The Fermi Thread - Part 3

Printable View

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

03-16-2010, 08:11 AM
ajaidev

Quote:

Originally Posted by situman

Isnt it about time that the ATi 5 series go thru a refresh? Is it still a 6 month cycle for refreshes/new architectures?

They are lazy because there is no one to fight with. 5990 and 5870 EF 6 are coming and so are custom designs of 5870 with 2gb and 1gb.

I really hope that ATi outs a 5950 "2*1440" something like that will not only be value for money but perform really well even against a rumored GTX 480 Ultra :yepp:
03-16-2010, 08:24 AM
highoctane

Quote:

Originally Posted by herderien

the OS is only handling CPU related ram, the GPU ram is handled by the gpu, the CPU is not really aware of it.

My last off topic here, video ram does decrease available memory address space which in turn limits available address space for system ram with a 32bit OS.

A clear explanation of address space: http://www.dansdata.com/askdan00015.htm
03-16-2010, 08:29 AM
Serra

The Fermi Thread - Part 3

This is, hopefully, the final installment of "The Fermi Thread". As before, this thread exists as without it the news section is overwhelmed by Fermi-related rumors, the majority of which are entirely without substance.

The Rules:
- This thread is about Fermi and things related to it only
- Do you know something about Fermi that isn't here? Post it!
- Don't play with the tags. We can track those and depend on proper tagging for search engine traffic.
- Play nice or you'll get the hammer

Link to Part 1 (82 pages in record time)

Link to Part 2 (109 pages)

If anyone has links to nVidia they feel are credible / official news feel free to send me a PM and I'll review and toss it up. Seriously, I know there is more info out there than the outdated stuff I have below!

Nvidias white paper on the arch: http://www.nvidia.com/attach/2957021...rt&primitive=0 [Jan 19, 2010]
More info: http://www.nvidia.com/object/gf100.html
03-16-2010, 08:33 AM
-Sweeper_

Quote:

Originally Posted by jaredpace

Collection of Fermi Benchmarks: courtesy phk, asuka, obr, kakkoii, kravis, heise.de, computerbase, expreview, & pctuning.cz

http://forums.techpowerup.com/showthread.php?t=117000

:up:
03-16-2010, 08:51 AM
ajaidev

Quote:

Originally Posted by annihilat0r http://www.xtremesystems.org/forums/...s/viewpost.gif

GTX 480: 600-625mhz
GTX 470: 600mhz

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

So the only difference between them is the bus width and 32 shaders? 32 shaders don't mean anything performance-wise. If it's the bus width that will distinguish the two cards, it means the 470 is bandwidth limited even at 600mhz which I don't think is possible.

I think 480's clocks have to be higher.

Thread 3 was unavoidable the real question is will it get around the 90Pages mark till April 6-8th :D
03-16-2010, 08:53 AM
acicula

Quote:

Originally Posted by ajaidev

Thread 3 was unavoidable the real question is will it get around the 90Pages mark till March 6-8th :D

just moderate anything that starts with i think or relies on unsourced information. should be managable then :)
03-16-2010, 08:56 AM
Praz

[QUOTE=acicula;4290064]

Quote:

Originally Posted by ajaidev

just moderate anything that starts with i think or relies on unsourced information. should be managable then :)

lol. Then there would be no thread.
03-16-2010, 09:24 AM
Etihtsarom

So what is the latest specs and pricing rumor?
03-16-2010, 09:32 AM
highoctane

Somebody got some die shots which supposedly is 530 mm2, however the site seems to have taken down the post.

http://vr-zone.com/articles/nvidia-g...hots/8610.html
03-16-2010, 09:32 AM
Mk

Thanks to jaredpace
http://www.xtremesystems.org/forums/...19&postcount=2
03-16-2010, 10:57 AM
spursindonesia

Quote:

Originally Posted by highoctane

Somebody got some die shots which supposedly is 530 mm2, however the site seems to have taken down the post.

http://vr-zone.com/articles/nvidia-g...hots/8610.html

So, diesize is around 530 mm^2, like what has been told by Charlie for months, who would have thought that ? LOL :rofl:

Way to go Mr. Demerjian, bravo. :clap:
03-16-2010, 11:54 AM
ElSel10

Quote:

Originally Posted by spursindonesia

So, diesize is around 530 mm^2, like what has been told by Charlie for months, who would have thought that ? LOL :rofl:

Way to go Mr. Demerjian, bravo. :clap:

Thats true but he was wrong about a lot of other things. He said ORNL cancelled their Fermi supercomputer and he said that Fermi was castrated to 448 SPs, then later admitted back to 512, both of which were incorrect (480 it is).
03-16-2010, 12:16 PM
E30M3

General performance:
gtx285 75%gtx470
hd5850 80%
hd5870 90-95%
gtx470 100%
gtx480 110-115%
hd5970 140%
03-16-2010, 12:19 PM
Vardant

Quote:

Originally Posted by spursindonesia

So, diesize is around 530 mm^2, like what has been told by Charlie for months, who would have thought that ? LOL :rofl:

Way to go Mr. Demerjian, bravo. :clap:

Quote:

GF100 is about 550mm^2 in size

http://www.semiaccurate.com/2010/02/...-and-unfixable

Bravo :rolleyes:
03-16-2010, 12:26 PM
Sn0wm@n

anyone willing to remove the his off of fermi so we can have a look at the die shot when its released ??? :D
03-16-2010, 12:33 PM
Olivon

http://tof.canardpc.com/view/33f87b9...88b1a27fc2.jpg
03-16-2010, 12:38 PM
highoctane

EDIT: I think this is what I think it is....

http://translate.google.com/translat...sl=zh-CN&tl=en

Nekked:
http://img251.imageshack.us/img251/8346/11484455.jpg
03-16-2010, 12:40 PM
Chumbucket843

Quote:

Originally Posted by spursindonesia

So, diesize is around 530 mm^2, like what has been told by Charlie for months, who would have thought that ? LOL :rofl:

Way to go Mr. Demerjian, bravo. :clap:

charlie knew that the die is 23.x * 23.x and 104 dies per wafer. he never gave a specific measurement afaik. that could be anywhere from 530mm2 to 570mm2.
03-16-2010, 12:42 PM
Sn0wm@n

even if charlie didnt give a specific number he came right with alot of things regarding fermi .... now lets hope that if nvidia stas with tsmc for their 28nm node that they wont have problems with yield and stuff like that so they can shrink as fast as they can without problem
03-16-2010, 12:56 PM
eric66

Quote:

Originally Posted by ElSel10

Thats true but he was wrong about a lot of other things. He said ORNL cancelled their Fermi supercomputer and he said that Fermi was castrated to 448 SPs, then later admitted back to 512, both of which were incorrect (480 it is).

how about this nvidia wasn't sure which one to use ? since not only charlie changed his info but many others too :ROTF:
03-16-2010, 12:58 PM
takamishanoku

Quote:

Originally Posted by Mk

Thanks to jaredpace
http://www.xtremesystems.org/forums/...19&postcount=2

Yup this should be sticky'd. Saves going through 200 pages and 2 threads
03-16-2010, 01:21 PM
ElSel10

Quote:

Originally Posted by Chumbucket843

charlie knew that the die is 23.x * 23.x and 104 dies per wafer. he never gave a specific measurement afaik. that could be anywhere from 530mm2 to 570mm2.

http://www.semiaccurate.com/2010/02/...-and-unfixable
03-16-2010, 01:26 PM
annihilat0r

Quote:

Originally Posted by ElSel10

http://www.semiaccurate.com/2010/02/...-and-unfixable

550mm2, 448SP with 600MHz core. Let's wait and find out how many of those were correct.

On an unrelated note, I found out that I had 141 posts in the Fermi Part 2 thread. :eek:
03-16-2010, 01:34 PM
455olds

Quote:

Originally Posted by E30M3

General performance:
gtx285 75%gtx470
hd5850 80%
hd5870 90-95%
gtx470 100%
gtx480 110-115%
hd5970 140%

You switched it around. The 5870 is rumored to be faster than gtx470.
This is the order that was originally posted.
General performance:
gtx285 75%
hd5850 80%
gtx470 90-95%
hd5870 100%
gtx480 110-115%
hd5970 140%
Here is where it was posted. http://www.xtremesystems.org/forums/...19&postcount=2
03-16-2010, 02:00 PM
zalbard

Quote:

Originally Posted by highoctane

EDIT: I think this is what I think it is....

Hope they have WAY more chips that that!

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