NVIDIA 600 Series Mobile GPUs Spotted at CES in Samsung Laptops!

[Russian]

This one is by yours truely

While at this year's CES, we received a tip from an unnamed source that there were possibly some Kepler GPUs located in some Samsung laptops on the floor. So, we decided to go and check it out for ourselves and surprisingly enough, there were indeed some Mobile 600 series GPUs.

http://bit.ly/NVIDIA600M

[-=DVS=-]

Mobile kepler ? doubtful, more then likely fermi based gpus Geforce 610M , GT 630M , GT 635M they have been on nvidia site for a while now , doesn't look anything special.

[Russian]

Mobile kepler ? doubtful, more then likely fermi based gpus Geforce 610M , GT 630M , GT 635M they have been on nvidia site for a while now , doesn't look anything special.

Did you check out the link? There are two laptops, with different GPUs with the same shader count but different clocks. One is the 630M and the other is the N13P-GT...

[Lanek]

so a 630M with 70mhz less on the core .... many constructor decrease the speed of the gpu/cpu for meet thermal demand ... i will not really take any conclusion about it without a benchmark

[Russian]

so a 630M with 70mhz more on the core ? will be interessant to know the process used.

That's the confusing part, the GT 630M is clocked higher than the N13P-GT. That's why I'm so confused. If it was a simple die shrink, why would they reduce the clock with the exact same shader count, memory frequency, etc?

[Lanek]

That's the confusing part, the GT 630M is clocked higher than the N13P-GT. That's why I'm so confused. If it was a simple die shrink, why would they reduce the clock with the exact same shader count, memory frequency, etc?

lol yes sorry, i have allready corrected my post.. As i said, it happend offtly constructor do it for thermal demand or Power consumption... but dont forget this was on an exposition model ..

If the card die shrinked in 28nm, or with a better quality 40nm process, why dont keep the same clock ? and just take the increase of perf ? can be for meet a particular tdp ( Nvidia or Samsung ).

We hope Kepler have increase perf/clock ( and this will be logic ) .. But without a benchmark, this is a bit hard to say this is why this difference on clock.

[Russian]

lol yes sorry, i have allready corrected my post.. As i said, it happend offtly constructor do it for thermal demand or Power consumption... but dont forget this was on an exposition model ..

If the card die shrinked in 28nm, or with a better quality 40nm process, why dont keep the same clock ? and just take the increase of perf ? can be for meet a particular tdp ( Nvidia or Samsung ).

We hope Kepler have increase perf/clock ( and this will be logic ) .. But without a benchmark, this is a bit hard to say this is why this difference on clock.

I agree completely. I'm at a loss. I think its possible, though, that since this is not quite a final GPU that it could be down-clocked until its finalized? Idk.

Its all speculation beyond what we see in screenshots. I sort-of regret not benchmarking the two when i had a chance...

[zanzabar]

all gforce 6xx parts are 1 for 1 rebadges of the 500 series and are OEM only, its the same as the gforce 100, so this is not news at all, the new parts are the 700 series.

[lutjens]

The GPUs may simply be relabelled 500M series chips...we all know NVidia has done this a few times...

Edit: Zan, you beat me to it...

[paulbagz]

all gforce 6xx parts are 1 for 1 rebadges of the 500 series and are OEM only, its the same as the gforce 100, so this is not news at all, the new parts are the 700 series.

Thought the 7XX naming series hadn't been confirmed?

-PB

[Russian]

all gforce 6xx parts are 1 for 1 rebadges of the 500 series and are OEM only, its the same as the gforce 100, so this is not news at all, the new parts are the 700 series.

Ahuh. Says who? I'm not entirely convinced that its true... Just because they did it once for Fermi doesn't mean they'll do it again.

[Tao~]

all gforce 6xx parts are 1 for 1 rebadges of the 500 series

just like the 500 was of the 400 series - its all Fermi.
this might be extraneous, but the midlevel amd chips have also remained unchanged 5650=6550=7550.

[zanzabar]

Ahuh. Says who? I'm not entirely convinced that its true... Just because they did it once for Fermi doesn't mean they'll do it again.

it was announced a while ago, and samsung/foxconn have already confirmed this for all but the top. and TSMC has not reported any meaningful 28nm shipments to NV, they did 2 wafers for MS and it got out then production got out this week but they did confirm when amd got large 28nm shipments.

just like the 500 was of the 400 series - its all Fermi.
this might be extraneous, but the midlevel amd chips have also remained unchanged 5650=6550=7550.

they redid the 500 power tracing on most but the lowest end chips (that never really hit market in the 400), its like the 100, and 300 were its a direct rename for OEM only. and if not why would they have a 96 shader chip with a 128 bit buss, that would not fit on a 28nm chip without allot of creative core cutting.

[Aberration]

it was announced a while ago, and samsung/foxconn have already confirmed this for all but the top. and TSMC has not reported any meaningful 28nm shipments to NV, they did 2 wafers for MS and it got out then production got out this week but they did confirm when amd got large 28nm shipments.



they redid the 500 power tracing on most but the lowest end chips (that never really hit market in the 400), its like the 100, and 300 were its a direct rename for OEM only. and if not why would they have a 96 shader chip with a 128 bit buss, that would not fit on a 28nm chip without allot of creative core cutting.

How do you figure?

[zanzabar]

the 108 (the one that one is with 96 shaders and no cuts) is 116mm^2, to shrink that to 28nm u would have less than 90mm^2, so no 128bit buss.

[AffenJack]

all gforce 6xx parts are 1 for 1 rebadges of the 500 series and are OEM only, its the same as the gforce 100, so this is not news at all, the new parts are the 700 series.

Wrong, Asus already confirmed that there are Kepler 600 Series Cards, so 600 Series will be a mixture of Fermi and Kepler.

[Russian]

Its probably not Kepler simply because of the naming sequence, but speculations were that it was...

Also, I'm not entirely sold on your belief that the naming schemes are correct since there have been no indications from my sources of that being true.

[Lanek]

Fermi not kepler .. article from August 2011: http://semiaccurate.com/2011/08/23/n...lineup-leaked/


( Some are still 40nm in this list most probably, and this N13P-GT most likely)

Dont worry, if they want to call them Kepler, they will do it.

[Tim]

Fermi on 28nm ey, that's interesting. Makes sense from a business perspective. Wasn't Fermi supposed to be on 28nm to begin with but due to TMSC having problems 28nm was postponed?

[Lanek]

No, the card was made for 40nm ... you do a mistake with second generation in 32nm as AMD HD6K, but TSMC have drop completely the 32nm at this time, and goes then directly for 28nm.

Nividia have been lucky this time, they was late with the GTX4xx in their programm, they have need again 3months more for release the middle range (GTX460 ), and finally needed to work a lot again on the initial Fermi for release the GTX5xx...
If AMD have got the possibility to release the HD6K as it was normally planned in 32nm.. things will have been maybe different. ( i say maybe, cause impossible to know what it would have been )

[AffenJack]

Fermi not kepler .. article from August 2011: http://semiaccurate.com/2011/08/23/n...lineup-leaked/


( Some are still 40nm in this list most probably, and this N13P-GT most likely)

Dont worry, if they want to call them Kepler, they will do it.

Old info, but drivers show that N13P-GT is Gk107. The Table might be right, but Charlies Interpretation that these are all Fermis is wrong.
Bringing the Device-Id together with Product Name Table which leaked this is a GT650M.
Maybe the Driver isn't recognizing the card in the right way, because with this data it would be way too slow for a 650M.

[Aberration]

the 108 (the one that one is with 96 shaders and no cuts) is 116mm^2, to shrink that to 28nm u would have less than 90mm^2, so no 128bit buss.

Still doesnt answer the question.

[zanzabar]

Still doesnt answer the question.

how so, the native 96 shader part is 118mm^2, amd has said in the past that u cannot have a 128bit buss on anything 90mm^2 or under without an adapter pcb (and NV never had a part under 100mm^2 with a 128bit buss), so since kepler are 28nm and a 96 shader part from 40nm was about 118 and die size in the change should be about 60% of the 40nm u would have about 70mm^2, and since adapter pcbs are not used with gpus and would add alot to the cost they would not have that. so u could not fit the pin out of the 128bit buss without a massive change in design and added cost.

and why would they waste the low production 28nm wafers on bulk chips for OEM (were they make little profit)

[Lanek]

Old info, but drivers show that N13P-GT is Gk107. The Table might be right, but Charlies Interpretation that these are all Fermis is wrong.
Bringing the Device-Id together with Product Name Table which leaked this is a GT650M.
Maybe the Driver isn't recognizing the card in the right way, because with this data it would be way too slow for a 650M.

I will not use the driver list as any evidence.. they can change the sku name, it is still a GF108 ( note the sku change can just come by rework on transistor for leaking, and precision ). all thoses infos have been allready posted ( SA was just the first where i remember the table ), most low end gpu oem and desktop OEM will be renamed brand.... some will be still 40nm, other are pushed in the 28nm when it is possible and needed ( TDP wise ). Why do a new SKU, a new arch for a low end mobile gpu with 96cuda cores ? the margin are so low at this level, and performance are not important at all. at the moment you can watch a video, and play in a browser game.

thoses gpu are determined for be included on low end netbook.. they are not there for compet in any segment other than price. If Nvidia ( or AMD ) will put money and R&D by using their new arch in mobile parts, this is not in thoses price segment...

And like said Zanzibar... why waste 28nm production line in a so insignifiant segment ?

[AffenJack]

Drivers can be pretty credible in some ways like you also saw with Southern Island Launch. 3 Month before Launch you could easy identify nearly every fake, which was discussed here for 10 pages cause many here thougt it might be credible. Most fakes said there is only 1 SI Chip, whereas Drivers showed 3 Chips.
You also say Charlies list is wrong because it's not 28 nm and the naming scheme leaked by computerbase has also to be wrong.
And even if the driver's wrong, how do you get to a GF108? Why can't the driver recognise a gf108? If it's an at the moment unknown name, why wasn't it launched with the other Fermi 600m rebrands?

I'll stay with Occam's razzor and think the driver shows wrong info for kepler instead of sayin that 3 independent sources are wrong and the driver can't recognise an old chip.