality. then the difference woulg decrease...

 

[BenSkywalker]

Rellik-

You have PM

&quot;When I read 60fps at 1024 UHQ on a 300 celeron with a MX I care to ask
HOW? I get under 60 fps at 1024 HQ (everything max and radeon tweaked to quality) on a 933 P3.&quot;

Huh? Not a GeForce2MX, I was saying a Celeron 300 could do it. I have a GeForce DDR and for me a Celeron 300 would have very little impact for me at that setting if any at all. How you ask? Detonator3s

&quot;And the rendering of the GF is not at the same level as the Radeon(please, no flames) So if one was to TRULY COMPARE the cards, we should tune the radeon to speed and the gf to quality. then the difference woulg decrease...&quot;

I'm pushing mid 50FPS at 10x7 UHQ no TC with ani enabled, at the very least the equal of the Radeon in Quake3(Radeon seems to revert back to bilinear when enabling anisotropic for some odd reason, makes the texture filtering a bit lesser then the nV boards, not to mention it appears to be using a ~-2 LOD bias adjustment which results in some serious texture aliasing, though it takes better screenshots). When I enable TC(and when I enable TC it is nigh identical to the Radeon, click the link in my sig and check out the nV S3TC fix article for details) and disable ani then I'm pushing ~60FPS. Above I linked to Anand's first review of the GeForce and he was in the ~60FPS range with the pre Det2 drivers on a Celeron 300(though it was an SDR so the 10x7 32bit numbers were not up to speed). The Det3 drivers perform nearly identical with TC off as the Det2's did with TC on(I have tested quite a bit), and now with the Det3s you can fix the TC to match the Radeon's.