Any definite or official figure of RV350 Transistors?

[jerry_enCater]

Any definite or official figure of RV350 Transistors?

Somewhere say 60M such as Beyond3D, Tech-report
Somewhere say 75M such as THG

Who know the truth?
Which one is right?

 

[Dave Baumann]

I'm attempting to find out.

 

[Mephisto]

Any definite or official figure of RV350 Transistors?

Somewhere say 60M such as Beyond3D, Tech-report
Somewhere say 75M such as THG

Who know the truth?
Which one is right?

As a general rule, if THG figures differ from others, THG is wrong.

 

[jerry_enCater]

Any definite or official figure of RV350 Transistors?

Somewhere say 60M such as Beyond3D, Tech-report
Somewhere say 75M such as THG

Who know the truth?
Which one is right?

As a general rule, if THG figures differ from others, THG is wrong.

A french website is the same as THG.

 

[Evildeus]

HFR says 76m

 

[MuFu]

I would guess it is closer to 60M from the die size.

MuFu.

 

[Dave Baumann]

Official repsonce from Euro PR is 75M. I'm waiting on another responce, but there may be an explaination. The chip shot in the review was from an A12 sample, which could only go to 375MHz - these current ones may be on a larger die...

 

[Luminescent]

75M actually sounds more realistic. The NV25 contained ~63 million, with four integer pipelines (2 texture units each) and 2 DX8 vertex shaders. Thus, with complete fp support and the same number of pipelines (with added flexibility), it seems more natural for the RV350 to contain more transistors.

 

[Dave Baumann]

Odd. I just noticed that according to the label on the back of the board the chip on that board is the same ASIC/revision number as that found on the chip that only made it to 375MHz, so it appears (if that label is correct) they didn't go back for another spin.

 

[Pete]

I thought nV chips generally had more cache than ATi, so the RV350 could conceivably have more features in about the same transistors as the NV25. I suppose ATi knows what they're saying, tho--unless they're also trying to compete in transistor-counts, too.

 

[Tridam]

75M actually sounds more realistic. The NV25 contained ~63 million, with four integer pipelines (2 texture units each) and 2 DX8 vertex shaders. Thus, with complete fp support and the same number of pipelines (with added flexibility), it seems more natural for the RV350 to contain more transistors.

On the other hand, RV350 has only one texture unit per pipeline -> smaller tex cache, smaller datapath... I'm not sure, but I think NV28 has 2 pixel shader integer units per pipeline. RV350 has 1 FP/FX unit per pipeline. RV350 doesn't have Hierarchical Z...

It's very difficult to find the RV350 transistors count with a comparison to NV28/25.

It makes more sense to do the same with R300. 107 millions – 4 pipelines – 2 geometry units – hierarchical Z, memory controller of 256 bits (instead of 512 bits) and smaller internal datapaths and caches. 107 millions -> 76 millions: - 28%. Do you think that all these elements are only 28% of the R300 transistors count ???? It makes no sense. 60 millions seems more realistic. 107 millions -> 60 millions: -43%.

Don't forget that RV250/280 has 35 millions of transistors. Actually, ATI seems very good at making good chips without a high transistors count. So I think Pete's assumption could be true…


If I'm not wrong, ATI has always stated that RV350 has 60 millions of transistors. It's what they say me many times. It would be very disappointing if ATI were saying 76 millions to have a bigger number than the NV31's 75. I hope that it's not the case

 

[jerry_enCater]

75M actually sounds more realistic. The NV25 contained ~63 million, with four integer pipelines (2 texture units each) and 2 DX8 vertex shaders. Thus, with complete fp support and the same number of pipelines (with added flexibility), it seems more natural for the RV350 to contain more transistors.

On the other hand, RV350 has only one texture unit per pipeline -> smaller tex cache, smaller datapath... I'm not sure, but I think NV28 has 2 pixel shader integer units per pipeline. RV350 has 1 FP/FX unit per pipeline. RV350 doesn't have Hierarchical Z...

It's very difficult to find the RV350 transistors count with a comparison to NV28/25.

It makes more sense to do the same with R300. 107 millions – 4 pipelines – 2 geometry units – hierarchical Z, memory controller of 256 bits (instead of 512 bits) and smaller internal datapaths and caches. 107 millions -> 76 millions: - 28%. Do you think that all these elements are only 28% of the R300 transistors count ???? It makes no sense. 60 millions seems more realistic. 107 millions -> 60 millions: -43%.

Don't forget that RV250/280 has 30-35 millions of transistors. Actually, ATI seems very good at making good chips without a high transistors count. So I think Pete's assumption could be true…


If I'm not wrong, ATI has always stated that RV350 has 60 millions of transistors. It's what they say me many times. It would be very disappointing if ATI were saying 76 millions to have a bigger number than the NV31's 75. I hope that it's not the case

RV250/280 has 30-35 millions of transistors???
What do you know??

 

[Tridam]

RV250/280 has 30-35 millions of transistors???
What do you know??

Sorry, 30 is wrong but 35 is right. Actually, ATI stated 36 millions of transistors for RV250. A little more for RV280 maybe. It's roughly the same for the RV280.

 

[jerry_enCater]

RV250/280 has 30-35 millions of transistors???
What do you know??

Sorry, 30 is wrong but 35 is right. Actually, ATI stated 36 millions of transistors for RV250. A little more for RV280 maybe. It's roughly the same for the RV280.

Can you give me some proof such as official announce,or some reliable sources?
Thank you!

 

[Tridam]

RV250/280 has 30-35 millions of transistors???
What do you know??

Sorry, 30 is wrong but 35 is right. Actually, ATI stated 36 millions of transistors for RV250. A little more for RV280 maybe. It's roughly the same for the RV280.

Can you give me some proof such as official announce,or some reliable sources?
Thank you!

I don't think it's in the official announcement docs. ATI said it to me at the RV250 launch (July 2002) and I write it. I think they said it to many editors so read some articles on the internet Every good site must be talking of 36 millions transistors. Beyond 3D is one of them (http://www.beyond3d.com/reviews/ati/radeon9000pro/index.php?page=page2.inc).

 

[Stefan Payne]

75M actually sounds more realistic. The NV25 contained ~63 million, with four integer pipelines (2 texture units each) and 2 DX8 vertex shaders. Thus, with complete fp support and the same number of pipelines (with added flexibility), it seems more natural for the RV350 to contain more transistors.

but who knows what the NV25 has what it not really needs.
Such as the NSR (the Register Combiners).

 

[Xmas]

but who knows what the NV25 has what it not really needs.
Such as the NSR (the Register Combiners).

The register combiners are the core functional units of the NV25 pipelines.