NVIDIA Tegra Architecture

leads me to believe that this Kepler will have a hard time just matching PowerVR, let alone leap-frogging it.

LOL, right. Well I'm sure we will find out by CES 2014 (when real products will likely be announced and demoed based on Kepler.M), but I do think that you will be proven very wrong at that time. Personally I doubt that there will be an ultra low power GPU that can match Kepler.M until second half of 2014.
 
I don't recall Nvidia saying Shield would be on a 6-month refresh cycle?

Shield is tied to the refresh cycle of Tegra. While there was a lot of time between Tegra 3 and Tegra 4 (close to 1.5 years in between), there will be relatively little time between Tegra 4 and Tegra 5 (close to 9 months in between).
 
It would have to sell well for them to continue, wouldn't it?

Or do they need it to showcase their SOC and whatever content deals they have -- those optimized for Tegra games.
 
Things aren't looking particularly good for Tegra 4. Looks like Nokia has opted for Qualcomm's S800 instead of Tegra 4 for their WinRT slate. WinRT may not be that that big in comparison to the Android market, but it represented a significant volume for Tegra 3 (almost all WinRT devices were Tegra 3 I believe), and one of the few reasons that Nvidia's Tegra division did as well as it did in the last fiscal year.

It makes me wonder if Microsoft may also be going with the S800 instead of Tegra 4. If so, that would put quite the squeeze on Tegra 4 for this fiscal year.

Regards,
SB
 
Tegra's latest guidance for Fiscal Year 2014 already reflects loss of Windows RT business for the year. Tegra revenue related to Windows RT was never very high in the first place because the platform failed to gain traction in the market.
 
Microsoft made millions of Surface RT tablets. They didn't sell many of those, but they still paid NVIDIA for the chips.
 
Personally I doubt that there will be an ultra low power GPU that can match Kepler.M until second half of 2014.

Well, if it end up in final product ( with proper implementation ) in 1H 2014

then you' ll right . after see how it will compete against ( G6400 MP4 )

but for 2H 2014 it won't hold up in comparison with Qualcomm Adreno 420 GPU, should be implemented in 20nm even !

http://www.brightsideofnews.com/news/2013/1/30/qualcomms-apq8084-leaks2c-rumored-powerhouse.aspx
 
Well, if it end up in final product ( with proper implementation ) in 1H 2014

then you' ll right . after see how it will compete against ( G6400 MP4 )

but for 2H 2014 it won't hold up in comparison with Qualcomm Adreno 420 GPU, should be implemented in 20nm even !

http://www.brightsideofnews.com/news/2013/1/30/qualcomms-apq8084-leaks2c-rumored-powerhouse.aspx
if Adreno 420 comes in 20nm, then its too late to compete with Logan. Mass production of 20nm will not be available than very late of 2014. So it will have to face Parker and its ridiculous power efficient Maxwell GPU.
Then you will say that Parker will be late so it will compete with Adreno 520 ! and so on and so on...
Conclusion, better to stick with Logan vs Rogue fight for the coming months ;)
 
if Adreno 420 comes in 20nm, then its too late to compete with Logan. Mass production of 20nm will not be available than very late of 2014. So it will have to face Parker and its ridiculous power efficient Maxwell GPU.
Then you will say that Parker will be late so it will compete with Adreno 520 ! and so on and so on...
Conclusion, better to stick with Logan vs Rogue fight for the coming months ;)

No not as you expected you Now Adreno 330 already in the market since June actually !

we have Galaxy S4 LTE-A & Xperia Z Ultra ..

I guess Adreno 420 should be in the by late 1H 2014 or early 2H 2014 :D
 
There's little point in splitting up SoC power consumption when one part influences another (GPU driver uses CPU cycles, GPU architecture sets memory bandwidth requirements). You can only take the whole package, anyway.

No, all signs point to Kepler.M coming to market in 1H 2014 (in fact, there is even a reasonably good chance that products will come to market by end of Q1 2014).
Which would put it right in the middle of Apple's release cycle.
 
but for 2H 2014 it won't hold up in comparison with Qualcomm Adreno 420 GPU, should be implemented in 20nm even !

Heh, I don't think that Adreno 420 will be able to keep up with Kepler.M in terms of peak performance level. Keep in mind that Kepler.M has enormous performance headroom. Top-end Kepler.M variants like Shield 2 should approach 400 GFLOPS throughput (for lack of a better performance metric), and top-end Kepler.M tablets should be within ~ 20% of that throughput.

Xmas said:
Which would put it right in the middle of Apple's release cycle

Yes, that is correct, and that is probably a good thing because the ipad is always an attention-getter. NVIDIA's timing is very good with Kepler.M
 
That was already described in the Anandtech article linked earlier. Logan's ULP Kepler GPU will have 192 CUDA "cores" operating at up to ~ 1GHz. The remarkable thing is that this level of peak performance and performance per watt will be achieved in the ultra mobile space on a 28nm fabrication process. Not many people expected that (at least I didn't). For other vendors to match or exceed the peak performance of Logan's Kepler GPU will probably require a somewhat mature 20nm fabrication process, which will not happen until closer to Q4 2014.
 
Heh, I don't think that Adreno 420 will be able to keep up with Kepler.M in terms of peak performance level. Keep in mind that Kepler.M has enormous performance headroom. Top-end Kepler.M variants like Shield 2 should approach 400 GFLOPS throughput (for lack of a better performance metric), and top-end Kepler.M tablets should be within ~ 20% of that throughput.



Yes, that is correct, and that is probably a good thing because the ipad is always an attention-getter. NVIDIA's timing is very good with Kepler.M

I have a hard time believing any of nvidias numbers, predictions or what ever...next to worthless.

The claims made about tegra 4 alone compared to what we ended up with is laughable. ..72 "core" gpu is barely any better than the ipad 4s a6x, and gets beaten by a smartphone chip from its main rival.
http://gfxbench.com/compare.jsp?cols=1&D1=Asus+Eee+Pad+(Tegra+4,+retina)

Also, the reference tablet/shield ended up noticeably slower than non nvidia built shipping products, to the surprise to no one, they have no competitve smartphone SOC all of this year and a scarce roster of OEMs lining up to stick a tegra 4 into a tablet.

Next year, unless im mistaken, they wont be moving to 20nm for mobile keplar where as qualcomm likely will for adreno 420, which also likley has more thermal room to play with thanks to the kraits..if clocked at a similar frequency.

Qualcomm is shipping smartphone SOCs today that are likely pushing 120+ gflops on 28nm (adreno 330) exynos 5420 will soon be shipping (smartphone?) With mali t628 mp6 @ 600mhz...pushing out something like 115 gflops also on 28nm.
Rogue will be released soon which will likely smoke both of those...thats this year on 28nm in smartphones.

Even if nvidia is remotely correct about its perfomance boasts (history make me sceptical) its rivals with be pushing hard and will likely match it.

You will note I have not mentioned AMD..another challanger in the tablet space.
 
@ ams

I already saw the video you posted earlier and I really impressed about Logan power efficiency against ipad 4 !

Regarding 20nm technology I still believe we'll see it by 2H 2014 even with even with Qualcomm first processor build with ARMv8 ... Qualcomm has been known as first company to bring 28nm to the market and we might experience the same thing with 20nm technology..

more coverage ..
http://www.fudzilla.com/home/item/31552-20nm-mobile-processors-in-2014
 
Qualcomm is shipping smartphone SOCs today that are likely pushing 120+ gflops on 28nm (adreno 330) exynos 5420 will soon be shipping (smartphone?) With mali t628 mp6 @ 600mhz...pushing out something like 115 gflops also on 28nm.
Rogue will be released soon which will likely smoke both of those...thats this year on 28nm in smartphones.

Adreno 330 has 128 ALUs running at 450MHz

128 ALU x 2GFLOPS x 1.125 x 0.450 = 129.6 GFLOPS

for Adreno 420 we didn't know about exactly, but from the leaks info know it'll run at 500MHz
if Qualcomm leaves the number of ALUs unchanged then we can calculated as :

128 ALU x 2GFLOPS x 1.125 x 0.450 = 144 GFLOPS

~

Now how much we expecting for Rogue & Tegra 4 ?!
 
source for APQ9084 + Adreno 420 ..

qualcomm-apq8084-quad-core-CPU-adreno-420-GPU.jpg
 
Back
Top