NVIDIA GF100 & Friends speculation

Deg, I've just read the conclusion there for a (moderate only, true) laugh.

e.g
It's a little bit of a bet on games embracing Tessellation and Ray Tracing but Nvidia hedged their bet by equaling the performance of the GTX-295 without the new technologies factoring in. Play a game with Tessellation and/or Ray Tracing and performance is more than double.

Remembered that 4870 vs GTX 260 review the did a while ago though
 
Yeah, it seems 470/480 is at best equal or a little faster than the 5850/5870, whilst using more power, being hotter, and noisier. The power/noise videos at [H] are completely off-putting just because of the noise alone. At the same time, it's not the fastest card you can buy - if frames are what's important, you can still trump the 480 with a 5970.

All in all, Fermi has turned out to be what I expected - a problematic product that was forced out the door late and compromised, and manages to only match the competition, but at the cost of more money, heat, power, and noise. It's a stopgap product to get Nvidia into the DX11 market, but it's costing them in terms of the larger die and poor yields. Nvidia will have to improve things over the next couple of quarters, either with a respin or improved TSMC process.

There is also no sign of mainstream products to compete with AMDs full range, while AMD may be able to bring out it's next generation by Q3/Q4, just as an improved Fermi and it's mainstream versions begin to come through in quantity. With lowered prices, factory overclocks, and two gig cards from AMD, competition will get even harder for 470/480.
 
Conclusions




At it appears Charlie was right about the Dirt 2 demo DX9 issue on the 4x0 cards. In actual gameplay the 5870 is faster than the 480.

If you take a closer look at benches they dont all run with same settings. For example in Metro 2033 GTX480 has 4xAA and 5870 has 'AAA' yet both perform about the same. Obviously GTX480 is doing a lot more.

And for DIRT2 benches all other sites point to better perfomance on GTX480 and again at Hardocp they got different settings. Seems a quite biased site that dont bench on equal terms.
 
So far it looks exactly where most people thought it would be: between the 5870 and 5970. So not a hilarious farce but rather a pretty solid entry.

I wonder how ATI will respond to this.
 
Mixed thoughts. On the good side, it (GTX480) came up faster than the 5870 quite consistently. 13% faster by Tech Power Up. Given all the problems, that's quite an accomplishment.

On the bad there are a lot of problems. The low memory clocks turned out to be a bad controller by Nvidia, major problem there. The other big issue is actually less texture fillrate than GTX285. Both point to a rather borked/unbalanced design. Almost as if the GPGPU parts costs Nvidia so much they weren't able to put enough texture specs in and keep the card at reasonable power/size. The third big issue is power, which the card really struggles with already and is way higher than the AMD parts (320W 480 vs 212W 5870 on furmark). Oh yeah, and it is equaled or slightly exceeded by the 5870 on Crysis.

The alleged 5890 should be able to catch up to that +13% or close. But then if rumors are true Nvidia will have the 512 SP part (6.7% more shaders) and maybe higher clocks with B1, so it might be right back to 10-15% faster.

Overall this is not going to put any pressure on AMD to lower prices. If anything they could probably raise them a few bucks.

I have to wonder how texture/bandwidth limited the card is, and if Nvidia missed a good chance to blow AMD out of the water with a part that performed 2X GTX285. By the same token I often think "what if" for AMD, what if they didn't stick so stringently to their small die strategy and stick a few more quads or whatever in Evergreen, they might easily be the absolute fastest (single GPU) right now in which case Nvidia would have been really screwed.
 
Last edited by a moderator:
Pretty disappointing when viewed solely in the context of gaming. Not really surprising given the route they took with double precision though.

Bouncing Zabaglione Bros. said:
if frames are what's important, you can still trump the 480 with a 5970.
I think you mean GTX 480's in SLI... you know, if frames are important.
 
Hmm, I'd like to see a comparison of Extreme Tessellation of the GTX480 vs the 5970.

AFAICT, HardwareCanucks did it in Normal mode where the 5970 still beats the GTX480.

[edit]And thus the HC server dies[/edit]
 
Last edited by a moderator:
lol... sarcasm save fail.

/passes out fixed die

Bouncing Zabaglione Bros. said:
Probably still cheaper when you take into account the server case and multiple PSU's you'll need to put them in when looking at running two 480s!
No fair, you're changing the rules.... frames are important, not cost or heat.
 
What is this?

http://www.nvnews.net/vbulletin/showpost.php?p=2219145&postcount=1490

Accelerated Jittered Sampling - Improving Anti-Aliased shadows
You guys must recognize this, you play a game, have 8xAA on and everything looks nice and dandy. Then you look at a shadow and notice it's all blocky and messed up. Here's where a new sampling method kicks in, it's called Accelerated Jittered Sampling. Now the geek explanation (and really I had to look this up) is this: Jittered Sampling is a stirred process in which values are sampled equally over a rectilinear subspace. The exact position of the respective sample in each sub rectangle is thereby varied randomly.

Take a look at the first example below (lower left image). You'll notice the weird blocky shadows. This is in 3DMark Vantage with AA levels turned up. So in layman's terms, what does it do then? Well, that banding is removed and replaced by noise. The result are much more smooth non-blocky shadows.

Let's have a peek at a scene again, where now Accelerated Jittered Sampling is applied:

imageviewa.jpg
imageview2.jpg


So there you go, to tackle this issue GF100 has Accelerated Jittering Sampling which improves the quality of AA shadows and sure, will likely also bring some additional AA performance.
 
This "TDP: 250W" card is consuming more power than the 300w dual-gpu cards most places I've seen and like 120w more than 5870, so I guess it's "250W - as long as you get a really good handpicked sample" (like the one @ guru3d)
 
There's a hint of something special in the Anandtech review, though: minimum framerates. Check this out:
http://www.anandtech.com/video/showdoc.aspx?i=3783&p=9

For the average framerates, in these benchmarks the 480 GTX is ~10% faster than the HD 5870, and quite a bit slower than the 5970. Ho hum, nothing special here.

But look at the minimum framerates!
The 4x0 cards just wipe the floor with the competition! The 470 GTX bests the 5870, and the 480 GTX is trading blows with the 5970! This means, to me, that despite the average framerates, the 4x0 cards, at least for Crysis: Warhead, are producing much better gameplay.

Now, the question is: is this unique to Crysis: Warhead? Or is this a pattern that repeats with other games as well? Is it something that only happens in newer games? Sadly, Anand doesn't help us here, as he only tested the minimum framerates in this one game. Did any other reviewers look at the minimums in a wider variety of games? I know Kyle over at HardOCP did, but it's difficult to tell where the cards stand due to how he tests.
 
Back
Top