9500 Pro vs. 9600 Pro vs. 5600 Ultra

[cvearl]

I thought this was interresting. It's a reply I made to someone at Rage3D that felt I was bitching about the 9600 Pro being Sub standard compared to a 9500 Pro. Whatcha all think will be the best in this $200 card shootout???

Here's the numbers. They worry me a little. I am afraid that IF the 9600 Pro is slower when it comes down to really stressing the card, I won't be able to reverse my decision to get a 9500 Pro because they will have all but vanished up here in Canada.

Herer's the facts so far... (edited to add Geforce FX 5600 Ultra)

9600 Pro Memory bandwidth = 9.6 GB/s
9500 Pro Memory Bandwidth = 8.6 GB/s
FX 5600 Ultra Memory Bndwth = 11.2 GB/s

9600 Pro Pixel and Texel speed = 1,600 MP/s
9500 Pro Pixel and Texel speed = 2,200 MP/s
FX 5600 Ultra Pix and Tex speed= 1,400 MP/s ??? That Low????

9600 Pro Triangle setup speed = 175 MT/s
9500 Pro Triangle setup speed = 275 MT/s
FX 5600 Ultra Vertices (Is that the same?) = 88

9600 Pro Rendering Piplines = 4
9500 Pro Rendering Piplines = 8
FX 5600 Ultra Rend Piplines = 4

9600 Pro Vertex Shaders = 2
9500 Pro Vertex Shaders = 4
FX 5600Ultra Vert Shaders = 2

Really looks like the final showdown is 9600 Pro vs. FX 5600 Ultra.

As for 9500 vs. 9600; What do you see in these numbers? Does the 12% memory advantage of the 9600 Pro make up for the 40% disadvantage in fill rate and the 50% disadvantage in Rendering pipelines and Vertex Shaders? If with all of this in mind, the 9600 Pro beats a 9500 Pro in stress tests, it reshapes what we have all come to understand about 3D performance and how it is delivered. I have a funny feeling that the 9500 Pro will come out on top though. If so there better still be one somewhere that I can buy. I think the only thing that they can use to truely stress these two ni a comparison will be UT2003 at max details and high res with all the eye candy enabled.

Also to quote nVidia -- The GeForce FX 5600 GPUs deliver 30% more performance at half the price of the GeForce4 Ti 4600. In what cases I wonder... Hmmm....

This is going to be a VERY interresting couple of weeks.

Charles.

[davidy]

I was under the impression that the 9600 pro wasn't really meant to outperform the 9500 pro, but to provide a part in that region that had drastically lower production costs.

[cvearl]

I had a simular reply to me and this was my response...


quote:
--------------------------------------------------------------------------------
Originally posted by bobzdar
One of the key points was a redesigned memory controller (or improved) to optimise it for use at 128bit width. The 9500Pro's memory controller is optimised for 256 bit bus, so it doesn't perform at peak performance that the already smaller bus can offer. I'm thinking this may have a larger effect than some think.
--------------------------------------------------------------------------------



I really hope this is true. Anand explained the design decisions behind the 9600 Pro. The 9500 Pro was making next to no margin for ATI because it was so expensive to produce and they needed a midstream card to complete the family after the release of the 9700 Pro. The 9500 Pro was rushed and they had very little time to cut anyhting out of the R300 design to make it cheaper. This time around they had plenty of time to produce a product that would give them a better return on thier dollar. This means that the 9600 Pro is not so much for "improving the gamers experience in the mainstream market" but more so to increase profits at ATI. In other words, the 9600 Pro release may be more for ATI than it is for consumers. I guess we will see.

Lastly, I beleive that HArdOCP said that the 9600 Pro is essentially a higher clocked 9500 "NON pro" with a slight memory controller enhancement. That sounds sucky beyond reason to me. Aside from the higher clocks, they are identical on paper.

Charles.

[5150 Joker]

Originally posted by cvearl
I had a simular reply to me and this was my response...


quote:
--------------------------------------------------------------------------------
Originally posted by bobzdar
One of the key points was a redesigned memory controller (or improved) to optimise it for use at 128bit width. The 9500Pro's memory controller is optimised for 256 bit bus, so it doesn't perform at peak performance that the already smaller bus can offer. I'm thinking this may have a larger effect than some think.
--------------------------------------------------------------------------------



I really hope this is true. Anand explained the design decisions behind the 9600 Pro. The 9500 Pro was making next to no margin for ATI because it was so expensive to produce and they needed a midstream card to complete the family after the release of the 9700 Pro. The 9500 Pro was rushed and they had very little time to cut anyhting out of the R300 design to make it cheaper. This time around they had plenty of time to produce a product that would give them a better return on thier dollar. This means that the 9600 Pro is not so much for "improving the gamers experience in the mainstream market" but more so to increase profits at ATI. In other words, the 9600 Pro release may be more for ATI than it is for consumers. I guess we will see.

Lastly, I beleive that HArdOCP said that the 9600 Pro is essentially a higher clocked 9500 "NON pro" with a slight memory controller enhancement. That sounds sucky beyond reason to me. Aside from the higher clocks, they are identical on paper.

Charles.

I agree with the assessment that ATi made the 9600 as a way of increasing their gains but I don't think they'd release an inferior part either. Guess we'll soon find out--keep in mind this card should have really good overclocking potential since it's on .13u and it has better compression than the 9800 pro as well as improved hyper-z.

[Dazz]

I think the Radeon 9600Pro will be faster then the FX5600Ultra due to the higher fillrate, if you look back the Radeon 9500 had a 256bit bus yet due to the kapt pipelines it was considerbly slower, so if the fillrate is low would that bandwith even be used?

[c4c]

Originally posted by Dazz
so if the fillrate is low would that bandwith even be used?
Sure:Anti-Aliasing. Multisample AA uses much less fillrate compared to super sampling, but still eats up the bandwidth.


From the specs listed, it seems the 5600 Ultra has about 87.5% of the fillrate of the 9600. However, the 9600 has about 85.7% of the bandwidth of the 5600 Ultra. Just based off this I would suspect the ATI would perform slightly better in normal rendering, but the FX would have a slight speed advantage in AA situations. However, architectural differences and drivers may be what makes either one of these shine. Its all up in the air until we see some hard numbers ;)

One thing about Anti-Aliasing though. Assuming both cards use the same AA as their bigger brothers, then ATI will have superior image quality in straight 4x MSAA. The 5600 could make most of this up with 4XS AA (and in fact do better for alpha textures), but this mode uses a mix of Super Sampling and Multisampling, which will eat up more fillrate (and performance). I hope reviews of both cards will include performance/quality comparisons of the various types of AA..