NVIDIA G84: Architecture and Video Performance

Hey I got the right die size. First time ever.

Jawed

 

Congrats.

Now he tells us!

 

Interesting to see how G84 video playback performs given that it's now arguably as important as gaming performance, depending of course on the intended use of the card.

We found G84 to be a bit less than ace in the gaming department when compared to the past generation of products. Maybe a little way down the line, after some pricing and SKU shuffling, things might be different.

 

Looks like NV had made some tweaking in the Z/Depth occlusion part in G84, judging by those skyrocketing coefficients:



Now, if G80 had this... with its 64 INT8 sampled texels per clock.

 

the 2nd MUL of G86(8500GT) does work with Rel 158.X but not work with Rel 100.

http://www.pcinlife.com/article/graphics/2007-04-18/1176863400d349_7.html

 

It would be sad if the MUL was somehow broken in G80. Does B3D have a 8500GT to verify this result with?

 

Either I must have missed something incredibly obvious or I'm really out of touch with reality : Does or should the above belong in the article? I mean, it's fine to speculate in forums and such but I honestly don't see any point in comparing (and that is essentially how I read it, a comparison) the actual hardware to prior speculations about it. A few words in the article about what was previously speculated wouldn't actually be too out of place but the above don't sit quite so well with me -- the way it sounded to me, it's like the hardware ain't as good as what you had speculated on prior to this article and therefore it isn't as good as you expected... based on speculations. Very strange IMHO.

Just MHO.

 

Hopefully I made it clear that the speculation was ours and ours alone, internally. We go through the process for everything that's upcoming as a matter of course, to be as on the ball as possible when it comes to the final analysis.

The balance of features/performance/area is a fine one at times, and it's that which I wanted to put across. It's a pretty crucial part of the G84 puzzle, actually, which is why I went with that angle to set the piece up and then tie it up at the end.

 

There is often a fine line between speculation and historical context, depending on who's doing it and how schooled/versed they are in the latter. In other words, not all speculation is created equal. For instance, was the "speculation" above informed by the historical context of 6600GT and 7600GT? I'd say it was.

 

Oh no, I didn't mean to say the speculations weren't B3D's (although B3D's own speculation must somehow be influenced by the speculations in its forums) or that speculations can't ever exist in such articles (in fact, I said it wouldn't be too out of place). I was just saying I would've preferred different phrases instead of what I quoted above (especially the second para).

My point being it just doesn't sound right, nor logically reasonable, to me for a reviewer to be somewhat disappointed with a piece of hardware because it didn't live up to speculations (while mentioning the speculations in the review) -- it's interesting to note that the last page of the article started off mentioning speculations before progressing onto actual performance and competing parts... IMHO it would've made a big difference the other way around for a "Conclusion" page. A straight-forward actual performance-based opinion while considering other competing parts would have been what I personally preferred and, to be fair, this is indeed what I read and preferred in the rest of that last page of the article -- for the reviewer to opine that the G84 may not be an excellent, or even good, $199 DX10 part based on its performance as well as taking into account existing competing parts (even DX9 ones) would've been sufficient for me. Somehow, I read both of the quoted paras (especially the second para, again) as throwing unnecessary "negative spin" on a part that I think (but ain't sure) NVIDIA would not have liked to have been compared to what anyone speculated it to be. I think there must be some reasonable responsibility to primarily depict or review a product for what it is, not for what it was expected to be.

I just think a product should be reviewed based on its actual merits, not on expectations or speculations, both of which existed at both the start and the end of the article -- it was like the article was written with your speculation at the front and back of your mind and what existed in between (the actual product's performance, price-performance, its competing products) was not the priority.

But hey, B3D has every right, of course, and this is just an opinion that I hope some take as constructive. It would be interesting to see if all future major hardware reviews here start and end like this one

 

If only it were so simple. Hardware analysis (at least for me) is about more than just a snapshot-based approach. At the risk of sounding twee or up my own backside, there's a bigger picture to consider when appropriate. Given we can't always afford to revisit every product down the line to see if it lived up to any pre/leaked and launch-day promise (since that's such a big part of the picture here), if there's informed speculation to be made that says something like, "well, if they'd done this instead, and it's what we expected them to do, then we think it might have stood up a bit better overall", then it seems prudent to me that we try and make it. Hardware vendors don't always get it right (and neither do we).

Stand up and be counted and all that. It's arguably a great thing that we're proved wrong somewhere down the line, too, so we learn something new.

And no, the way the piece was setup and concluded certainly isn't a template for every other that'll follow.

 

Well, numbers certainly don't lie.

We're not talking about the value of a company's shares based on its history of products.

Personally, I think most IHVs are clever enough to make informed (and certainly very important) decisions such as to release a product like the G84 as it stands -- no, IHVs don't always get it right but we're not talking about a GFFX here.

Wishing for a product to have been different would be nice if backed up by theoretical performance data and not a logical assumption that it would've simply been a better product. Now, speculating if an extra, say, $20 would be justified if the product had a bit more of something would have been interesting...

[edit]BTW, I view "hardware analysis" to be distinctively different to a "product review". Maybe that's the problem I have with this article...

 

Well this is going to make for interesting reviews going forward. Will B3D be disapointed if R600 doesn't run at 1 GHz core and isn't at least 40% faster than G80, like so many expected?

 

Were any of those "so many" us, as published in formal content? For instance, if it isn't 512-bit, which we offered on the front page, I'd certainly expect it to be mentioned in the review.

 

Ultimately, this your site and you can (largely) write whatever you want in your reviews. I just think you are losing some credibility by comparing a product to unsubstanciated speculation on it. Paper tigers and all. There are many valid points to make, and most reviewers (including yourselves) have made them. I don't think this one is a particularly valid one, that's all.

 

Well, duly noted (and talked about at our staff meeting earlier in some depth). Like I said, it's not a tactic to be wheeled out constantly, and in hindsight some tiny changes to the wording here and there would have delivered things a bit better and I misjudged things, but it is what it is.

Note that the speculation wasn't completely unsubstantiated either, and we do try our best to get the pre-analysis as right as possble. Some of our early thoughts on the chip before it showed up were formed after conversations with NVIDIAns, whatever that's worth.

 

Future performance gap?

So, basically, informed speculation was that the 8600 series would have 1/2 the renderers of the 8800 core, and instead was created with 1/4.

My magic 8-ball sees an 8700 series in the murky future when NVidia needs a new toy to annoy AMD after it tries to split the gap of the 8600-8800.

 

I wish people would put the common names as well
I have no idea what a G71 or a RV570 is ?