reported 9600 clock rates and prices

D

Dave H

Regular
Apparently we're getting:

9600 Pro (128MB): 4x1, 400/300 - $200
9600 128MB: 4x1, 325/200 - $170
9600 64MB: 4x1, 325/200 - $150

:? The Pro part looks alright, but 9500 Pro will probably outperform it even if it does have all the new 9800 core enhancements. And the NP clocks...128 bit wide 200 MHz DDR?! A bit pricey for that, IMO. And again I can't imagine a 9500 NP won't smoke it.

Should add that [H] is reporting 9600 has the "DX9++" :)roll:), i.e. F-buffer support, so the other 9800 enhancements are likely IMO.

Well...if they can hold those prices they should have some pretty nice margins. What were the rumored clocks on NV31 again?
 
Dave H said:
Apparently we're getting:

9600 Pro (128MB): 4x1, 400/300 - $200
9600 128MB: 4x1, 325/200 - $170
9600 64MB: 4x1, 325/200 - $150

:? The Pro part looks alright, but 9500 Pro will probably outperform it even if it does have all the new 9800 core enhancements. And the NP clocks...128 bit wide 200 MHz DDR?! A bit pricey for that, IMO. And again I can't imagine a 9500 NP won't smoke it.

Should add that [H] is reporting 9600 has the "DX9++" :)roll:), i.e. F-buffer support, so the other 9800 enhancements are likely IMO.

Well...if they can hold those prices they should have some pretty nice margins. What were the rumored clocks on NV31 again?
Click to expand...

well

ATis presentations that they sent out are a bit flawed it seems
but anyways
DX9++, SMartshader 2.1 is a part of the RV350 spec.
but the presentation ATi sent out doesn't mention anything about the F-buffer when it comes to the RV350

I dunno..

it does however clearly state Hyper Z III and not III+ for the RV350
 
If the 9600 Pro is slower than the 9500 Pro (looks like it might be) it's going to be the same 9000/8500 thing all over again.
 
The 9600 PRO has more bandwidth, and ATI told me that in most cases it see's the 9500 PRO out because of that. There actually some discussion as to whether its 400/350 and not 400/300 as well.

The thing that astonishes me about the RV350 is the size of the core. The one they showed at London had no shim around the chip, so if you turned it on its side and looked under the heatsink you could see the core - I'd estimate it to be about 5mm wide, really very small.
 
Interesting. So I guess "DX9++" only means FP support for 3d textures and the like? I thought it was a reference to "DX9+" being Nvidia's term for increased length shaders. Sounds even more poorly conceived than C++!

(Not that RV350 would really benefit from the F-buffer, just that I dislike the naming convention.)
 
Thanks for the prices... looks like the 9600 is going to be a really cheap card. Bit of overclocking, it could get up to 9500 speeds.
 
ATI are not really serious about "DX9++", its really just a jab at NVIDIA's marketting. They said that there is no such thing as "DX9+", as you may have heard (i.e. from NVIDIA), there's only DX9 and you are either compliant or not -- but if you want to go around quoting that then list ours as DX9++ because here a list of features that we support and the competition doesn't. (It should be noted that PS/VS 2.0 Extended is actually documented separately from PS/VS 2.0 though).
 
The prices are from Ante P's review at nordichardware, and the review at hardocp has the same ones. I think these are estimated prices at this point, though, not final.

And I wouldn't necessarily count on being able to OC the 9600 NP to 9500 NP speeds, given that the issue is the memory and not the core clock. What I mean is, it's a safe be the core will be overclockable because it's the same chip as in the 9600 Pro. But the DRAM is going to be plain 200 MHz DDR, so it's not at all a given that it will OC up near 275 MHz.
 
DaveBaumann said:
RV350 is a development of R300. There is no F-Buffer in RV350.
Click to expand...

Oh well... seems you´re right. There´s a typo in the R9600 papers saying it has SmartShader 2.1 but the specsheet says 2.0 (but HZIII+ and SV2.1)

Lars
 
There were several reasons I was hoping it was not a 4 pipeline part :-? (yes, even besides not wanting to have to hide from Joe... :p ). It seems to make good business sense to have the RV350 replace the R300 if it can, but in shader performance it just seems like it is going to be lacking with the info we have.

I do find transistor counting results interesting (watch out! I'm doing math before breakfast again! :oops: ):

R300-RV350 (107-75)=~32 million transistors for 4 full FP pipelines and 2 vertex shaders.. :!: . I would have expected more return from dropping those items

R300-RV250 (107-30)=~77 million transistors for 3 vertex shaders, 4 full FP pipelines, and FP functionality instead of PS 1.4 functionality for 4 pipelines.

Do I have the figures that are pretty close to the actual transistor counts?

If so, the comparison of 77 to 75 above is very confusing for me here. The first semi-consistent theory that comes to mind is N-patch hardware support is present on both the RV350 and the R300, takes up significant chip real-estate, and is broken (atleast on the R300).

Other thoughts involve wondering about the complexity required by the change from HyperZ-II to HyperZ-III for the RV250 to the R300 (main difference seems to be color compression), cache size changes, and the improved Aniso/Trilinear.

But, sick, no breakfast, and having just woken up, I may be a bit lost in the woods at the moment.
 
DaveBaumann said:
The thing that astonishes me about the RV350 is the size of the core. The one they showed at London had no shim around the chip, so if you turned it on its side and looked under the heatsink you could see the core - I'd estimate it to be about 5mm wide, really very small.
Click to expand...

I find that size hard to believe. Assuming it were square (no guarantee), thats only 25 sq mm. Knowing the size of our part, and its composition, and its process size, and comparing that to the reported 70M transistors, I'd guess that it (the RV350) would have to be at least 50-75 sqmm.
 
As far as I know, exact frequencies (core/mem) for these products has not been annouced yet. The only thing I know is that we've said that the 9600Pro would have a core "around" 400 MHz. Our partners might of pre-announced some clocks, but those could change.

The RV350 core is larger than 25 mm^2, but it is very small for what it can do. Compared to a 9500pro, well it's hard to say; given the higher clocks, it will have higher fillrates and will win a lot of benchmarks over the 9500pro. It's higher memory speed and efficiency will help too. But for pure shader execution, the 9500pro has the advantage of a 8x3=24 shader instructions per cycle, vs the 4x3=12 shader instructions of the 9600. Even with the speed advantage of 30%, the 9500pro might win some benchmarks. It's actually hard to say (reviews and benchmarks will give people a better idea). Both are excellent products :)
 
RussSchultz said:
DaveBaumann said:
I'd estimate it to be about 5mm wide, really very small.
Click to expand...

I find that size hard to believe. Assuming it were square (no guarantee), I'd guess that it (the RV350) would have to be at least 50-75 sqmm.
Click to expand...

Well, let's see now. 50 sqmm would work out to about 7 mm wide (again, assuming it's square). I think that a difference of 2 mm is well within the margin of error of what Dave's eyes can probably estimate while squinting to see between the heatsink and the pcb.

Let's try not to be too hard on him for that. :D

I assume the point he's trying to make is that it's really tiny.
 
tamattack said:
sireric said:
The RV350 core is larger than 25 mm^2, but it is very small for what it can do.
Click to expand...

Any chance you could let us know what the die size actually is? :LOL:
Click to expand...

If I hated my job and enjoyed being prosecuted, I'm sure I would tell you!
 
You must log in or register to reply here.

Similar threads

S
ATI to launch Radeon 9500 and 9700 on the 18th
4 5 6
Replies
110
Views
15K
Pete
P
Back
Top