NVIDIA GF100 & Friends speculation

[trinibwoy]

They released two demos that were pretty impressive IMHO... the Mecha thing and the depth of field one. Both are more impressive from a technical standpoint than anything, but that's the point, right?

Yep, but I was thinking of something more comprehensive like they've done in the past. Also, where's the tessellation stuff? I remember seeing at least one demo of some adventurer guy but don't know if it was ever released.

I can imagine tessellation being faster on Fermi, but I'm not convinced that anything other than a synthetic case that is heavily over-tessellated is going to demonstrate it, but we'll see. For everything else I just can't see Fermi being significantly faster, and I can foresee it being slower in some areas.

Sure seems that way. But while everyone is caught up with Cypress comparisons I really want to see how it stacks up against GT200 because the current numbers indicate lower architectural efficiency than that beast. Quite a feat if you were to believe Jawed :smile:

Of course I have no numbers for Fermi, but I can vouch for the impressiveness of Cypress *especially* in future-looking workloads. That said, I'll be super-happy if Fermi blows me away in even a couple areas!

What kind of workloads are those? Haven't heard any noise on the topic so far, not even from AMD.

 

[jaredpace]

what memory is it using then. The memory information posted here sugests 800mhz but there is no 800mhz just 1.2ghz memory undervolted and down clocked to 800mhz.

I think hes posting a response to you either making a typo, or assuming the 470 has 1.5gb vram. I suspect the memory is samsung brand K4G10325FE-HC04 .4ns 5gbps, just like what's on the 5800 cards.

 

[Silus]

Sure seems that way. But while everyone is caught up with Cypress comparisons I really want to see how it stacks up against GT200 because the current numbers indicate lower architectural efficiency than that beast. Quite a feat if you were to believe Jawed :smile:

The GTX 480 being faster than the HD 5870, doesn't quite make this NVIDIA's R600, but if the performance difference over GT200 is really reflected in these supposedly legit benchmarks, it's still a huge failure. GT200 over G80/G92 was much better than this...on all counts, even with the low hot clock frequencies that GT200 had when compared to G80 and especially G92.

 

[NathansFortune]

Yep, 480 will be in reviewer hands and shipping...... How many releases and reveals can we do again? Does shipping 5 to newegg count as released?

I think for Nv even 2 shipped to a retailer will count as launched!

GTX480 is definitely going to be in short supply until B1 is here and they get yields a bit fixed. GTX470 will be in better supply depending on what the SP count is. If it is 384 then it could be readily available, but at 448 NV could even have a shortage there. Maybe a 384 SP part will make up the GTX460 if they release one.

Either way it all seems too big and expensive to be really competitive with ATi in pricing terms and the current leaked benchmarks aren't positive enough to suggest a massive performance delta to justify a higher price. If the leaks are from a 384 SP chip then I think Nv are in the clear, but a 448 chip would spell concern.

 

[GZ007]

I think hes posting a response to you either making a typo, or assuming the 470 has 1.5gb vram. I suspect the memory is samsung brand K4G10325FE-HC04 .4ns 5gbps, just like what's on the 5800 cards.

I think they will use the low voltage 800MHz gddr5 if the card will have 800MHz. If they would use the normal 5gbps ram than u could increase your bandwith to 200 GB/s without a big deal and that wouldnt help to much the 480gtx.

 

[Ailuros]

I think they will use the low voltage 800MHz gddr5 if the card will have 800MHz. If they would use the normal 5gbps ram than u could increase your bandwith to 200 GB/s without a big deal and that wouldnt help to much the 480gtx.

The primary reason IMO for clocking the memory at 800MHz is power consumption (which chimes in well with the rumored ~220W TDP for the GTX470). Even if the boards would have 5gbps GGDR5 I wouldn't go as far to push the memory frequency up to 1.25GHz and not worry that it'll blow up heh.

 

[fehu]

http://www.fudzilla.com/content/view/17953/1/

 

[Skinner]

http://www.fudzilla.com/content/view/17953/1/

Without AA.

 

[WeiT.235]

Old story again. The Unigine 1.1 in the Youtube is irrelevant to the benchmark shown later in the same video. IMHO, it just misled you guys.

It is considered as the complete version of result demonstrated in nv's Fermi architecture deep dive. Check the 60-120 sec.
Thanks Migi06

Notice the 27 FPS of 5870? It' coincides with the score of 5870 at 1920x1200 4x AA 16x AF with tessellation enable in the Unigine bench by PCI, which I believed they probably got the card of 470 in hand. In their bench, the 470 scored 29 FPS at the same setting.

Heise.de didn't detail the setting of Unigine in the leaked bench, but it scored the 29 FPS, 27FPS at 4xAA for 470 and 5870 respectively, surprisingly same as the results of PCI. If they got the completely different source, it is quite safe to say that both bench should be in the same setting, given CPU is nor so significantly important as 3DMark in such DX11 intensive benchmark. PCI should deserve the credit this time.

Unigine benchmark
4xAA [fps]: 29/22/27 (GTX470/HD5850/HD5870)
8xAA [fps]: 20/19/23 (GTX470/HD5850/HD5870)

So all of these seem to confirm the Unigine result of 5870 is the genuine in the show of Fermi deep dive. If it makes sense, and nv is believed to not change the setting for gtx480 immediately after benching the 5870 or change the specs of 480, the 43 FPS of 480 given by nv will be probably benched at 1920x1200 4x AA 16x AF with tessellation on, IMHO. 1.6x faster than 5870...

 

[SimBy]

I wonder what happens when you turn on AA & AF. It all points to not much advantage in real world apps that don't do insane amount of tess. And I doubt any game will use that much tess anytime soon. We'll see what Dirt 2 and AvP benchmarks tell us. But my guess is GTX480 is barely faster in those games with max settings.

Non the less, for extreme amounts of tess GTX480 has clear advantage, trouble is, the rest of arch is lacking. So ahead of it's time is the right conclusion?

 

[Mindfury]

So all of these seem to confirm the Unigine result of 5870 is the genuine in the show of Fermi deep dive. If it makes sense, and nv is believed to not change the setting for gtx480 immediately after benching the 5870 or change the specs of 480, the 43 FPS of 480 given by nv will be probably real, IMHO. 1.6x faster than 5870...

480--43fps
470--29fps

I can't believe GTX480 is nearly 50% faster than GTX470，unless NV push GTX480 clock over 800Mhz....

 

[WeiT.235]

480--43fps
470--29fps

I can't believe GTX480 is nearly 50% faster than GTX470.....

I just realized it...

 

[SimBy]

480--43fps
470--29fps

I can't believe GTX480 is nearly 50% faster than GTX470.....

I thought one is with AA and the other is without.

 

[Jawed]

Neither the red nor the green lines on the pair of graphs matches. Additionally, the graph from the YouTube video shows both cards with lower minima. e.g. HD5870 dips to about 16 as opposed to 20 in the older graph.

Anyway, GTX480 should have ~60% more bandwidth than GTX470, so it seems plausible that the gap is huge.

Now that we know that tessellation on ATI is hardly setup limited, attention turns to DS which in theory is chewing through a huge amount of data - i.e. DS could be bandwidth bound on NVidia.

What's causing it to be so slow on ATI is a puzzle.

Jawed

 

[Mindfury]

http://www.fudzilla.com/content/view/17952/1/

If you have been waiting for Fermi, Geforce GTX 480 and GTX 470 and you really want to spend some 600€ / $600 for a high end graphics card, you better be ready to buy it on March 26th.

Despite the fact that Nvidia will have a massive volume launch with thousands of cards available in Europe and USA, they might easily sell out in matter of hours.

People close to Nvidia believe Fermi cards will sell like hot cakes and that there will be huge demand for them as soon as they launch. The reason is quite obvious a single GPU Geforce GTX 480 will end up faster than any Radeon HD 5870 card and in some new games, the advantage will be significant. In some games Radeon might end up faster or very close, but in the big picture Nvidia’s big chip is going to win.

Just be ready to order your card on the first day as they might sell out easily, or be ready to wait a week or two for a second batch.

 

[rpg.314]

Despite the fact that Nvidia will have a massive volume launch with thousands of cards available in Europe and USA, they might easily sell out in matter of hours.

That's hardly inconsistent with ~5k-10k launch volume being speculated.

 

[WeiT.235]

http://www.fudzilla.com/content/view/17952/1/

If it is true, I can't see any reason to spend a 600 euro card which is just marginally faster 5870, unless you are crazy for a bigger key chain several years later...

 

[Malo]

http://www.fudzilla.com/content/view/17952/1/

If you translate the actual text, basically they're saying there will only be a short supply of cards (thousands, not tens of thousands, coincides with the only 5000 available at launch rumor), and for $600 you will get a card that is at least equal, if a little faster than a $400 card, except for some corner cases?

And how is "thousands of cards" considered a massive launch? They're just trying to build hype. Of course they will sell like hotcakes and there will be high demand. The first release with limited numbers will always sell to all the crazy enthusiasts which don't care how much it costs, they must have the latest and greatest. And high demand is a given since once they sell the 5000 or so, there won't be anymore for awhile!

Ignore stuff like this, just PR crap. I just want to see some real reviews.

 

[Bouncing Zabaglione Bros.]

If it is true, I can't see any reason to spend a 600 euro card which is just marginally faster 5870, unless you are crazy for a bigger key chain several years later...

If you're going to spend that kind of money, why not get a 5970 or 5990 and get something even faster again?

 

[leoneazzurro]

That's hardly inconsistent with ~5k-10k launch volume being speculated.

5-10k ARE thousands

IIRC, at the ATI 5000 launch, the rumors spoke about "tens of thousands", I wonder if this is the difference we could expect in the availability of Fermi cards...

EDIT: I see no that Malo also wrote the same things...