The Fermi Thread - Part 3

Printable View

Show 100 post(s) from this thread on one page

03-21-2010, 07:23 PM
highoctane

Quote:

Originally Posted by LordEC911

But the original statement still stands, 40nm wafers are more expensive than 65/55nm wafers and yes, by a few thousand dollars per wafer.

There you go again, where are you coming up with this range, was it in the silicon quarterly or something.... :confused:

I call shenanigans, the wafers are more expensive by only $23.99 per wafer and you can take that to the bank, for real...
03-21-2010, 07:35 PM
Amorphous
A huge cost is R&D. You worry how much the wafers and heatsinks cost, but the real point to consider is R&D. The Fermi architecture, just like the G80 architecture, was a long time in development and expensive to produce. But consider a few points:
- The architecture is being made into three seperate high-end products, the cheapest of which is the GeForce part. Quadro parts are often priced twice (or more) the price of GeForce parts, and Tesla parts even more.
- The architecture is extremely modular. Scaling up for the next generation after GTX 480 is almost as simple as CTRL+V. Scaling down is just as easy.
- Longevity. Fermi has it. It's an extremely advanced architecture with a ton of features built around the best part of DX11, tessellation.
I haven't seen anything that's made me believe that NVIDIA will be selling GPUs at a loss. Consider that even with HD5800 out performing NVIDIA parts, NVIDIA is gaining more of the market, selling a ton of GPUs and making a lot of money. Even IF they were selling GF100 at a loss, they aren't in AMD's financial shoes.

Besides, why should we care if they make or lose money on a product so long as the price and performance are competitive?

Amorphous
03-21-2010, 07:40 PM
weston

Quote:

Originally Posted by Amorphous

Besides, why should we care if they make or lose money on a product so long as the price and performance are competitive?

Amorphous

you had me until this...

nVidia has only gained market share in the low end and OEM markets, laptop makers are dropping them due to issues over the last 2 years, and OEMs are tired of renaming parts for new models. nVidia will be fine if the first round of fermi doesn't do well, but they better hope they get some money from somewhere. No business can afford to sell all it's parts at a loss for very long.
03-21-2010, 07:49 PM
LordEC911

Quote:

Originally Posted by highoctane

There you go again, where are you coming up with this range, was it in the silicon quarterly or something.... :confused:

I call shenanigans, the wafers are more expensive by only $23.99 per wafer and you can take that to the bank, for real...

The difference is, I never asked you to believe me...

Quote:

Originally Posted by Amorphous

A huge cost is R&D. You worry how much the wafers and heatsinks cost, but the real point to consider is R&D. The Fermi architecture, just like the G80 architecture, was a long time in development and expensive to produce. But consider a few points:

I haven't seen anything that's made me believe that NVIDIA will be selling GPUs at a loss. Consider that even with HD5800 out performing NVIDIA parts, NVIDIA is gaining more of the market, selling a ton of GPUs and making a lot of money. Even IF they were selling GF100 at a loss, they aren't in AMD's financial shoes.

Amorphous

Right... Lots of money is spent on R&D.
Lots of money is also spent on operational expenses.

Ummm... pretty sure Nvidia lost marketshare the last couple of quarters.
03-21-2010, 08:21 PM
highoctane

Quote:

Originally Posted by LordEC911

The difference is, I never asked you to believe me...

No difference, I still think its a contrived figure...

Quote:

Originally Posted by LordEC911

Ummm... pretty sure Nvidia lost marketshare the last couple of quarters.

Quote:

Intel was the leader in Q4'09, elevated by Atom sales for netbooks, as well as strong growth in the desktop segment. AMD gained in the notebook integrated segment, but lost some market share in discrete in both the desktop and notebook segments due to constraints in 40nm supply. Nvidia picked up a little share overall. Nvidia's increases came primarily in desktop discretes, while slipping in desktop and notebook integrated.

http://jonpeddie.com/press-releases/...er-also-beats/
03-21-2010, 08:25 PM
Sn0wm@n

but they have less market shares then last year
03-21-2010, 08:47 PM
Solus Corvus
Quote:
Originally Posted by Amorphous

The architecture is extremely modular. Scaling up for the next generation after GTX 480 is almost as simple as CTRL+V. Scaling down is just as easy.
Can we then analogize scaling down as using the DEL key? :D

Scaling down for the mid and low range would be a good idea. They'd get a much higher yield and they could fill those market segments with nice Fermi arch chips instead of evolved G80 arch chips.

Quote:

Besides, why should we care if they make or lose money on a product so long as the price and performance are competitive?

Well, I for one also consider the market and social impact of my purchase. I'd like strong competition in the GPU market. What happens now in the early days of gpu computing could affect how the market looks for decade. But we all will have our own reasons for buying a product.
03-21-2010, 09:01 PM
LordEC911

Quote:

Originally Posted by Sn0wm@n

but they have less market shares then last year

Yeah, is the table wrong?
It shows a decrease for everyone but Intel...
03-21-2010, 09:06 PM
highoctane

You would have to be a subscriber to get the full article.
03-21-2010, 11:11 PM
ajaidev

Quote:

Originally Posted by Solus Corvus

Can we then analogize scaling down as using the DEL key? :D

Scaling down for the mid and low range would be a good idea. They'd get a much higher yield and they could fill those market segments with nice Fermi arch chips instead of evolved G80 arch chips.

Well, I for one also consider the market and social impact of my purchase. I'd like strong competition in the GPU market. What happens now in the early days of gpu computing could affect how the market looks for decade. But we all will have our own reasons for buying a product.

That would DEL the whole die what you need to use is the backspace key :)

I hope its scalable otherwise we will have Gx3xx in low-mid end based on GT200 and G92 cores.

My reason for buying is funding the gpu makers till holodecks come :D and play some games in the mean time.
03-22-2010, 12:12 AM
mao5

5970,5870,5870 2GB 5850 VS. 480, 470 final scores:

http://bbs.pczilla.net/attachments/m...ecd61398ab.png
03-22-2010, 12:22 AM
insurgent

Total scores for what though?
03-22-2010, 12:24 AM
Gamekiller

Quote:

Originally Posted by insurgent

Total scores for what though?

Troll points it seems. I thought someone kept these threads clean? :rolleyes:
03-22-2010, 12:31 AM
K404

Quote:

Originally Posted by mao5

5970,5870,5870 2GB 5850 VS. 480, 470 final scores:

http://bbs.pczilla.net/attachments/m...63b9b97a63.jpg

WTH is that? I've never seen so little context or relevant annotation in a graph :p:

Thats the graph makers fault BTW, not yours. Unless you made it.........
03-22-2010, 12:42 AM
Coldon

Quote:

Originally Posted by mao5

5970,5870,5870 2GB 5850 VS. 480, 470 final scores:

http://bbs.pczilla.net/attachments/m...63b9b97a63.jpg

I are can be like troll?

No but seriously, why even bother posting something like that, give me 10 minutes in excel and i can make you a more believable but completely fabricated graph.

To all the ATI fanboys in this thread, fermi might not be insanely faster or cheaper than the 5xxx series but at least it can run two screens :p: (so do i get my troll points yet?)
03-22-2010, 12:47 AM
mindfury

mao5=P2MM，a infamous chinese ATI fanboy...

Sometime he maybe has some real inside source,sometime he just post some bull:banana::banana::banana::banana:...
03-22-2010, 12:47 AM
mao5

Quote:

Originally Posted by Coldon

I are can be like troll?

No but seriously, why even bother posting something like that, give me 10 minutes in excel and i can make you a more believable but completely fabricated graph.

To all the ATI fanboys in this thread, fermi might not be insanely faster or cheaper than the 5xxx series but at least it can run two screens :p: (so do i get my troll points yet?)

oh, buddy, can you complain to the AIC?
03-22-2010, 12:52 AM
mindfury

Quote:

Originally Posted by Coldon

I are can be like troll?
To all the ATI fanboys in this thread, fermi might not be insanely faster or cheaper than the 5xxx series but at least it can run two screens (so do i get my troll points yet?)

http://images2.sina.com/newscenter/c.../526582840.jpg

http://images2.sina.com/newscenter/c...2162868449.jpg
03-22-2010, 12:54 AM
eric66

Quote:

Originally Posted by Coldon

I are can be like troll?

No but seriously, why even bother posting something like that, give me 10 minutes in excel and i can make you a more believable but completely fabricated graph.

To all the ATI fanboys in this thread, fermi might not be insanely faster or cheaper than the 5xxx series but at least it can run two screens :p: (so do i get my troll points yet?)

it runs two screens problem was solved long time ago wow what a hater
03-22-2010, 12:55 AM
insurgent

He specifically wants two monitors, three just don't cut it xD
03-22-2010, 01:00 AM
mindfury

Quote:

Originally Posted by mao5

whatever, I don't care your s hit

You forgot your “Radeon 7" bull:banana::banana::banana::banana:....

Quote:

Originally Posted by mao5

buddy, my source told me you can see "Radeon 7 Series" on the computer screen when you power on the Evergreen machine, ok? not the signature pic from sb else, thx.

http://forum.beyond3d.com/showpost.p...postcount=1761
03-22-2010, 01:05 AM
zalbard

Quote:

Originally Posted by Sn0wm@n

anyone want to take a wild guess as to when tsmc will have their 28nm process ready for nvidia's shrinked fermi ?????

ill start and say Q4 2011

6-9 months tops, as for most cards, I guess.

Quote:

Originally Posted by HelixPC

on a serious note, seeing how the memory will make minimal performance increase for most people, im sure nvidia chose lower frequency gddr5 to keep cost down, thats my guess.

The cost difference can't be significant. Must be something else. Heat, perhaps...

Quote:

Originally Posted by Chickenfeed

Thanks for that. Further supports the possibility of a 480Ultra down the road.

No reason for them not to release it once they can.

Quote:

Originally Posted by fellix_bg

The metallic cover has definitely something to do with the heat dissipation.

It shouldn't make a significant difference in heat dissipation.
It's probably built that way to allow a bigger radiator while keeping an acceptable width.

Quote:

Originally Posted by orangekiwii

Wheres my fermi 512sp at 750 mhz?

Wasn't that the intended? I want to see benchmarks of that not the 'dud' 480sp version

750MHz is too high for a reference card.
And 512sp version is supposedly coming later...

Quote:

Originally Posted by To(V)bo Co(V)bo

Can somebody tell me how much profit nvidia and there AIB's will make off of each card at these current prices?

Nvidia isnt able to price these cards much more considering the prices ATI is selling parts for. Nvidia is bound and shackled by their lack luster performance. Ati's selling prices have a death grip on nvidias bottom line.:brick:

I'm sure they will make some profit. On both Geforce and (of course) Tesla cards. The price they are selling them for is bound to be higher than the manufacturing cost.

Quote:

Originally Posted by ***Deimos***

WHAT IS THE BIG DEAL?

512 vs 480. That's **only** 6% worth.

End of the World because nvidia ships a card with less than all shaders?
Broken card? Give me a break. 8800GT was **only** 112 out of 128 shaders, and one of best selling cards.

It's the only high end card with disabled parts I can think of.

Quote:

Originally Posted by annihilat0r

http://translate.google.com/translat...tm&sl=tr&tl=en

According to this, Nvidia might have already sold huge numbers of Fermi GPUs as Tesla cards to the Chinese government. Which means a probability of very low availability of GeForce cards.

There would be low availability in any case.
And we can't complain, Tesla is their primary market... At least they'll make some serious $$$! :yepp:

Quote:

Originally Posted by saaya

breaking news! ati MIGHT have sold even higher numbers of rv870 gpus to the chinese government!

source

LOL, nice source! :rofl:
I doubt they sold many... But they sold a lot of 4870x2s, I think.

Quote:

Originally Posted by Chruschef

This actually makes a lot of sense to me, GPUs are capable of hacking much faster than CPUs because of the mathematical nature of GPUs . . . and the Chinese govt. is allegedly quite fond of hacking things.

What's up with all the hate? :shrug:

Quote:

Originally Posted by M.Beier

Probably posted somewhere here, but there could be 2 editions of the gtx 480 coming out.

One with 512 shaders as well as the 480 edition. Its probably going to be an ultra with an ultra like price. Hopefully they bump the clocks up a bit like they have done in the past.

Unlikely IMO... Or at least right now. But there should be definitely something similar coming out a bit later.

Quote:

Originally Posted by dduckquack

i didnt know the chinese government played a lot of 3dmark

:rofl:

Quote:

Originally Posted by BeepBeep2

There are probably more computers in Seoul, South Korea than there are in North Korea as a whole...

Anyway, why is nvidia selling Fermi to the chinese gov't.?

Because Tesla cards cost a lot more than Geforce cards. And Tesla is the primary reason for creating Fermi arch (GPGPU, etc). So they are just following their plan.

Quote:

Originally Posted by onethreehill

GTX 480 and GTX 470 specifications
http://www.hardware-infos.com/news.php?news=3484
Translated
http://translate.google.com/translat...84&sl=de&tl=en

Looks about right, I guess.

Quote:

Originally Posted by xdan

And also it seems that until 6 april the stock will be very low...almost inexistent...and the cards will be overpriced by sellers in some countrys like Romania. :down:

This is typical for high end cards these days, though. Gotta blame TSMC, I suppose...

Quote:

Originally Posted by annihilat0r

Performance wise, I am not waiting for any surprises. Prices are good, TDP is not as bad as it was "hyped" to be, these cards seem to be somewhat decent.

Good definition! :up:

Quote:

Originally Posted by Vardant

These are supposed to be new.

http://i39.tinypic.com/14xp3qh.png

I bet they are not using 10.3 drivers that are supposed to be MUCH faster in Dirt2.
Interesting, nonetheless.

Quote:

Originally Posted by SKYMTL

Honestly, it isn't even in my system yet. I am trying to re-benchmark as many cards as possible with the new 10.3a and 197-series drivers before popping the new stuff in. :up:

Great! Going to be an excellent comparison then! Looking forward to it! :up:

Quote:

Originally Posted by Solus Corvus

Dirt2 is also a DX11 title with tessellation - something we have been told repeatedly that Fermi is particularly good at.

It doesn't use much tessellation at all.
Mostly for crowds of people and some minor effects...

Quote:

Originally Posted by eric66

what is power draw of 480 ?

250W.

Quote:

Originally Posted by Olivon

http://tof.canardpc.com/view/623af3e...7e064c137c.jpg

1.5-2x gaming performance of GTX285? 5870 is already faster than that on average...

Quote:

Originally Posted by btdvox

Will there be a 2+ GB version of the GTX 480?

Doubt that, just Tesla cards I think.

Quote:

Originally Posted by takamishanoku

Cheers mindfury for those terrific pics. The heat sink looks quality (as does the card in general)!

Yeah, stylish and sturdy. I like it, too.

Quote:

Originally Posted by ShadowFox19

Being the card has two DVI and one mini-HDMI outputs, will they be able to support two monitors (via the DVI's) and a TV (via the HDMI) at the same time?

Doubt that, since it needs SLI for 3 monitors...

Quote:

Originally Posted by weston

whoa already 33 pages :O

still no real info, so what's the point

Oh, there is real info.

Quote:

Originally Posted by Amorphous

A huge cost is R&D.

Which should hopefully be covered by Tesla sales.
Good point, though.

Quote:

Originally Posted by mao5

5970,5870,5870 2GB 5850 VS. 480, 470 final scores:

Fail graph.
03-22-2010, 01:08 AM
Coldon

Quote:

Originally Posted by eric66

it runs two screens problem was solved long time ago wow what a hater

I was trolling with that comment, jeez relax.

but honestly the flicker issue is still present with the 10.2 drivers and the new 10.3 drivers. I got 4 friends with 5870s, the asus card doesn't flicker for some reasons but the MSI and club3D cards both have flickering on the second screen when connected to my dual 1920x1200 monitors. I really wanted to get a 5870 but seeing as I work on my machine and have dual screens specifically for that reason, the 5870 is not a card I can risk buying.
03-22-2010, 01:14 AM
mindfury

Quote:

Originally Posted by Coldon

but honestly the flicker issue is still present with the 10.2 drivers and the new 10.3 drivers. I got 4 friends with 5870s, the asus card doesn't flicker for some reasons but the MSI and club3D cards both have flickering on the second screen when connected to my dual 1920x1200 monitors. I really wanted to get a 5870 but seeing as I work on my machine and have dual screens specifically for that reason, the 5870 is not a card I can risk buying.

Keeping mem frequency at 3D clock will fix the problem. The asus card doesn't flicker because it's bios keep mem frequency at 1200Mhz in mutli-monitor.

Quote:

Originally Posted by Dave Baumann

Memory clock switching does not occur in mutli-monitor scenarios for the with hardware PowerPlay systems for the reason OpenGL guy commented on (which, as he pointed out, I've also commented on previously). With R700, the flash in multimon mode wasn't so much of an issue because it basically only does it when you boot a game, and in many instances it may also be changing resolution as it loads up, so that wasn't deemed an issue; with hardware PowerPlay, because we can get much lower lows and because it is hardware activity based PowerPlay states will move around frequently - both at the desktop and even in games - and so in multi-panel mode we have restricted the memory clock switching so that the screen flashing as it moves states on some of them do not occur.

Even though, in multi-mon modes, we have to restrict the engine clock a little more than we can in single panel modes, we still achieve better power savings than we could with 2D/3D switching.

http://forum.beyond3d.com/showpost.p...9&postcount=49
03-22-2010, 01:32 AM
WeiT.235

Quote:

3DMark Vantage Extreme
HD5970: 12339
GTX480: 9688
HD5870: 8912
GTX470: 7527
HD5850: 6848

http://www.hardwareluxx.de/community...9-post364.html

Show 100 post(s) from this thread on one page