Calling ALL Video Card EXPERTS!!

Untruest · Mar 20, 2005

I am checking a Visiontek xtasy radeon 9200 256mb as being the best pci card on the market based only on its specs. But if u have experience knowing very well what the following features really do in games, please critic.

• SMARTSHADERª 2.0
• SMOOTHVISIONª 2.1
• HYPER Zª III+
• TRUFORMª 2.0
• 4 Parallel Rendering Pipelines
• Programmable full-scene anti-aliasing
• Advanced Anisotropic Filtering
• Lossless Z-BUFFER Compression and Fast Z-BUFFER CLEAR enhances memory bandwidth up to 25%

• ATI¨ Radeon¨ 9200VPU
• 250 MHz VPU Core
• PCI
• 256MB DDR Memory
• 128-bit Memory Interface
• Memory Clock: 200MHz (400MHz DDR)
• Dual 400MHz DACs
• API Support: DirectX¨ 9.0, OpenGL

www.visiontek.com

Here is what i think,

Basicaly this is a directx 9.0 compatible 9200, with 256mb.

But what caught my eye is: truform 2.0 (no big deal), Hyper z III+ (used on 9800s) and smartshader 2.0 ( pixel & vertex shader above 2.0 i think)

I think 4 rendering pipelines is low if u can only process 1 pixel per pipeline.

And i am also worried about T&L. on the 9100 it had 2nd generation T&L which is why i think it was a success but on this 9200 they don't even mention if it has T&L which is why i am lead to believe it has the same T&L as a 7500.

I am buying a pci soon, and i can't decide whether this card is worth it. If u know of a better pci card please inform us.

TheGreatGrapeApe · Mar 20, 2005

It is one of the best you can get (other two options are FX5200/5600)

It does have hardware T&L, but the R9100 will be faster because it uses the R8500 core which is more capable IMO than the R9000/9200 core.

But no it's no PS2.0 capable.

Being DX9 compatible means it conforms to the specs (which include subsets), but it is NOT fully DX9 capable, speaking of which neither is the FX5200. Maybe the FX5600 is better but I even doubt that it can handle true DX9/PS2.0.

Bein stuck with PCI limits your options. You best IMO is the R9100, after that it's a toss-up between the 3 others. Maybe the FXs might come in handy around the era of Longhorn, but by then you'll need a new computer anyways so that's of little concern.

- You need a licence to buy a gun, but they'll sell anyone a stamp (or internet account) ! - RED GREEN GA to SK :evil:

Untruest · Mar 21, 2005

It's interesting what u say, so the 5200 who clearly states to be directx 9.0 compatible may not be fully capable of handling it?

I am not sure what an IMO is, but in any event, i saw a 9250 today with 2nd generation T&L similar to the 9100 with 256mb at 500mhz 128 bit.

U obviously bring up an importance about the core that is being used. I always ignored whether it was a 9200 or 9000 and payed more attention to the specs.

Could u name some advantages a core could have over another?

BTW i saw a pci 5700le. Can't quite remember the specs. But it should also compete.

Another thing is apparently the 9250 i am talking about is on a x800 core. Check it out at www.bestbuy.com

Crashman · Mar 21, 2005

The 5200 supports full DX9 features but is so extremely weak that turning these features on results in unplayable framerates in games.

9200 is a more powerful solution, even if it's just using DX8 hardware.

If you compared AGP cards rather than PCI, you'd get the results 5200<9200<9100<8500<9500<Ti4200<9600XT when playing DX8 games, and the 5200 looks even worse with DX9 features enabled, which can't be shown with the 9200, 9100, 8500, or Ti4200.

Only a place as big as the internet could be home to a hero as big as Crashman!
Only a place as big as the internet could be home to an ego as large as Crashman's!

cleeve · Mar 21, 2005

If you can find a PCI 5600/5700 or Radeon 9100, get it.
Otherwise, the 9200 you have there will be good, it's almost as good as those other options because any card you put in there will be limited buy the PCI bus.

________________
Radeon 9700 PRO (o/c 332/345)
AthlonXP 3200+ (Barton 2500+ o/c 400 FSB)
3dMark03: 5,354

Untruest · Mar 21, 2005

Enabling DX9 features, only calls for better Visual effects/graphics,right?

IF it does, the 9200 from visiontek and the 5200 may both support pixel 2.0 and vertex 2.0 but they are just to weak to get any decent FPS. Can this be overcome through overcloking?

I am one of those people that think that the difference from pci to agp isn't marginal. Given that u keep most of the data local to the video card and minimize processor-video card data transfers, that any pci card can compete with its agp twin.

How do u suppose i will be really limited by a pci card ( becaue of bandwidth? How much dara is really being transfered per second anyway? 10mb?)

And what about the difference in Core, does it really do anything?

Thanks,

cleeve · Mar 21, 2005

1. All Readons below 9500 are DirectX 8 parts. The 9200 does not support PS or VS 2.0.

2. The 5200 has architecture so horrible that it cannot really run any game using the DirectX 9 shaders, in fact it's mostly slower than the 128-bit Radeon 8500, 9100, and 9200s.

Overclocking can't really help it, the problem lies within it's architecture.

3. Core (architecture) is king. What's most important on a videocard:

1. GPU (the architecture, # of pipelines, etc) - MOST IMPORTANT
2. Bus (Old PCI vs. new AGP or PCI-express)
3. Memory Interface (64, 128, or 256-bit)
4. Clockspeeds
5. AMOUNT of RAM (Least important, assuming you have the minimum spec for the game*)

To clarify... all the factors above are relevant to videocard performance, but this is the general heiarchy.

GPU architecture is the most important. How many pipelines the card has and how it uses them, which advanced affects the processor can handle.

Second most important is the bus. An old PCI bus cannot compete with the AGP or PCI-express bus. It simply can't move the data fast enough, and framerates will cap out quickly at any resolution past 800x600.

Third most important is the memory interface. This relates to how FAST the memory is accessed, which in most cases is much more important than now MUCH memory there is. A 64-bit memory bus is only half as fast as a 128-bit memory bus, and a 256-bit memory bus is twice as fast as that.
Be wary of cards like that have the "SE" or "LE" designation... in most cases, it means their memory bus is half of what the non-SE versions are, and this has a large impact on performance

Forth most important is clockspeeds. Higher clockspeeds on both the GPU and RAM mean higher performance. But remember, 400 Mhz memory on a 64-bit bus is the same speed as 200 Mhz memory on a 128-bit bus.

Least important is the AMOUNT of RAM. Developers create games for the lowest common denominator, and that means they are still making 64-meg texture sets for games. But the newest games are starting to make use of 128-megs of RAM, so 128-megs is the minimum. A few games will show a slight performance increase with 256-megs of RAM, but these number less than 5 I can think of.

256-Megs of RAM is, in most cases, a marketing tool for uneducated people to make a buying decision on. But trust me, you'd rather have a 64-meg Geforce4 Ti4200 than a 256-meg Radeon 9600SE.

________________
Radeon 9700 PRO (o/c 332/345)
AthlonXP 3200+ (Barton 2500+ o/c 400 FSB)
3dMark03: 5,354

Crashman · Mar 21, 2005

Since I don't have time now for a detailed answer, I'll provide this response to bump your post up the list for others to see.

Only a place as big as the internet could be home to a hero as big as Crashman!
Only a place as big as the internet could be home to an ego as large as Crashman's!

Untruest · Mar 22, 2005

The radeon 9200 is indeed a directx 8.1 card, however the one i am talking about made by visiontek has smartshader 2.0. According to ATI, smartshader 2.0 supports pixel and vertex shader 2.0.

Check out the websites but if u still think i am wrong then state how is it false advertisement.

In any event, i've already given up on that card ever since i found a 9100 will 4 rendering pipelines with 2 pixels per pipeline. 128 bit memory interface, which isn't the speed cleeve but rather the size at which data is being transferred.

It's true that agp 8x far exceeds the pci bus in bandwidth, but i am curious to know how many megs is really being transferred between the card and the cpu or the card and system ram.

(Trying to get back to the topic,lol) I am trying to take that advice about not getting a directx 9.0 card if it were weak because it would just lag even more with directx 9.0 features unabled. Before I make a final decision. Do u know if the 5700le is packing enough heat to run directx 9.0 features.

cleeve · Mar 22, 2005

If it's a 92xx series card, it doesn't have smartshader 2.0... that's a typo on the visiontek site.

Here's a great review of PCI videocards:

<A HREF="http://www.sudhian.com/showdocs.cfm?aid=445" target="_new">http://www.sudhian.com/showdocs.cfm?aid=445</A>

________________
Radeon 9700 PRO (o/c 332/345)
AthlonXP 3200+ (Barton 2500+ o/c 400 FSB)
3dMark03: 5,354

TheGreatGrapeApe · Mar 22, 2005

128 bit memory interface, which isn't the speed cleeve but rather the size at which data is being transferred.

LOL!

Yeah Cleeve, ya' n00b, dontcha know that !?!

- You need a licence to buy a gun, but they'll sell anyone a stamp (or internet account) ! - RED GREEN GA to SK :evil:

cleeve · Mar 22, 2005

You can move 128-bits of data through a 128-bit interface twice as fast as you could move it through a 64-bit interface.

I maintain the validity of the speed analogy, screw you guys!

________________
Radeon 9700 PRO (o/c 332/345)
AthlonXP 3200+ (Barton 2500+ o/c 400 FSB)
3dMark03: 5,354

TheGreatGrapeApe · Mar 23, 2005

LOL!

You can move 128-bits of data through a 128-bit interface twice as fast as you could move it through a 64-bit interface.

But what if it's 4 seperate 32 bit parcels running on @ 150 mhz on the 128bit interface versus 350mhz on the 64bit interface? Wouldn't that 64bit interfaces have the faster traffic? 'Cause like you said the 128bit is faster? WTF!?!

In which case the SE cards stand for S-OOoo E-xtreme!

Damn now I'm all confused. Just after I figured out what effective means, like I want frickin' ineffective RAM, what do you think this is a Xabre!

Speaking of which, what does GDDR stand for, I've been told by Eden it stands for Gee'-Dat'-Der'-Ram, but if that's the case what's the 3 stand for? 3 for 1 discount, third times a charm, good things come in 3s, Ram for 3D cards [MAtrox uses GDDR2?], any clues?

I think they should simply solder some extra dimms to the board and then you can upgrade without gettng a new card, right?!?

You guys just make it so complicated to try and find a new card for minesweeper that I'm getting a little anxious about the Longhorm Minesweeper being too much for whatever choice I end up with! :frown:

- You need a licence to buy a gun, but they'll sell anyone a stamp (or internet account) ! - RED GREEN GA to SK :evil:

magneezo · Mar 23, 2005

I saw a 9200 on sale at Wal-Mart for $68.00.
They have an AIW 9200 there as well for $129.
They had PCI on one.....but I think it was the plain 9200.
Supposed to be a sharp card related to the old 9100 or the original 8500 chipset from ATI.
Don't expect to scream through DX9 games like Doom or Far Cry.
But games like RTCW, it will rock the house.

Untruest · Mar 23, 2005

lol, thanks for the article, i actually read it a long time ago.

Why is it not easy for a third party company to add a feature to a card such as T&L but it is easy to remove them.

I believe the 9500 was a 9700 but with less pipelines, all they did was removed a feature. WHy can't they add just as easily?

my best bet is this 9100 64mb ddr 55$ US that i found, but i am trying to explore directx 9.0 before i buy it.

cleeve · Mar 23, 2005

The 9500 is a full 9700 GPU with 4 of it's 8 pipelines disabled. but those pipelines are still physically on the GPU. They're just disabled with a different package and BIOS.

Just like a Radeon X800 PRO uses only 12 pipelines, but has 16 pipelines physically on the GPU, like an X800 XT.

________________
Radeon 9700 PRO (o/c 332/345)
AthlonXP 3200+ (Barton 2500+ o/c 400 FSB)
3dMark03: 5,354

Untruest · Mar 23, 2005

oohhh ok, i c what u mean

Well something new just occured to me.

IF a graphics card, decent in specs ( 250mhz/400mhz, 4 pipelines, 128 bit interface ... ) has 16 textures per rendering pass, doesn't that help out a lot in its therotical fill rate? Apperently something like this can consume a lot of memory bandwidth, can could we calculate the amount 16 textures in one pass can take up.

Can someone clarify why the amount of textures in a pass is important again?

I remember the geforce 2 and 3 were similar in therotical fill rate but because the geforce 3 can perform a max 4 textures in one pass, this made it a better card.

cleeve · Mar 23, 2005

Let's use a simplified model, comparing two cards: the Radeon 9100 and 9200.

The 9100 has 4 pipelines with two texture units per pipeline;

whereas the 9200 has 4 pipelines with one texture unit per pipeline.

More textures per pass means that in a situation where multiple textures are being applied to the same pixel (which happens more and more as games get more complex), the videocard can do more work per clock cycle. But this is only applicable in situations where multiple textures are being used.

i.e. if each pixel has only two textures applied, the Radeon 9100 and 9200 will behave the same.

BUT if the pixel has 3 or 4 textures applied, the Radeon 9100 can do the work in one clock cycle.
The 9200 on the other hand would need 2 clock cycles to complete that pixel, because it could only process two textures per pass.

So the 9100 would work twice as fast in that case.

Remember, this is a simplified model, but it gets the point across, and demonstrates why the 8500/9100 series usually beats the 9000/9200 series, even when the 9000/9200 series has a clockspeed advantage.

________________
Radeon 9700 PRO (o/c 332/345)
AthlonXP 3200+ (Barton 2500+ o/c 400 FSB)
3dMark03: 5,354

Untruest · Mar 23, 2005

Now i am kind of confused between a "texture unit per pixel pipeline" and a "texture per rendering pass". I think ur implying that they are the same but i think i've learnt differently.

Lets take the 9100 for example. It has 4 pixel pipelines and 2 textures units per pipeline. Now i have a pixel that needs 8 textures.

My question is, Can the 4 'pixel pipelines' work on that one pixel during the same cycle? Or does each 'pixel pipeline' have different pixels in them and the one that needs 8 textures has to re-enter the pipeline 3 more times for a total of 4 cycles to be textured?

Now the 9100 can do 6 textures per pass, so does that mean it would take two passes to texture our pixel? And how long is one pass? Does it always take that amount of time?

cleeve · Mar 23, 2005

Each pixel pipeline can only work on one pixel per cycle (pass).

If you have four pipelines than can handle 2 textures each, and process a single pixel with 4 textures, it still takes 2 passes regardless of the number of pipelines you have.

More pipelines means you can process more pixels during one clock cycle.

________________
Radeon 9700 PRO (o/c 332/345)
AthlonXP 3200+ (Barton 2500+ o/c 400 FSB)
3dMark03: 5,354

Untruest · Mar 23, 2005

A 5700 can process 16 textures per pass. Do they mean that each pipeline can process 4 texturing units? (There are 4 pixel pipelines on a 5700)

IF this is the case, then wouldn't that be better than a 9100?

cleeve · Mar 23, 2005

The 5700 *is* better than a 9100, although not by that much for PCI cards, because the bus limits the effectiveness of all of them.

On AGP/PCI-e the 5700 is a good deal better than the 9100.

It's the 5200 that sucks.

________________
Radeon 9700 PRO (o/c 332/345)
AthlonXP 3200+ (Barton 2500+ o/c 400 FSB)
3dMark03: 5,354

Untruest · Mar 24, 2005

So am i right about the whole pipeline thing??

On another page, it said something like ; "6 textures per pass with 8 texels per clock" what can u make of it?

cleeve · Mar 24, 2005

A Texel is a textured pixel, so this would indicate that the card could texture 8 pixels with up to 6 textures each per clock cycle.

At the end of the day, this stuff is good to know, but benchmarks is the stuff that counts. You should be studying those more than theory if it's time for a purchase.

________________
Radeon 9700 PRO (o/c 332/345)
AthlonXP 3200+ (Barton 2500+ o/c 400 FSB)
3dMark03: 5,354

Untruest · Mar 24, 2005

I'd say that, that's not too bad.

Because they mention it indirectly, i am not sure if there is 4 pipelines with 2 texturing units or 4 pipelines with 1 texturing unit.

On a site, it said 2 pipelines with 2 texturing units, but i thought that couldn't make sense since the card can process 4 pixels per clock(u need at least 4 pipelines to do that)

How many pipelines and texturing units in each would u say there is.

Calling ALL Video Card EXPERTS!!

Distinguished

Champion

Distinguished

Polypheme

Illustrious

Distinguished

Illustrious

Polypheme

Distinguished

Illustrious

Champion

Illustrious

Champion

Distinguished

Distinguished

Illustrious

Distinguished

Illustrious

Distinguished

Illustrious

Distinguished

Illustrious

Distinguished

Illustrious

Distinguished

Share this page