jimmyjames123
Regular
NVIDIA's CES 2010 press conference webcast has been available since last night:
http://ispss.istreamplanet.com/nvidia/
When did you start at IMG/PowerVR? Congrats!
Tegra 2 Announcement
- Thread starter Arun
- Start date
NVIDIA's CES 2010 press conference webcast has been available since last night:
http://ispss.istreamplanet.com/nvidia/
When did you start at IMG/PowerVR? Congrats!
Why do you think that? 32bit LPDDR2 667 should offer 2.7GB/s bandwidth. That might be "enough" for the cpu, but it looks to me like it could be pretty limiting for a (non-TBDR at least) renderer. I think the last nvidia gpus with such low memory bandwidth on the desktop were about 5 years ago (GeForce 6100/6200) and those versions which only had that little memory bandwidth (the normal ones had about twice that) were very limited by the lack of bandwidth.
If you look at the Win CE6 user guide for the Harmony kit, you'll see a screen grab that shows remote execution of an app on the hardware. If I'm not mistaken, it shows the 2D and 3D (did we know Tegra 2 had multiple domains?) clock as being 240 MHz.
Now, I'm not discounting that NVIDIA did actually improve the 3D hardware in terms of unit counts, but given Tegra 1 was 120 MHz and NVIDIA are claiming 2x performance in 3D versus the previous generation....
It smells like there's been no improvement other than clocks to that part of the chip to me, and I'm starting to wonder if 240 MHz is the ceiling for the smartbook/netbook/tablet/Harmony SKU that was announced, with any smartphone version either being lower clocked T20, or a different chip altogether.
Now, I'm not discounting that NVIDIA did actually improve the 3D hardware in terms of unit counts, but given Tegra 1 was 120 MHz and NVIDIA are claiming 2x performance in 3D versus the previous generation....
It smells like there's been no improvement other than clocks to that part of the chip to me, and I'm starting to wonder if 240 MHz is the ceiling for the smartbook/netbook/tablet/Harmony SKU that was announced, with any smartphone version either being lower clocked T20, or a different chip altogether.
brain_stew
Regular
Wishmaster
Newcomer
And why is that a bad thing? Their GPU was already best in class so if doubling the clock can make it competitive why not do it? I would be more worried if they'd do it for the third time with tegra3 but for now I'm ok with higher clocked GPU if its good.
But probably they slightly improved the GPU and clocked it higher to keep the developments cost as low as possible. Probably T3 will be just T2 CPU but higher clocked(due to 28nm technological process) with completely new GPU and new video engine.
It's not like they need to do drastic changes every generation. Improve what can be improved and change what needs to be changed.
Best in class in terms of what exactly? I'm not very good in estimating die area from die shots but the graphics part, but if my rather dumb estimate should be close to reality and the GPU on T2 is roughly over 10mm2@40LP and it's still only 2PS/2VS/2TMUs@240MHz, then the perf/mm2 isn't exactly what I'd call ideal.
In any case I'm still not that certain that they haven't double PS/TMU unit count in T2.
If any competing part would cash in just peanuts that would be a notion. Look at the hypothesis above and consider that if it should be true that they haven't increase unit count (which I still doubt) in theory you could squeeze easily a 2MP SGX543 into that die area with a more modest frequency than 240MHz and still have a healthy lead in performance and capabilities overall.
Wishmaster
Newcomer
I'm not saying they didn't either. What I'm trying to say is that even if they only clocked it higher it still is better than what we have in iPhone 3GS and better than both snapdragon1 and 2.
Besides as I said they probably improved GPU from the first generation both on software and hardware part but without any revolution which will probably come with 28nm process.
BTW, when is 28LP coming up? Late this year, mid next year, late next year....?
iPhone 3GS is a smartphone; what you're seeing as Tegra2 for the moment is the highest end variant which isn't necessarily targetted at smartphones unless you intend to run around with a battery backpack for them. SGX535 is Tegra1 comparable material and it's newest Intel iteration is being clocked at 400MHz under 45nm which is over 3x times the T1 frequency.
Process technology advancements aren't exclusive to just one manufacturer. If NV's TegraX benefit from those in the future, the same applies for all other designs. I'm not even sure if there will be any 28LP, but if then by the time it starts production any "revolution" will most likely face the generation beyond what's available today from the competition. In that case that supposed revolution will have to go beyond what's inside the SGX545 for instance, unless of course you're willing to believe the fairy tales that Tegra3 might be a GF100 grandchild just as Tegra2 was supposed to be a GF9 grandchild. The reality check for the latter will tell you that T2 might not even reach GF6 capabilities in the end.
Wishmaster
Newcomer
If the drivers are good this time maybe they will be able to squeeze something from it
Yeah you're right but nvidia along with qualcomm should be one of first to show 28nm SoC's. Besides I never said that T3 will be GF100 based, what I'm trying to say from the beginning that maybe T3 will be completely new just like GF100 is now compared to G80 architecture which was used for 2 generations. That's all.
If the drivers are good this time maybe they will be able to squeeze something from it
Because you have any comparable 3D measures from a T1 powered netbook?
So far I haven't seen any miracles in terms of final device integration either. Let's see how the 40/45nm generation shapes up within this year and we can then talk about the less foreseeable future with a gazillion of "ifs" and too generous expectations in the meantime.
Of course are there very good chances for T3 to rely on a more advanced embedded architecture.
And since you mention G80 it was manufactured on TSMC's 90nm, whereby R600 arrived half a year later on TSMC's 80nm. Shall I go on and elaborate why bets on process technology like above can often turn out to be completely meaningless?
from what's know so far, tegra2 does not seem to be any more OCL/CUDA friendly than tegra1 is, or, IOW, not friendly at all. combined with no SIMD, that would position tegra2 as one of the very few (the only?) pocket computing platforms without viable provision for running custom number-crunching software, whereas virtually all the competition has some means to do some heavy lifting, be that OCL, SIMD, proprietary vector engines or arbitrary combinations thereof. being the only one who can't do something which everybody else on the market can is not an advantageous position - there's no guarantee that tomorrow a killer number-crunching app for handhelds won't come out (say, hypotetically, a new AV codec). what will nv do then - come up with 'CUDA for tegra'? perhaps release a NEON-enabled SKU?
i think that a mindset of 'haldhelds are glorified media players - you support a few codecs (in silicon) and your're good number-crunching-wise' will bite them in the ass, mid-to-long term. of course, having the next nintendo handheld secured may allow nv to not worry much about that.
i think that a mindset of 'haldhelds are glorified media players - you support a few codecs (in silicon) and your're good number-crunching-wise' will bite them in the ass, mid-to-long term. of course, having the next nintendo handheld secured may allow nv to not worry much about that.
What number crunching apps to you envision is going to run on a tablet ?
I would say not spending Si and power on an extension that will go unused by 99.9% of the T2 systems is an excellent choice. Especially considering that the bandwith isn't there to support number-crunching in the first place. Besides Cortex A9 already has a fairly capable FPU.
As a mobile client SOC T2 seems fairly well rounded, IMO.
Cheers
Perhaps game physics or, maybe, photo stitching software? <shrug>
The same question would apply for shaders in the embedded market and many other things. So far evolution though in the embedded market seems to trail fairly good desktop developments in terms of capabilities and your question for benefits in terms of heterogenous computing applies then to all platforms (while on a smaller degree on embedded platforms).
Last time I checked heterogenous computing was all about handling in a more efficient way all system resources and if a lower clocked GPU can reach goal X by several times faster than any embedded CPU out there you're usually on the winning and not the losing side when it comes to overall power consumption.
Of course is T2 well rounded (if you ignore things like perf/mm2 and in extension power consumption despite the funky marketing claims) but that still doesn't make it best in its class just because the IHV that manufactures it starts with a capital N.
In fact the competing solutions will have not in the least anything to worry about in terms of die area/power consumption/hw capabilities etc. except NV's possible faster turn times which might be a headache for the large semiconductor manufacturer competitors to change as competition heats up.
I never claimed T2 would be best in class. NEON was probably on their wish list but was curbed once the die-diet set in. If the T2s prime target market is going to be tablets I can see why.
In tablets, media playback and webbrowsing performance is going to be the most important features. Media playback is well covered by their hardware, web browsing would gain *nothing* from SIMD. Ditching NEON and using the area and power to increase single thread performance makes sense in that context.
I'm sure you're right. NV's competitors probably need to worry more abouyt NV's software stack than their hardware. Having solid Linux drivers and hardware assisted media playback in Flash could be differentiating factors.
Cheers
If you narrow it all down to media playback and webbrowsing I don't think even a GPU was necessary after all.
If that's narrowed again down to linux, flash and stuff like video decoding then I don't think their competition has anything to worry either.
No one is saying that it's an underwhelming SoC, yet it still isn't exactly an 8th wonder either. If the sticker wouldn't state NVIDIA but something like a way less known manufacturer it wouldn't turn even 1/10th of the heads it does today.
Let me see how the trend goes: any intermediate efforts for heterogenous computing as one example will be just marketing wash (while ironically there are already manufacturers working with things like image processing f.e.) and when someone like NV comes along finally with OpenCL support it will be damn important. I guess for NVIDIA evolution works in completely different speeds; cutting edge for the GPU market and for the embedded markets feature advancements only when deemed necessary or else when they can finally squeeze it into a reasonable amount of die area.
In the meantime there are rumors circulating in the background that a major deal rumored to be connected to T2 is in jeopardy due to too high power consumption. For the sake of the irony I was up to recently fairly sure that they have it in their pocket. I am worried personally since I actually want competition to heat up as much as possible. But since T2 is basically a multimedia/web browsing thingy I don't think it's particularly relevant either.
I wonder if SIMD can't be put to good use for various small graphics tasks for which sending commands to some 2d/3d hardware would be more expensive. That'd make Web browsing faster.