Brilliantdeve
Newcomer
NVIDIA Tegra Architecture
- Thread starter french toast
- Start date
-
- Tags
- nvidia
I wouldn't say that it is totally inefficient. It's just that in a relatively small and thin handheld device, Cortex A15 isn't able to stretch it's legs (in Shield the Cortex A15 is wickedly fast for a handheld device). Keep in mind that pretty much all high performance quad-core SoC's can be pretty power hungry at peak performance levels.
HP Slatebook x2 web browsing, gaming, boot time, UI smoothness: http://www.youtube.com/watch?v=jEt_akYAByw&list=UUyPhjvvD73RpEmq-FaQEq1g
Apparently this device is selling for $455 shipped (with coupon code SVD8492, see here: http://www.gadgetreview.com/2013/08...2-coupon-slashes-5-off-the-tablet-hybrid.html) with keyboard attachment and with extra battery inside the keyboard all included. Not too bad. In my opinion, this was a smart way to design and sell a T4-powered tablet device.
Apparently this device is selling for $455 shipped (with coupon code SVD8492, see here: http://www.gadgetreview.com/2013/08...2-coupon-slashes-5-off-the-tablet-hybrid.html) with keyboard attachment and with extra battery inside the keyboard all included. Not too bad. In my opinion, this was a smart way to design and sell a T4-powered tablet device.
ok we know you hate NV so please stop moaning, it becomes really pathetic.
The thing is that bringing together a competitive SoC (CPU+GPU+Modem) is a huge task, in fact, a multi-year multi-billion dollar project. Competition is fierce, Qualcomm is an experienced player with deep pockets. intel, no need to talk about it. Samsung is another giant...
Nvidia is trying to get a pie of this very lucrative business and even if the result is a mixed one for now, they need to be there and their plan seems to be legit. Tegra was a first try, Tegra 3 was an average part but many times better, Tegra 4 has much higher performance but not so power efficient, Tegra 5 looks like it will solve the GPU power efficiency and feature set. Finally Parker is the one that will tell us if nvidia can compete in this field... or not ! With custom ARM V8 project Denver 64 bit cores + Maxwell ridiculous power efficient GPU class + their mature soft modem = best NV shot. If they fail with Parker, maybe it will be the end of their mobile dreams, but something tells me that they will be extremely competitive...
Rendez-vous in 2015
Somebody has to be a counter to ams's insufferable cheerleading, all based on one canned benchmark.
Yeah that's the point that should be obvious to everybody. Nvidia isn't up to the task. Now they are desperately trying to remake Tegra into a gpu licensing business because they already tried and failed, over and over, at making their own silicon. That's exactly why we got the Logan demo to deflect from the pathetic performance of their current offering. This is what Nvidia always does when they have nothing.The thing is that bringing together a competitive SoC (CPU+GPU+Modem) is a huge task, in fact, a multi-year multi-billion dollar project. Competition is fierce, Qualcomm is an experienced player with deep pockets. intel, no need to talk about it. Samsung is another giant...
Nvidia is trying to get a pie of this very lucrative business and even if the result is a mixed one for now, they need to be there and their plan seems to be legit. Tegra was a first try, Tegra 3 was an average part but many times better, Tegra 4 has much higher performance but not so power efficient, Tegra 5 looks like it will solve the GPU power efficiency and feature set. Finally Parker is the one that will tell us if nvidia can compete in this field... or not ! With custom ARM V8 project Denver 64 bit cores + Maxwell ridiculous power efficient GPU class + their mature soft modem = best NV shot. If they fail with Parker, maybe it will be the end of their mobile dreams, but something tells me that they will be extremely competitive...
Rendez-vous in 2015
Did any of you ever stop to consider why - if Nvidia really had such a huge advantage - they are so desperate to license it out to competitors all of a sudden? Surely THIS TIME would really be it, with their incredible perf/W dominance brought by "Kepler.M"? I mean why license out your huge advantage (according to ams) when you could just use it to your own benefit and finally make inroads into the market you've been failing to break into for years?
Last edited by a moderator:
jimbo, you should trop trolling and go somewhere else to discuss products that you actually have an interest in. At the end of the day, Tegra 4 is one of the fastest ultra mobile SoC's on the market today, and most T4-based tablets/phablets will be reasonably priced and will have similar system peak power consumption and peak temps compared to most other high end ultra mobile quad-core tablets/phablets.
P.S. Stop blaming NVIDIA and Intel for AMD's lack of foresight and lack of investment in ultra low power processors and the associated ultra mobile technology that goes with it.
P.S. Stop blaming NVIDIA and Intel for AMD's lack of foresight and lack of investment in ultra low power processors and the associated ultra mobile technology that goes with it.
Last edited by a moderator:
french toast
Veteran
Well as others have said, I think to determine gpu efficiency and even cpu efficiency you would need to include more data into the equation then just flops and outright benchmarks, power consumption is probably the most important issue which you have left out, something which has been criticised heavily about nvidias prehistoric mobile gpu uarch.
We can completely throw out any nvidia marketing slides, reference tablets and especially shield...as they are about as realistic an indicator as trying to measure your table width holding your arms out and guessing.
Nvidia had enough goes at mobile soc to have been able to come up with a better solution than tegra 4...in some ways they have gone backwards. .as it cant fit into a smartphone chassis without a mains charger plugged in and an aspestos case, where as tegra 1, 2&3 could.
The shadow core turned out to be an agricultural solution to power consumption on tegra 3 - if not an inventive one -...so why did they use it again? Why not just go with big_little like everyone else? Why use 5 cortex A15's when it must have been obvious in testing that power consumption was going to be massive?
If they really did intend to quit smartphones at the high end and aim for tablets, why equip such a comparatively weak, outdated and power hungry gpu to go alongside those beastly a15s?...
No I think they went for a tegra 3 straight swap like for like and just cocked it up. ..tegra 4 is essentially the bulldozer of the mobile world im sad to say.
Tegra 4i on the other hand looks very interesting indeed, I camt wait to see it in devices.
Evidence? There must be a really good reason why T4 has been so overwhelmingly beaten. So what is it? If it's "competitive" like you claim, Nvidia wouldn't have to build their own hardware to sell it.
It's not competitive. It's not even close to being competitive on any metric.
AMD's lack of multi-billion dollar losses you mean? Both Intel and Nvidia have wasted billions and are no closer to cracking the mobile market. Some might say they are further away than ever.
@ french toast: big.LITTLE vs. 4+1 is not the issue at all. The issue is that peak performance of the Cortex A15 needs to be capped in order to fit the power constraints of a relatively small and thin handheld device. In reality that doesn't matter too much, because Cortex A15-based SoC's are still really fast compared to almost any other ultra mobile SoC out there. As for Tegra 4, it absolutely stomps on Tegra 3 with respect to CPU and GPU performance. In fact, GPU performance is up by a factor of up to 6x, so the perf. per watt is actually quite a bit better on T4 compared to T3. That is hardly what I would call a backward move. As with any ultra high performance quad-core ultra mobile SoC, peak power consumption tends to go up as peak performance goes up.
Of course it is competitive. The CPU performance is at the top of the heap, and the GPU performance is as good or better than virtually any other ultra mobile SoC other than S800 (and that is a much more expensive SoC with cost of integrated modem included and a 50% larger die size area to boot).
NVIDIA and Intel have already cracked the ultra mobile market to some extent with what is sampling right now. In the meantime, AMD is burning through cash due in part to massive penalty payments to Global Foundries.
Somewhat higher, really?
SoC Shootout: x86 vs. ARM
http://gfxbench.com/result.jsp?benc...e=device&version=latest&data-source=1&site=gl
Links for die size and cost? I mean tbh ams, if Nvidia has all the advantage you claim they do, why is it that Qualcomm has all the design wins?
Nobody is sampling Nvidia stuff, because nobody trusts them or even cares. Kepler might be the best ever mobile SoC but Nvidia's bullshit has burned every bridge they had, so it wouldn't even matter. When you promise high and deliver low like Nvidia does time after time, no company worth their salt would believe anything coming from them. Tegra 1-4 were abject failures. Every one with high promises that weren't delivered. That's the only fact that matters.
Is that $500m supposed to be in profit or the gross revenue from selling the chips? And if it's the latter does the $400m in incremental investment not include those manufacturing costs?
Honestly having a hard time even seeing where nVidia is going to gross $500m selling Tegras this year...
Yes, really. If you look at a graphics benchmark where these ULP GPU's actually have anything close to playable and smooth framerates, the difference is "only" ~ 20% comparing shipping devices: http://gfxbench.com/result.jsp?benc...true&arch-MIPS=true&arch-x86=true&base=device . That is a significant difference, but nothing extraordinary either.
You can find info on die size area if you do some reading. Even though NVIDIA has cost savings with the T4 SoC vs. the S800 SoC, the reason Qualcomm's SoC's are very popular in general is due in large part to their strength in the baseband modem business.
$500 million is Tegra revenue across all lines of business. It is my understanding that the $400 million incremental cost for Tegra includes all costs associated with Tegra (including manufacturing costs). As for how they can reach $500 million revenue for the year, well that is still way down from last fiscal year, and their automotive business has doubled vs. last year.
EDIT: the incremental investment R&D cost for Tegra for this Fiscal Year 2014 is actually $300 million (not $400 million). The manufacturing cost is unknown. The R&D costs are much lower for Tegra this year because much of the R&D is now leveraged with the core GPU R&D.
Last edited by a moderator:
I must be pretty out of the loop, because I never understood the point of these high end mobile SoCs in cars. What is the GPU power used for, prettier 3D models in navigation maps and a slicker GUI? And what are all those CPU cores being put to use for? At any rate, what kind of volume are we talking here, because at best I expect it to only be in higher end models, or am I wrong on this?
I wouldn't believe the ASP for Tegra chips this fiscal year is > $25 (could be much lower) so that'd mean > 20m chips to hit $500m revenue. Unless they're selling something else that gets counted as Tegra income. If we really are talking > 20m chips then I still don't see how it's going to happen this fiscal year.. by the way, I'm not even sure if that means FY2013 or FY2014..
I wouldn't believe the ASP for Tegra chips this fiscal year is > $25 (could be much lower) so that'd mean > 20m chips to hit $500m revenue. Unless they're selling something else that gets counted as Tegra income. If we really are talking > 20m chips then I still don't see how it's going to happen this fiscal year.. by the way, I'm not even sure if that means FY2013 or FY2014..
These mobile SoC's are being used for Navigation systems, Infotainment systems, Digital Instrument clusters, Rear Seat entertainment, Video/Image processing, Audio processing, Driver assistance, etc.
Texas Instruments is actually partnering with NVIDIA in the automotive, consumer, and embedded space (which is more evident now that TI has refocused away from producing their own ultra mobile SoC's): http://www.ti.com/lit/sg/slyt510/slyt510.pdf
Texas Instruments is actually partnering with NVIDIA in the automotive, consumer, and embedded space (which is more evident now that TI has refocused away from producing their own ultra mobile SoC's): http://www.ti.com/lit/sg/slyt510/slyt510.pdf
Last edited by a moderator:
All of that extra R&D that you refer to is a part of the $300 million incremental R&D cost. Cost of goods sold on $500 million revenue is probably ~ $250 million (since Tegra gross margins are only slightly below the current corporate average). So that means that Tegra revenue above ~ $600 million for the year would lead to profitability. Tegra revenue of ~ $500 million would lead to a relatively mild loss of $50 million for the year.
Similar threads
