Early 470 & 480 benches

[Benetanegia]

Kristin Kreuk is my future wife, she just doesn't know it yet.

No... beep... no, no... error... error... beep beep! not computable.

She is mine. beep... mine! miiiiiiiiine!! beeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeeep....

GTX480 = $650 retail. Prices will most likely be higher due to lack of supply and obviously demand will be there for a new Nvidia product. GTX470 = $450-$500.

Any sources? I have not heard but rumors. None of them had any better credibility than the one in my link.

 

[erocker]

Any sources? I have not heard but rumors. None of them had any better credibility than the one in my link.

Purely fabricated in my head. Sorry, I should of put a "My guess on pricing is..." in front of it. But I am somewhat basing it off the same sources. It could be around $600. If it has equal performance of my two 5850's for the same price, I'll get one. I'd love to get some mobo slots back.

 

[johnnyfiive]

Thats frickin' ODB in your BC2 sig!

 

[bobzilla2009]

450 euros is $615 btw, if they are $450/450e then nvidia are screwing over Europe by nearly 40% extra. the hd5870 has a RRP of £279 (or 309e) so even if it releases at 450e, it's still over 140e more.

 

[KainXS]

GTX480 dropped to 480 cores.

http://www.brightsideofnews.com/news/2010/3/15/geforce-gtx-480-has-480-cores2c-aibs-confirm.aspx

GTX470 ?

exactly double that of a 280 now, ^^

More Importantly

Also, we received news about targeted prices for the parts, and they look quite competitive against six-month old ATI 5800 Series: according to the benchmarks leaked some time ago, GeForce GTX 470 will beat out HD 5850 and in some cases, even HD 5870 - for the price of HD 5850 - we were told that the ballpark figure is $299-349, while GTX 480 will run between $449-499.

 

[bobzilla2009]

it'll be nice if those prices are right, can anyone say a $300 hd5870 and a $200 hd5850? in the end if nvidia try to compete with current amd pricing they're going to be in for a big fight after 6 months of amd premium pricing to rake in the money. Which is awesome for us.

 

[KainXS]

If they're right I would be amazed because it would be the cheapest high end , new gen, nvidia card in years by far.

and at that price it would be pretty tempting for the 470

 

[Benetanegia]

if they are $450/450e then nvidia are screwing over Europe by nearly 40% extra.

And Ati. And Intel. And Asus. And and and.

That's the norm. Look at my previous link. The HD5870 sells for over 400 euros. Every card has a similar value as those in $ in the US.

UK prices are somewhat different and are better than EU prices right now. The reason is that the pound is very low right now. 1.1 Euros now and it used to be 1.6-1.8 not so long ago. The consecuence is funny, because UK prices can't change (market is not so flexible, explain customers why prices are suddenly so much higher...) and they could not do 1 dollar = 1 pound in the first place, because noone would buy anything. So the values had a lower numerical value than in EU and continue being the same lower numerical value than in euros now. What has changed is the pound-euro relation with the funny effect of making UK prices lower than EU prices (for people in the EU outside UK, of course).

In the example you posted:

price before = 279 * 1.5 = 418 euros
actual EU price back then= 400 euros
price now = 279*1.1 = 307 euros
actual EU price now = still 400 euros conclusion I might travel to UK on vacations soon...

 

[KainXS]

just a question here, if that BSN article is correct and the card is 480SP(meaning they disabled a SM) wouldn't that make the GTX480 only have 60TMU's and 15 tessellators since each SM is linked to 4TMU's and 1 tessellator.

 

[Benetanegia]

just a question here, if that BSN article is correct and the card is 480SP(meaning they disabled a SM) wouldn't that make the GTX480 only have 60TMU's and 15 tesselators since each SM is linked to 4TMU's and 1 tesselator

Yeah, it looks like, but who knows. I'm still skeptical about how they are disabling only 1 and 3 SMs. That makes one or three of the GPCs have one less SM, but the chip still has the same triangle setup and rasterizers (some with less SMs to work with). idk it's weird. I always thought that the second SKU would have an entire GPC disabled, it always made more sense to me, but what do I know? Nothing so...

And for all we know they could be disabling individual SPs. 2 SPs per SM in the GTX480 and 4 in the GTX470. That makes much more sense to me tbh and would make it a much better yielder (being able to disable defective SPs no matter where they are), which would explain the discrepancy on what different people say about Fermi yields. One could hear that Fermi has an average of x defects (let's say 9) on different parts of the chip and assume, based on that and on the typical harvesting methods, that 9 SMs (288 SPs) had been rendered useless (rendering the chip useless too), when in fact only 9 SPs were rendered useless and could still classify as a GTX480, if you know what I mean.

Pure speculation I know, but I still have vivid the claim that Nvidia made about Fermi being much more flexible and scalable and looking from the perspective of them disabling SMs, I don't know were that flexibility can be found. They'd be doing the same as in the past, where's that flexibility?

 

[HalfAHertz]

...
Also the GTX4xx series are far cheaper to produce than GTX2xx series, despite all the crap that you might have heard. That means that GTX480 costs less than GTX285 to produce, so it could essentially release for the same price and have better margins (of course it won't because they want to recover R&D, but a $600 price tag is not necessary at all). What I just said is not 100% true at the moment, but it will soon become 100% true right after launch due to higher volumes, and prices are chosen based on how much it will cost to produce when the card has been released, not on pre-release costs.

How so? It's the same size(bigger than?) as the 285 and produced on a crappier manufacturing process. Even if the PCB is less complex (only 384-bit bus), it needs expensive vrms and cooling - everything points to it being more expensive. I'm pretty sure the 285 has pretty high margins atm, and that the 480 won't come near it in the forceable future.
I'm not saying you're wrong but without solid proof your word is as good as my old gym socks.

 

[Benetanegia]

How so? It's the same size(bigger than?) as the 285 and produced on a crappier manufacturing process. Even if the PCB is less complex (only 384-bit bus), everything points to it being more expensive.

According to Nvidia yields are almost on par with GT200/GT200b yields which were never good to begin with. So even if the chip costs more, it's not much more, nowhere near as much as has been said. And I'm talking about the costs that the cards will have in the next months, which is what really matters: they can stretch profits per card now and expect better sales and profits later, yields will improve too. On the other hand, like you said 40nm node is a crappy one, but that doesn't mean exactly what most people directly assume. As a person that has worked for a mid-to-large scale bussiness buying/selling for the firm, I can tell you without a shade of doubt, that 40nm being a crappy node is not something that goes without consecuence. Don't believe ANY claim about how much a waffer costs, I can tell you that at least Nvidia (as TSMC's bigger customer by far) gets them cheap. Probably cheaper than 55nm at the time of GTX285 release or 65nm at GTX280 release.

That's only part of the costs. According to what they are saying out in the wild, the PCB is much much simpler (6 layers), the power circuitry is simpler and you can tell that just by looking at it and there are far fewer and cheaper components on the PCB. Also there's no NVIO chip which costs quite a bit too.

 

[HalfAHertz]

True, I totally forgot about the nvio chip. Well I hope you're right!

 

[Steevo]

So the benchmarks posted are all fakes, the Heaven one I have reposted in my thread is a fake. 470 is sporting 1280Mb.


And it looks like Nvidia is having TDP, clock, and yeild issues. Only a 480 CUDA core part. So cmon green team, throw us some spin on how this is a advantage and will cause the ultimate fail of ATI.

 

[[I.R.A]_FBi]

trojan horse?

 

[Zubasa]

trojan horse?

Only this time the horse goes on fire because of its TDP

 

[yogurt_21]

So the benchmarks posted are all fakes, the Heaven one I have reposted in my thread is a fake. 470 is sporting 1280Mb.


And it looks like Nvidia is having TDP, clock, and yeild issues. Only a 480 CUDA core part. So cmon green team, throw us some spin on how this is a advantage and will cause the ultimate fail of ATI.

see I curious as to why they lowered the gtx480 to 480sp but left the gtx470 at 448? doesn't that leave them rather close?

It smells like gtx275 vs gtx285 all over again.

 

[erocker]

I read the 512 shader parts are being used for Tesla workstation graphics.

 

[MarcusFoX]

Good thread here.... we'll see when this card is released, if the overclocking head room is even anywere near the the 5xxx series card, it will interesting to see how muct heat this baby produces.

 

[jamsbong]

No doubt this card is faster eventhough it is only a small margin. I think as dev uses more tessellation like the heaven benchmark. Fermi will shine a bit more. However, by the time Dev. utilizes tessellation, there is probably a Fermi2 or Fermi2.5. Moreover, I doubt ATI will be standing still, I expect them to respond with a HD6xxx when Fermi2 comes around.

One thing really worries me is the 1.8amp fan! I imagine it will be loud as a vacuum cleaner or a screaming jet! Moreover, there 1st page screenshot showing a clear 79% fan load at idle (or just after after a benchmark) and that is not promising.

It is kinda strange that the GDDR5 ram in FERMI is running at such low speed. I suspect either the power supply is not clean (not enough phase) or the GPU consumes too much power and there is not enough power for the RAM. Anyway, I'm kinda surprise how little the performance increase is with current games, consider the fact there is more fillrate, more bandwidth and more core processors compared to Evergreen.

 

[bobzilla2009]

79% fan speed to knock it down to 34C, nice. with my hd5870 and it's far less powerful (but still bloody loud at 71%) fan, my card goes to 21C @ 79% (even when overvolted to 1.350V, at stock it goes lower).

If that isn't a driver bug, this card is not going to suitable to anyone who either likes to hear or doesn't have in ear headphones.

 

[EarthDog]

So the benchmarks posted are all fakes, the Heaven one I have reposted in my thread is a fake. 470 is sporting 1280Mb.


And it looks like Nvidia is having TDP, clock, and yeild issues. Only a 480 CUDA core part. So cmon green team, throw us some spin on how this is a advantage and will cause the ultimate fail of ATI.

So are these real or fake????

 

[Steevo]

Mine are real, the Nvidia ones are a fake. Confirmation from Nvidia seems to show 1280MB on the 470, not 1200 like it shows, but you can edit the BIOS of any card to say anything you want.

 

[EarthDog]

The heaven bench you posted is real and is not fermi, thats your 58xx, right? The first post results for Fermi are fake...correct?

 

[KainXS]

based on what I am seeing ima say this is what the specs will be

GTX480
Core Clock: 650
Shader Clock: 1300
Pixel Fill Rate: 31200
Texture Fill Rate: 39000
M Controllers/ROP Partitions: 6
Tesselators 15
Rops: 48
TMUs: 60


GTX470
Core Clock 625
Shader Clock: 1250
Pixel Fill Rate: 25000
Texture Fill Rate: 35000
M Controllers/ROP Partitions: 5
Tesselators 14
Rops: 40
TMUs: 56

GTX470 only has 40 Rops because the ROP Partitions as far as the Nvidia Whitepaper states are coupled and automatically scale.

The ROP units are organized in six groups of eight. Each group is serviced by a 64-bit memory controller.

Turning off one memory controller turns off 8 rops.
The GTX480 only has 15 Tess and 60TMU's because as BSN states, the card has a disabled SM to deal with heat.