R3400

jvd

Banned
The major advantage of RS400 over RS350 is the adoption of a real DirectX 9 IGP. Even though RS350 utilizes a Radeon 9100 Pro for IGP, it is still technically DX8.1. RS400 will integrate the Radeon 9600 Non-Pro GPU onboard, which should give Intel's 915G a run for its money. If ATI's roadmaps are to be trusted, this is a bit of a surprise for us. Our original roadmaps in April market RS480 to launch with DX8.1 instead of DX9. Unfortunately, graphics performance is only the secondary function of an IGP chipset, and thus the northbridge must be up to snuff. ATI's memory controller has been its downfall many times before in the past.

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2134

Looks interesting . 9600 gpu on board video would be a very nice jump over the nforce 2 intergrated video and intell intergrated video
 
RS350 was classed as having integrated 9200 - it wasn't until it turned up do you find that there was no VS and half the pipelines. Although there will almost certainly be DX9 functionality, it'd be best to wait to see the details of the implementation before you can consider its performance - having said that, I can't see ATI wanting to let the performance go to Intel, as that would be somewhat of an embarrasment.

Hopefully some more of the HyperZ techniques will translate across as this should be useful when relying in system bandwidths.
 
Cool thanks for the info.

On a side note do u know if ati is planing on moving onto amd chips ?
 
DaveBaumann said:
RS480 was the codename that was bandied around at the AMD conference a while back.

nice hopefully it will be nice.

Wouldn't mind that 3 monitor support that ati was doing
 
Anandtech said:
ATI's memory controller has been its downfall many times before in the past.
Funny they should still be mentioning this. Wasn't it their article that showed ATI performing much better than even Intel's chipsets once a Prescott is thrown in? I don't recall the 925X doing nearly as well with Prescott.

I'd be very surprised if ATI decreases performance from RS350. So long as they don't, they should have a real winner on their hands. Now it's just a matter of when they can roll it out of the door...
 
DaveBaumann said:
RS350 was classed as having integrated 9200 - it wasn't until it turned up do you find that there was no VS and half the pipelines.
Still, its performance was quite close to the 9200 - slower than a 128bit 9200, but faster than the 64bit version. And it supported exactly the same features (not all in hardware though). Maybe it would have been more honest to call it "9200 derived".
However, I don't think it's possible this time to remove 2 pipelines, as the 9600 consists of a single quad pipeline (and if I understand that correctly, it might be non-trivial to break it up). And I doubt ATI wants to engeneer a completely new chip. So my bet would be a 4 pixel pipe chip (this part identical the the 9600), though possibly without (or with 1?) vertex shader. Though maybe you know more about it than I do...
Hopefully some more of the HyperZ techniques will translate across as this should be useful when relying in system bandwidths.
To what of the HyperZ features are you refering? RS350 also retained all HyperZ features that the 9000/9200 had, I doubt it's different this time. But I don't think Hierarchical-Z will make a comeback.
Are there no longer plans for IGPs for Athlon64 systems? There were some rumours that it is a problem because the IGP couldn't access ram directly. AMD told though that it shouldn't be a major problem, but fact is there are NO IGPs at all available for A64 systems today.
 
mczak said:
Are there no longer plans for IGPs for Athlon64 systems? There were some rumours that it is a problem because the IGP couldn't access ram directly. AMD told though that it shouldn't be a major problem, but fact is there are NO IGPs at all available for A64 systems today.

Chipset:

http://www.via.com.tw/en/k8-series/k8m800.jsp

System:

http://tech-report.com/reviews/2004q2/shuttle-sk83g/index.x?pg=1

But let's just say that it's 3D performance is nothing to brag about :)
 
However, I don't think it's possible this time to remove 2 pipelines, as the 9600 consists of a single quad pipeline (and if I understand that correctly, it might be non-trivial to break it up).

This was the same situation with 900/9200 and IGP 9100.

To what of the HyperZ features are you refering? RS350 also retained all HyperZ features that the 9000/9200 had, I doubt it's different this time.

R(v)2x0 has no early Z rejection and the Z compression is less. There is also no colour compression for AA.
 
DaveBaumann said:
However, I don't think it's possible this time to remove 2 pipelines, as the 9600 consists of a single quad pipeline (and if I understand that correctly, it might be non-trivial to break it up).

This was the same situation with 900/9200 and IGP 9100.
I was under the impression though that the 4 pipelines are tied together more closely for this generation. That might just be my perception though :?
IF 2 pipelines can be more or less easily removed, it would actually make a lot of sense - ATI now sells 9550 with a chip clock of only 250Mhz, so cutting two pipes and clocking it a lot higher (something like 450-500Mhz probably should be no problem) should just give the same performance while save a lot of transistors (and should probably still beat the GMA900). It's probably safe to say ATI is not aiming for 9600Pro performance...

To what of the HyperZ features are you refering? RS350 also retained all HyperZ features that the 9000/9200 had, I doubt it's different this time.

R(v)2x0 has no early Z rejection and the Z compression is less. There is also no colour compression for AA.
Ah, forgot about those. Does that stuff use a lot of transistors? If so I'd think especially color compression (no competitor offers fast AA on IGP or even low-end card) is likely to go.
 
mczak said:
If so I'd think especially color compression (no competitor offers fast AA on IGP or even low-end card) is likely to go.

I would actually remove the whole MSAA unit and instead add more shading power. Dunno if that's feasible to do though.
 
DaveBaumann said:
RS350 was classed as having integrated 9200 - it wasn't until it turned up do you find that there was no VS and half the pipelines.

Although that wasn't too surprising (did anyone really think it'd have hardware vertex processing?). I'll be surprised if the integrated 9600 core has more than 2 pipelines or if it has hardware vertex processing (although I won't mind being pleasantly surprised). Bandwidth limitations make more than 2 pipelines a bit redundant anyway (unless the core runs at a very low speed, which it probably won't).
 
ET said:
DaveBaumann said:
RS350 was classed as having integrated 9200 - it wasn't until it turned up do you find that there was no VS and half the pipelines.

Although that wasn't too surprising (did anyone really think it'd have hardware vertex processing?). I'll be surprised if the integrated 9600 core has more than 2 pipelines or if it has hardware vertex processing (although I won't mind being pleasantly surprised). Bandwidth limitations make more than 2 pipelines a bit redundant anyway (unless the core runs at a very low speed, which it probably won't).

Theoretical memory bandwith is also increasing from 6.4 GB/s to 10.66 GB/s.
And when shaders are used more, memory bandwith is less bottleneck.

So IMHO 4 pipelines is a very reasonable for RS400.
 
jvd said:
How big is the current rv350 ? How big would it be on 110nm or 90nm .

I recall the threads about RV350 being a very small die. And the 0.11u RV370 (X300) should be about 20-30% smaller yet, and would be the logical core to slap onto RS400.
 
What is nvidia going to do in the future?

The fx5200 is crap, and I do not see an integrated type video from the nv4x class yet so are they just screwed on integrated video for now?
 
Back
Top