ATI RV740 review/preview

[homerdog]

http://vr-zone.com/forums/418725/hd-4770-rv740-performance-and-price.html

Blimey, that last slide says it all.

They're calling G92 an "old architecture/process shrink" in comparison to the "new architecture" RV670. I know I know, RV670 was more than just a die shrink, but so was G92! And they conveniently don't mention G92b in there. Sorry for the OT. :smile:

80W TDP for the 4770, so the 4750 should be alright without an external power connector. OEM boxes need some love too

 

[Arty]

They're calling G92 an "old architecture/process shrink" in comparison to the "new architecture" RV670. I know I know, RV670 was more than just a die shrink, but so was G92! And they conveniently don't mention G92b in there. Sorry for the OT. :smile:

RV670 brought in DX10.1, so they could probably get away with it.

80W TDP for the 4770, so the 4750 should be alright without an external power connector. OEM boxes need some love too

Fudzilla says 4750 will also have an 6 pin.

 

[Jawed]

Could you please PM me with the references you use since i am in the fourth year
of electronics engineering and our graduation project will be a fully working Graphics card!

From scratch?

Vector graphics?

So i need as much references i can to educate myself and my team regarding this stuff..

Start by asking specific questions here:

http://forum.beyond3d.com/forumdisplay.php?f=43

AMD's and NVidia's websites have piles of technical stuff for you to look at. Both have technical discussion forums, too.

Games Developer, Graphics Hardware, Eurographics, Gamefest and Siggraph conferences are all good places to start. You'll drown!

AMD has in-depth hardware information in the form of register-level guides and architecture programming guides. These guys find it pretty interesting:

http://phoronix.com/forums/forumdisplay.php?f=43

I'm sure there's a wiki for that stuff somewhere.

Patent searching is a good sport, too:

http://gb.espacenet.com/search97cgi/s97_cgi.exe?Action=FormGen&Template=gb/en/advanced.hts

that's reasonably approachable (as opposed to the US site).

Jawed

 

[Fusion]

Oh, thanks alot for your effort, Much appreciated...

No, not from scratch, we have relations with technical people inside AMD/ATI so they could provide us with the GPU itself (probably the RV730 chip)..

We'll build the system (Memory Interface, PCB Design, Signal & Power Integrity issues,
tweaks for some known design problems,...etc) as other companies like Sapphire, Gigabyte
and other brands actually do..

Hope I cleared my point..

 

[Jawed]

Ooh that sounds like a lot of fun - reminds me of a school project I did interfacing a 6522 to an AY-3-8910 sound chip to bring sound to games on my 6502 computer

This is my era (can't find scans of UK magazines):

http://www.atarimagazines.com/compute/index/

Where's that :drool: smiley?

Don't forget the cooler!

Jawed

 

[DegustatoR]

Blimey, that last slide says it all.

Why stop at the last slide? Let's check the other ones:
- the 2nd, 'world's first 40nm chip'. So i guess all these Altera chips doesn't exist, huh? And while Intel is using 45nm, they are using it for nearly a year now.
- the 5th, DX10.1. Ultra Quality shadows in SCS are the same between DX10 and DX10.1. The difference is in alpha textures AA which is slightly better under 10.1.
- the 7th. Gigaflops are pointless.
- aaand the 8th. The latest example of innovation from AMD would be RV790, am i right?
If that's not FUD then i don't know what FUD is.

 

[crystall]

If that's not FUD then i don't know what FUD is.

That's not FUD, that's marketing.

 

[Kaotik]

Why stop at the last slide? Let's check the other ones:
- the 2nd, 'world's first 40nm chip'. So i guess all these Altera chips doesn't exist, huh? And while Intel is using 45nm, they are using it for nearly a year now.
- the 5th, DX10.1. Ultra Quality shadows in SCS are the same between DX10 and DX10.1. The difference is in alpha textures AA which is slightly better under 10.1.
- the 7th. Gigaflops are pointless.
- aaand the 8th. The latest example of innovation from AMD would be RV790, am i right?
If that's not FUD then i don't know what FUD is.

You asure about the 5th slide stuff? IIRC those Ultra Quality shadows, illustrated in the pic, aren't available in DX10 in SCS, only DX10.1

 

[Jawed]

The latest example of innovation from AMD would be RV790, am i right?

Yeah, the cap ring is like "eh, if it's that good why wasn't it there in the first place?"

Jawed

 

[CarstenS]

You asure about the 5th slide stuff? IIRC those Ultra Quality shadows, illustrated in the pic, aren't available in DX10 in SCS, only DX10.1

I think the latest patch for SCS brought some change wrt to the shadows. The picture AMD paints surely was true a few months ago.

 

[DegustatoR]

You asure about the 5th slide stuff? IIRC those Ultra Quality shadows, illustrated in the pic, aren't available in DX10 in SCS, only DX10.1

Yes, i'm sure.
DX10.1 in SCS brings better alpha textures AA and some performance improvements. Shadows are the same between DX10 and DX10.1.

I think the latest patch for SCS brought some change wrt to the shadows. The picture AMD paints surely was true a few months ago.

Not really. Patch 1.5.07 added Ultra Quality shadows for both DX10 and DX10.1 renderers. Previously there was only High Quality (and there was no DX10.1 renderer at all).

 

[Kaotik]

Yes, i'm sure.
DX10.1 in SCS brings better alpha textures AA and some performance improvements. Shadows are the same between DX10 and DX10.1.


Not really. Patch 1.5.07 added Ultra Quality shadows for both DX10 and DX10.1 renderers. Previously there was only High Quality (and there was no DX10.1 renderer at all).

Ther article you linked doesn't show a single shadow comparison picture between 10 and 10.1.
Can't understand the text though, but here's what the translator (systran) says:

It is interesting to also note that if in version 1.5.04 and in the DX9- regime of [patcha] 1.5.07 we note no difference between the quality of [otrisovki] of shadows by [videokartami] Of geForce and Radeon, then in the new Ultra- regime she is located: on [videokarte] Of geForce is observed clearly expressed [banding] of the washed away boundary of shadow (penumbra, “penumbra”), which on [videokarte] Of radeon is furthermore smoothed by something of the type of [dizeringa]. I.e., Radeon HD 4800 X2 on the maximum quality in S[CHN] ensures not only the higher productivity (practically two times), but also the more high quality of the filtration of shadows! For the interest we opened use DX10.1 to Radeon, but the quality of smearing the edge of shadow this did not influence - i.e., this superiority on the quality over GeForce is caused not by support DX10.1, but by something other.

Now considering the patchnotes say the new shadows take advantage of DX10.1 if present, it seems pretty confirmed to me that there is a quality difference

 

[DegustatoR]

Ther article you linked doesn't show a single shadow comparison picture between 10 and 10.1.

Really?
http://www.ixbt.com/video/itogi-video/test/0811_scsdx101/scs_gfgtx280_1507_dx10_4xaa_1.png
http://www.ixbt.com/video/itogi-video/test/0811_scsdx101/scs_rhd4800_1507_dx10_4xaa_1.png
It's exactly the same quality for DX10.1 on 4870.

Can't understand the text though, but here's what the translator (systran) says

That's true (i wrote this text, btw), but it's hardly what AMD's showing in their presentation because the exactly same difference in shadow quality rendering is there even when you're not using DX10.1 on the Radeons.
It's either some kind of different penumbra filtering in DX10 renderer for GF and R or a bug (optimization?) in GF or R drivers, DX10.1 features has nothing to do with this.

 

[CarstenS]

Not really. Patch 1.5.07 added Ultra Quality shadows for both DX10 and DX10.1 renderers. Previously there was only High Quality (and there was no DX10.1 renderer at all).

Yes, really. First, there was a different codepath with an 8x8 Shadow-Map-Filter IIRC, which was exklusive to DirectX 10.1-cards. I don't know though, if that has been removed completely with Patch 1.5.08 or if it was changed so that DX10.0-hw would render the same shadows.

 

[DegustatoR]

Yes, really. First, there was a different codepath with an 8x8 Shadow-Map-Filter IIRC, which was exklusive to DirectX 10.1-cards.

Well, i'll recheck it, but my previous findings showed that this filtering may actually be working even in DX10 on DX10.1-cards.
So it's more like "AMD GPUs codepath" then DX10.1 codepath.

 

[CarstenS]

Maybe, but remember that Patch 1.5.08 was also the patch that dramatically improved performance on geforce-cards and also, that the orginal stalker was a TWIMTBP-game.

Assassin's Creed all over again?

 

[Kaotik]

Really?
http://www.ixbt.com/video/itogi-video/test/0811_scsdx101/scs_gfgtx280_1507_dx10_4xaa_1.png
http://www.ixbt.com/video/itogi-video/test/0811_scsdx101/scs_rhd4800_1507_dx10_4xaa_1.png
It's exactly the same quality for DX10.1 on 4870.

Well, it says DX10 for both, not DX10 and DX10.1?

That's true (i wrote this text, btw), but it's hardly what AMD's showing in their presentation because the exactly same difference in shadow quality rendering is there even when you're not using DX10.1 on the Radeons.
It's either some kind of different penumbra filtering in DX10 renderer for GF and R or a bug (optimization?) in GF or R drivers, DX10.1 features has nothing to do with this.

You're sure about the DX10.1 having nothing to do with it? I guess we'll take your word for it as I don't think there's any shader analyzation for them nor developer comments?

Anyway, I never meant the differnece would be same as shown in the AMD pics, just that there's a difference

 

[CarstenS]

Search for this document on AMDs GDC-website:
Stalker - Clear Sky A DirectX showcase.pps

Starting from Page 38, there's the shadowing approach for SCS explained (quite detailed). They also mention, why they can afford to use a 8x8 PCF Filter on 10.1-HW:

 

[Jawed]

http://developer.amd.com/gpu_assets/Stalker - Clear Sky A DirectX showcase.pps

Jawed

 

[Jawed]

It seems to me that this is possible using the NVAPI, anyway, so there shouldn't in the end be a difference between AMD and NVidia in the case of shadow filtering.

Jawed