AMD release 690-series core logic

[Rys]

<a href="http://forum.beyond3d.com/articles/amd690launch/"><img border="1" src="http://forum.beyond3d.com/articles/amd690launch/focus.png" align="right" width="75" height="75"></a>Today sees AMD release their long-awaited 690-series core logic, created by pairing AMD RS690 and AMD SB600, a combo which results initially in 690G and 690V SKUs.

We've scoped out the core logic specs and major details, and we've taken a quick look at an ASUS mainboard that hosts 690G too. The big question is: in early 2007, is AMD 690 and all that entails enough to carve out some marketshare in the IGP space? 30+ design wins seems to hint at an emphatic yes.

Check out our <a href="http://forum.beyond3d.com/articles/amd690launch/">initial look at AMD 690G</a>, and there's also a Hexist look at EQS's AMD 690G-sporting <a href="http://www.hexus.net/content/item.php?item=7972&page=3">AB1S-RS690MKM</a> for a retail board perspective.

[Geo]

Wheee. For the nostalgic, the first time I remember running across references to RS690: http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=2222 Whereupon we all went crazy for months over what "Kaleidoscope" was (answer: Avivo). Tho by that roadmappery we're pretty much a year late with RS690. . .

[BrynS]

There seems to be some discrepancy regarding the number of vertex shaders -- 2VS seems reasonable given that RV410 had 6 VS, but both of the reviews I've read so far (Firingsquad (http://www.firingsquad.com/hardware/amd_690_chipset_performance/page3.asp) & Tech Report (http://techreport.com/reviews/2007q1/amd-690g/index.x?pg=9)) report that 690G has no HW VS, instead the CPU will handle vertex processing -- a mix up in the reviewers guide?

Also, does the AVIVO portion of 690G represent ATI's previously mentioned UVD moniker, or is that new logic due in the next chipset/IGP (790G?)?

[Geo]

So far as I know, no UVD in there. The timelines wouldn't work out for that anyway, when you look at how long RS690 was in development.

[Rys]

There seems to be some discrepancy regarding the number of vertex shaders -- 2VS seems reasonable given that RV410 had 6 VS, but both of the reviews I've read so far (Firingsquad (http://www.firingsquad.com/hardware/amd_690_chipset_performance/page3.asp) & Tech Report (http://techreport.com/reviews/2007q1/amd-690g/index.x?pg=9)) report that 690G has no HW VS, instead the CPU will handle vertex processing -- a mix up in the reviewers guide?

Also, does the AVIVO portion of 690G represent ATI's previously mentioned UVD moniker, or is that new logic due in the next chipset/IGP (790G?)?
FS and TR are right, RS690 has no vertex processors. Fixed that misunderstanding. Also fixed are the HDMI revision (it's 1.3, not 1.2), the fact that 690V is single-link DVI only and the number of PCIe lanes in total (24, with 16 for PEG, 4 for general I/O and 4 for the ALINK connection to SB600).

While I was checking D3D9 caps earlier I should have spotted the VS config, sorry about that.

[BrynS]

No problem. I suppose in a low-cost IGP design where performance isn't a critical factor, the lack of HW VS shouldn't be a problem, especially when it (690G) produces favourable performance against existing IGPs. Perhaps I'm misremembering the original R4xx discussion back in '04, but I thought its VS logic was relatively petite, so for 690G to not have half (3 or maybe just 2) the VS units of RV410 seems a bit unexpected.

[bloodbob]

So the chip itself isn't even DX9.0 because it doesn't have vertex shaders ( you could call a TNT2 a DX10 card if you included software rendering ). Really this shouldn't be certified as Vista Basic ready cause doesn't that require DX9.0 capable card which well this chip aint.

[Dave Baumann]

It passes WHQL as a DX9 chip - everything that needs to be exposed as DX9 is exposed. Its WinEI scores are sufficient to recieve premium logo and enable the premium 3D desktop.

[Zaphod]

So the chip itself isn't even DX9.0 because it doesn't have vertex shaders ( you could call a TNT2 a DX10 card if you included software rendering ).Really? I was under the impression that they abandoned any requirement for hardware vertex shaders many moons ago. The MSDN DX9 FAQ (http://msdn2.microsoft.com/en-us/library/bb219721.aspx) even keeps boasting their "fully featured software vertex-processing pipeline (including support for custom vertex shaders)" which has "several different code paths, depending on the processor type" and delivers "a surprisingly high level of performance".

[Geo]

Didn't Intel do "DX9" that way too? At any rate, surely what matters is 1). Do valid DX9 instructions get performed per spec and 2). Performance relative to the competition market-segment-wise. From the benchies I saw this morning, #2 seems to be covered.