Nvidia Kepler Rumors Options

[gamble]

The Gtx 680 eta date kept changing so I switch my order to the Gtx 670, might as well save some money!

Looks like the Gtx 670 overclocks pretty good: http://www.gpureview.com/gpureviews-gtx-67...rticle-947.html

Kinda nice for a quick view!

[`angerfist]

are you disagreeing with me then agreeing?

Sorry ill write *sarcasm* next time so you can see ;D Oh and about the availability of the gtx 680, have you been to an msy lately? =/ I think you should go and not just show their pdf link. As for me im not someone to order online, I like to go in store and buy it (just personal preference) and NOONE has the card instore currently atm.

The ETA keeps changing because nvidia cant deliver according to the consumers demands. I think the successful amount of manufactured gtx 680s was something like 40%? But that could also be why everywhere else in the world if also lacking these cards. Maybe and hopefully the gtx 670 will see a better launch.

[nesquick]

are you disagreeing with me then agreeing?

Sorry ill write *sarcasm* next time so you can see ;D Oh and about the availability of the gtx 680, have you been to an msy lately? =/ I think you should go and not just show their pdf link. As for me im not someone to order online, I like to go in store and buy it (just personal preference) and NOONE has the card instore currently atm.

The ETA keeps changing because nvidia cant deliver according to the consumers demands. I think the successful amount of manufactured gtx 680s was something like 40%? But that could also be why everywhere else in the world if also lacking these cards. Maybe and hopefully the gtx 670 will see a better launch.

The supply problem is 3 fold, one the cards are slightly better than the AMD's offerings and two TSMC is having trouble supplying both AMD and Nvidia 28nm wafers and 3 when something just releases you can't expect massive levels of stock cause that would be a waste of time in stockpiling it whereas they could have a partial launch and people queuing up to buy said product.

I think it unrealistic to expect massive amounts of stock regardless of company for launch of a really good product, wasn't it like 3 weeks before stock for the 7970 started filtering down for example?

[nobody813]

GTC 2012 Part 1: NVIDIA Announces GK104 Based Tesla K10, GK110 Based Tesla K20

GK110 Packs 2880 CUDA Cores, 384-bit Memory Interface: Die-Shot

With its competition checked thanks to good performance by its GK104 silicon, NVIDIA was bold enough to release die-shots of its GK110 silicon, which made its market entry as the Tesla K20 GPU-compute accelerator. This opened flood-gates of speculation surrounding minute details of the new chip, from various sources. We found one of these most plausible, by Beyond3D community member "fellix". The source of the image appears to have charted out component layout of the chip by some pattern recognition and educated guesswork.

It identifies the the 7.1 billion transistor GK110 silicon to have 15 streaming multiprocessors (SMX). A little earlier this week, sources close to NVIDIA confirmed the SMX count to TechPowerUp. NVIDIA revealed that the chip will retain the SMX design of GK104, in which each of these holds 192 CUDA cores. Going by that, GK110 has a total of 2880 cores. Blocks of SMX units surround a centrally-located command processor, along with six setup pipelines, and a portion holding the ROPs and memory controllers. There are a total of six GDDR5 PHYs, which could amount to a 384-bit wide memory interface. The chip talks to the rest of the system over PCI-Express 3.0.

This post has been edited by nobody813: May 17 2012, 07:17 PM

[gamble]

Got my GTX 670 today. Just reinstalled windows 7 and updates. Downloading a few games, 3dmarks, and Heaven. Early numbers before OC on the memory are:

Without any tweaks to power or OC, the cards core goes up to 1025mhz and the memory is 3105mhz in EVGA scanner stress test with heavy mode enabled.
Without any tweaks to power or OC, the cards core goes up to 1150mhz and memory is still 3105mhz for Heaven 3.0 with everything maxed.
Temp with custom power band goes up to 66c with the fan speed matching temp of 66%.

Power boost set to 122% (highest):
Heaven crashes at 1250mhz halfway thru the bench without any notable artifacts(gpu offset 100mhz). Heaven passed 1228mhz (offset 75mhz). So somewhere in between is the sweet spot! Temp matching speed got up to 69c. GPU TDP got up to 107%.

EVGA Scanner stress test with heavy mode (gpu offset of 100mhz) clocks the gpu to 1250mhz at first then drops to 1215mhz. TDP goes up to 129%! Temp is around 70c. So, you can see how it is downclocking itself. Strange that Heaven failed at gpu offset of 100mhz but it was really running 1250mhz and not scaling down. Something to investigate! But need to stop for now after all day of messing with windows, downloads, and tweaking settings.

More to come!

[SceptreCore]

Can someone explain to me what the GTX600 series is at the core level. What is it based off? Is it Kepler? Tesla?

[gamble]

Is this what you were looking for? Kepler is in the 600 series and Tesla uses Kepler cores too:

http://www.tgdaily.com/hardware-features/6...ered-tesla-gpus
http://www.theinquirer.net/inquirer/news/2...m10-gpgpu-board
http://www.v3.co.uk/v3-uk/news/2174765/nvi...tesla-m10-board

[nesquick]

Can someone explain to me what the GTX600 series is at the core level. What is it based off? Is it Kepler? Tesla?

Kepler is the architecture, gefore and geforce tesla are just marketing names afaik.

[philo-sofa]

The two architectures in play are Kepler and Fermi. Tesla is branding name for workstation cards (a holdover from the GTX 200 series 'Tesla' architecture, which was the gen of 'compute' cards Nvidia offered).

As gamble says, some lower end 600 series cards use the older Fermi core (GT 610, 620 & 630 as well as some mobile units). The GTX 680, GTX 670, GT 640 and probably the upcoming '650 & '660) use the new Kepler core. Kepler being a much evolved (28nm) descendant of Fermi. Anandtech put it well:

"Kepler brings with it some very important architectural changes compared to Fermi, but at the same time it’s still undeniably Fermi. From a high level overview Kepler is identical to Fermi: it’s still organized into CUDA cores, SMs, and GPCs, and how warps are executed has not significantly changed... After doubling the size of the functional units in a SM, NVIDIA then doubled the number of functional units in each SM in order to grow the performance of the SM itself. 3 groups of CUDA cores became 6 groups of CUDA cores, 2 groups of load/store units, 16 texture units, etc. At the same time, with twice as many functional units NVIDIA also doubled the other execution resources, with 2 warp schedulers becoming 4 warp schedulers, and the register file being doubled from 32K entries to 64K entries."

This post has been edited by philo-sofa: Jun 8 2012, 10:06 AM

[Dasa]

gt640 isnt looking to flash at ~$100-110us rrp and slower than 6670 $65
its almost half the speed of 7750 $107
http://www.tomshardware.com/reviews/geforc...eview,3214.html

its price really needs to drop
hopefully nvidia can at least manage some better performance with the 650 660

This post has been edited by Dasa: Jun 8 2012, 01:34 PM

[gamble]

In case someone is interested, this was for about 30mins maps. BF3 MP maxed out with latest drivers 301.42 and latest 2gb game patch running at 1200c/6800m, max temp 67c. Temp would stay around that and the fan was matching temp at 67%. Power limit got up to 120% but usually sits around 114-116% when playing.

Gulf of Oman:
2012-06-08 22:35:47 - bf3
Frames: 147173 - Time: 1815774ms - Avg: 81.052 - Min: 37 - Max: 169

Grand Bazaar:
2012-06-08 23:07:15 - bf3
Frames: 139267 - Time: 1805181ms - Avg: 77.148 - Min: 37 - Max: 179

Sharqi Penisula:
2012-06-08 23:41:21 - bf3
Frames: 135426 - Time: 1774480ms - Avg: 76.319 - Min: 36 - Max: 191

This post has been edited by gamble: Jun 9 2012, 03:31 PM