DeltaChrome nitro

Sxotty

Legend
Ok http://www.theinquirer.net/?article=14588 yes I know the inquirer, but I was wondering is this sugesting that the regular part has clocks of 300/300, and the nitro part has 315/315 that seems like a very small difference between mid/high end products. I think I am confused about something here.

Anyway it is good to hear that their drivers are coming along well. It sounds like they may yet be a credible prescence.
 
Why is it there are 3dmark scores of this thing out in the wild, but that I can't find any bloody PICTURES of the board??? Weird!

Anyway, it's a bit disappointing to say the least a "high-end" part is so much slower than a chip that was released some 1 1/2 year ago despite having pretty much the same basic specs (8 pixel pipes, 4 vertex shaders, similar clock speeds)... Is this chrome-whatsitsname thingy running on a single-channel 128-bit memory interface or something? What is it that holds this thing back? Less than 12k 3dmark01 is pitiful, there must be GF4Ti boards that do more than that on a 3GHz P4 system.

Ok, so drivers might be dodgy, but for how long have they been working on this thing anyway? It can't be yesterday they got the first working driver build up and running!
 
Guden Oden said:
Why is it there are 3dmark scores of this thing out in the wild, but that I can't find any bloody PICTURES of the board??? Weird!

Anyway, it's a bit disappointing to say the least a "high-end" part is so much slower than a chip that was released some 1 1/2 year ago despite having pretty much the same basic specs (8 pixel pipes, 4 vertex shaders, similar clock speeds)... Is this chrome-whatsitsname thingy running on a single-channel 128-bit memory interface or something? What is it that holds this thing back? Less than 12k 3dmark01 is pitiful, there must be GF4Ti boards that do more than that on a 3GHz P4 system.

Ok, so drivers might be dodgy, but for how long have they been working on this thing anyway? It can't be yesterday they got the first working driver build up and running!

Its only got a 128bit memory bus
 
dan2097 said:
Guden Oden said:
Why is it there are 3dmark scores of this thing out in the wild, but that I can't find any bloody PICTURES of the board??? Weird!

Anyway, it's a bit disappointing to say the least a "high-end" part is so much slower than a chip that was released some 1 1/2 year ago despite having pretty much the same basic specs (8 pixel pipes, 4 vertex shaders, similar clock speeds)... Is this chrome-whatsitsname thingy running on a single-channel 128-bit memory interface or something? What is it that holds this thing back? Less than 12k 3dmark01 is pitiful, there must be GF4Ti boards that do more than that on a 3GHz P4 system.

Ok, so drivers might be dodgy, but for how long have they been working on this thing anyway? It can't be yesterday they got the first working driver build up and running!
Its only got a 128bit memory bus
Ditto for GeForce 4... which I think was Guden's point.
 
OpenGL guy said:
dan2097 said:
Guden Oden said:
Why is it there are 3dmark scores of this thing out in the wild, but that I can't find any bloody PICTURES of the board??? Weird!

Anyway, it's a bit disappointing to say the least a "high-end" part is so much slower than a chip that was released some 1 1/2 year ago despite having pretty much the same basic specs (8 pixel pipes, 4 vertex shaders, similar clock speeds)... Is this chrome-whatsitsname thingy running on a single-channel 128-bit memory interface or something? What is it that holds this thing back? Less than 12k 3dmark01 is pitiful, there must be GF4Ti boards that do more than that on a 3GHz P4 system.

Ok, so drivers might be dodgy, but for how long have they been working on this thing anyway? It can't be yesterday they got the first working driver build up and running!
Its only got a 128bit memory bus
Ditto for GeForce 4... which I think was Guden's point.

Whoops for some reason I thought he was comparing to a radeon 9700 :p

3d mark 2001 is a very poor bench for comparing new cards as to do well on it you have to app specifically optimize nature. Thus quite a lot of the result will depend on how aggressively (if at all) sis is optimizing the nature test. The gf4 when using drivers up to 44.90 has some very aggresive optimizatiosn for the nature test i.e. almost double the frame rates. There are less in later drivers (which gives people more evidence to say later drivers are slower on geforce 4s, which may be true but not to the extent outlined)
 
dan2097 said:
OpenGL guy said:
dan2097 said:
Guden Oden said:
Why is it there are 3dmark scores of this thing out in the wild, but that I can't find any bloody PICTURES of the board??? Weird!

Anyway, it's a bit disappointing to say the least a "high-end" part is so much slower than a chip that was released some 1 1/2 year ago despite having pretty much the same basic specs (8 pixel pipes, 4 vertex shaders, similar clock speeds)... Is this chrome-whatsitsname thingy running on a single-channel 128-bit memory interface or something? What is it that holds this thing back? Less than 12k 3dmark01 is pitiful, there must be GF4Ti boards that do more than that on a 3GHz P4 system.

Ok, so drivers might be dodgy, but for how long have they been working on this thing anyway? It can't be yesterday they got the first working driver build up and running!
Its only got a 128bit memory bus
Ditto for GeForce 4... which I think was Guden's point.

Whoops for some reason I thought he was comparing to a radeon 9700 :p

3d mark 2001 is a very poor bench for comparing new cards as to do well on it you have to app specifically optimize nature. Thus quite a lot of the result will depend on how aggressively (if at all) sis is optimizing the nature test. The gf4 when using drivers up to 44.90 has some very aggresive optimizatiosn for the nature test i.e. almost double the frame rates. There are less in later drivers (which gives people more evidence to say later drivers are slower on geforce 4s, which may be true but not to the extent outlined)
Except that what SiS does on this benchmark is irrelevant since DeltraChrome is being made by S3 Graphics, which is a division of Via ;)

In any event, many of the tests in 3D Mark 2001 are CPU limited on modern cards at default settings. However, this can help show how efficient the driver is.
 
As for why DC doesn't have the performance of the R300 despite some superficial similarities suffice it to say that there are many, many more factors which enter into an ASIC's performance than are actually seen. Cache sizes, register file organization, and other undisclosed factors have a large impact on performance as do pipeline setup, number of texture/shader units, etc...

EDIT: The relatively low bandwidth certainly doesn't help.
 
forgive my ignorance on all things S3, but is this DeltaCrome Nitro a board based on the first DeltaCrome GPU? I assume this is *not* the DC refresh. or maybe it is?

It will be interesting to see how S3 plays out the current DX9 generation (DC and DC refresh) as well as the next generation, DirectX10 (S3 Destination Films)
 
nobie said:
Hmm this would be a decent card for around $75.
No, no it wouldn't. The drivers are going to suck ass. Not worth having a video card to play games if it has bugs left and right.
 
Chalnoth said:
nobie said:
Hmm this would be a decent card for around $75.
No, no it wouldn't. The drivers are going to suck ass. Not worth having a video card to play games if it has bugs left and right.
I don't understand how you could have possibly reached this conclusion yet. (Or are you basing this solely on old S3, which is essentially a different, defunct company.) The card is still unreleased and there have only been offhand remarks about the drivers in these previews so far. Furthermore, the majority of things we've been hearing are good reports about how they've done vast amounts of bugfixing and performance boosting.

Megadrive: Yes, it is the same Deltachrome chip, upclocked a few MHz.
 
Heh. We'll see. I just don't believe that they'll be able to produce good drivers. Performance is only once small piece of the picture.
 
One thing to keep in mind while reading those graphs: trilinear. 'nuff said.
Anyway, I'll buy one. And then I'll start spamming their dev relations with OpenGL bug reports :D
 
AFAIK S8 only has one vector ALU per pipeline, that is not co-issue capable, so in comparison to other chips it can't execute anywhere near as many instruction per clock.
 
Lets put it this way...

I bought my GF4 Ti4200 128MB used for $75 a few months ago, and you can now buy them new for that price. I was reading the Firing Squad review of the DeltaChrome S8 Nitro and was curious how its performance stacked up against my (ever so slightly overclocked) Ti4200. So I installed the Quake 3 demo (old, but worthy for comparison's sake) and ran a timedemo. At 1280x1024 high quality, the DeltaChrome S8 Nitro apparently gets roughly 80 fps. With my Ti4200, I got double that.

So no, I'd say it's not even worth $75.
 
Actually, that sounds rather definitively like a driver efficiency issue. It would be interesting to track how that changes with driver releases compared to more GPU limited applications.
 
Quake 3 is old, but it's old and popular, and it's been around long enough that I don't see any reason a new card should be getting half the framerate of a Ti4200 in it.

But if you want a newer game to compare with, so be it. The DeltaChrome S8 Nitro got 24.5 fps in Splinter Cell at 1280x1024. My Ti4200 just got 25.4 fps, and that in the Class 2 adapter mode, not the Class 1 that the DeltaChrome is almost certainly running at. It's not the pathetic showing that Quake 3 was, but it's still nothing to be proud of, especially when it's being released a couple of years after the GF4, and it's planned to have an MSRP over twice the street price of a Ti4200.
 
Back
Top