Back in September, 2010, Nvidia Corp. unveiled its three-years roadmap that disclosed the company's plans to release Kepler family of graphics processing units in 2011 and Maxwell lineup of GPUs in 2013. A lot of happened since then and problems with fabrication processes apparently changed the roadmap of the company. As a result, Kepler family of chips will be relatively small, whereas Maxwell line will emerge only in 2014.
Without any doubts, Kepler architecture is among the most impressive products developed by Nvidia ever. However, at present Nvidia only has two chips - GK104 and GK107 - that utilize Kepler with one more - GK110 - announced for compute applications. Even the first-generation Fermi family featured four graphics chips, with the addition of two additional second-gen Fermi processors, Fermi family included six chips in total. According toVR-Zone web-site, Nvidia will refresh Kepler family no earlier than in March, 2013, in a bid to increase performance of its graphics offerings by 25% - 30%, meanwhile the first Maxwell GPUs are only projected to be unveiled in 2014.
The main reason for the delays is believed to be slow roll-out of new process technologies by Taiwan Semiconductor Manufacturing Company and its rivals.
Nvidia Maxwell will be a very important GPU family for the company. Maxwell will be Nvidia's first graphics processing units to contain project Denver 64-bit ARM-compatible general-purpose core, which means that Maxwell will be able to boot operating systems themselves. Maxwell will be the first top-to-bottom GPU architecture, powering everything from Tegra to Tesla.
Nvidia did not comment on the news-story.
Tags: Nvidia, Maxwell, Geforce, Tesla, Quadro, Tegra, Kepler, Fermi, 20nm
Comments currently:
8
Discussion started: 08/28/12 04:06:14 PM
Latest comment: 08/30/12 05:37:41 PM
Expand all threads |
Collapse all threads
[1-4]
1.
LOL, how's this news?! It's been known since july 2011.
http://news.softpedia.com...ompany-Slide-210023.shtml
And the official PDF from nvidia themselves dated june 2011, page 14:
http://www.nvidia.com/doc...ng-Sumit-June11-Final.pdf
0
1
[
Posted by: eddman
|
Date: 08/28/12 04:06:14 PM]
2.
25-30% faster only from GTX780? Sounds underwhelming. Looks like NV is not going to give us the full-fledged GK110 then. If the 25-30% estimate is true, HD8970 has a shot at perhaps beating the GTX780, especially at high resolutions where the 7970 GE already leads the 680 by 10%. NV may need more than 25-30% to reclaim the performance crown from the 8970 at enthusiast resolutions and triple monitor gaming. If HD8970/GTX780 are just 25-30% faster than the HD7970 GE, I am skipping a generation straight to Maxwell/HD9000 for a more tangible performance increase. I think GTX780 can be faster though if NV actually launched the full GK110 chip.
show the post
0
4
[
Posted by: BestJinjo
|
Date: 08/28/12 09:27:45 PM]
+ expand thread (3 answers)
- collapse thread
Unless the next Radeon has a 512 bits memory bus, the growth will be bigger for Nvidia at higher resolutions(256->384).
0
1
[
Posted by: cosminmcm
|
Date: 08/29/12 12:37:00 AM]
In % terms relative to GTX680, it may be true. We don't know if NV will release a full-blown 384-bit GK110 chip though. Based on those 25-30% estimates, doesn't look like that at all. Maybe it will be a 2304 SP / 320-bit GK110 derivative, who knows.
Right now a 7970 GE is already 9-10% faster overall depending on the source at 1600P, and a 1200mhz 7970 is 19% faster:
http://tpucdn.com/reviews...B/images/perfrel_2560.gif
That means at high resolutions a hypothetical 25% 8970 vs. 7970 GE may not be that much behind a 30-35% faster GTX780 vs. 680. AMD cards seem to do well as the workload increases. The same has been true with 4890 vs. 285, 5870 vs. 480, 6970 vs. 580, and continues with 7970 GE vs. 680. Even a stock 925mhz 7970 = 680 at 1600P. I have no doubt that a GK110-based 780 will probably end up faster at 1080P though. However, at that level of performance it would be almost wasteful to buy a GTX780 to play console ports at 1080P.
1
4
[
Posted by: BestJinjo
|
Date: 08/29/12 12:07:37 PM]
There is no leak on 8970 specs. It probably won't be much better than 7970, only beause to do that they will have to increase the die significantly. There will be no contest, GTX780 wins hands down.
For the examples that you gave above, the situation was different: Nvidia didn't push the clocks as much as they could, because of the high power consumption, while AMD pushed it almost to the max, there was not much you could squeeze out by overclocking. They did that because the little dies they used could't do much at moderate clocks.
0
1
[
Posted by: cosminmcm
|
Date: 08/30/12 01:03:14 AM]
3.
the most interesting feature would be: 64-bit ARM-compatible general-purpose core. bound/glued 8 or 16 of those together and NV would have a good share of x86 cores. more and more super computer will be using NV processors so what do we think the techies would want in their desktops?
0
2
[
Posted by: idonotknow
|
Date: 08/29/12 12:47:42 AM]
+ expand thread (1 answer)
- collapse thread
ARM core != x86
2
0
[
Posted by: rower
|
Date: 08/29/12 01:27:02 AM]
4.
Intel MIC on 14nm vs Nvidia Maxwell on 20nm (?) in 2014 will be very interesting to watch.
1
0
[
Posted by: lol123
|
Date: 08/30/12 05:37:41 PM]
[1-4]
Make Home Page
Add to Favorites
Xbitlabs on Facebook
XbitLabs is using Twitter
Windows Live Alerts
