**G92 versus RV570 racing target same goal: 10K 3DMark06
**
NOVEMBER WILL BE the slaughter place between ATI and Nvidia once again, and the swords will go into fiery battle. This time, the performance difference will be minimal, since both chips are targeting the very same price point, and offering very similar, if not identical features.
Certain Graphzilla minions kept on hyping the G92 as the best thing since sliced bread and a new performance leader, while G92 is now marking a mainstream chip. Mainstream chip that targets 8800GTX in 3DMark06. Performance is estimated at above 9500 3DMarks (2006, not 2005), depending on name of the partner that Nvidia is talking too. We received multitude of different numbers (all very similar), and Nvidia hopes that this policy will get the leaks out. Guys, rather keep the promises your sales reps are telling partners. They’re all operating on razor-sharp margins anyway.
The name of the G92_290 chip will be the GeForce 8700 GT/GTS, while RV670 should come to market as Radeon HD 2700GT/XT. Fastest G92 board is being developed under codename P393A01, and it will retail for 249 (we expect overclocked versions to be priced at 279 and 289 USD), and cheapest variants will go for 199-229 USD.
In short, G92 is simply half a G80, 65nm manufactured chip sitting on a regular 33x33mm BGA packaging, with a lot of things happening under the hood. This is not just your cut-down G80 chip, but rather a combination of higher-clocked part (over 800 MHz), and higher fill-rate than 8800GTS, even though the number of shader units is lower. Nvidia did not cut the number of ROP units, so expect hellish pixel and Texel fill-rate. RV670 took the same receipt as G92 compared to G80: cut the number of R600 units in half and you get RV670. 160 Superscalar shaders consisted out of 32 fatties and 128 regular scalars, blazing high GPU clock… like we said, RV670/G92 are siblings.
Both chips are gunning for 512MB GDDR3 memory by Qimonda. Part number for reference boards on both sides call for 1.0ns ones (new HYB chips), which surprised us at best. We did not expect that development of boards is like seeing siblings develop, but carrying different hearts. Both are manufactured at the same place, and same manufacturing process, though.
Have you had any doubt, DAAMIT is pitching for very same performance that Nvidia is achieving. 10.000 3DMarks is the whole grail, but 9500 should be doable. Performance estimates were based with Kentsfield machine, so this may change if you overclock the CPU at above 3.0 GHz level (QX6800 works at 2.93 GHz).
What is the codename of NV55 now (one that was alleged to us to be the G92 - with nice pictures of a 2-slot monster), remains to be seen. All we know is that this chip will go head to head against R680, both due to be released in Q1 2008. It will be interesting to see what will be the key amount of memory in January - R680 with 512/1024/2048 versus G80 refresh with 768 or 1536 MB. If Qimonda can deliver GDDR5, even a memory type could change, from current Samsung GDDR4 liebling.
Funny how cycle is the same, yet different. We would only like to know will Nvidia reset to ATI numbers once they reach 10000 mark after the GeForce name. So far, it is funny to see original 8800 (ATi FireGL, still sitting in my Museum) and compare it to today’s 8800. µ
P.S. Nvidia is spotting S/PDIF connector on the reference boards, meaning DVI Audio or HDMI adapter is quite a valuable feature to have. DisplayPort is also mucho importando, but not on the RV670 nor G92 reference boards. µ
Link->November 3D War brings ATI and Nvidia siblings