NVidia - how is it going as a company?

[rgallant]

@ FearfulSPARTAN: The Xiaomi Mi3 Tegra 4 design win with China Mobile is pretty big news for Tegra 4 for a variety of different reasons. Xiaomi is one of the fastest growing companies in history ($10 billion valuation after only three years in business). China Mobile is the world's largest wireless service provider. And the system performance of the 1.8GHz Tegra 4 variant in this smartphone form factor appears to be very good even compared to the 2.3GHz S800-8974AB variant (even though the latter does have better GPU-centric benchmark performance):

Antutu: 36582 vs. 36026
Quadrant: 18702 vs. 18975
Geekbench2: 4106 vs. 4139

So what is the difference between Tegra 2/3/4 vs. Tegra 5 "Logan" and Tegra 6 "Parker"? The difference is that, with Tegra 2/3/4, there was only one primary design team working with a less-than-modern GPU architecture. So, after Tegra 2 was done, they started Tegra 3, and after Tegra 3 was done, they started Tegra 4 (all the while having strict SoC die size area requirements to keep production cost low, and all the while being unable to heavily leverage existing modern-day GPU assets). Starting with Tegra 5 "Logan" and Tegra 6 "Parker", NVIDIA is designing two modern-day SoC architectures at the same time. This will make a big difference in both performance (with a modern-day GPU feature-set) and time-to-market.

When Tegra 5-powered devices first come to market (which should be as early as Mar-Apr 2014 since the "Logan" SoC started sampling in ~ June 2013), the peak GPU performance of Kepler.M will be unmatched in the ultra mobile space in my opinion. I do not believe that this level of peak GPU performance will be matched in this space until a somewhat mature 20nm fabrication process in the second half of 2014. And right around the corner in 1H 2015 will be Maxwell.M GPU and the circle will turn again in NVIDIA's favor. So instead of being one step behind with respect to GPU performance in the ultra mobile space, they will be one step ahead starting with Tegra 5 "Logan". The writing is on the wall so-to-speak.

won't intel have sub 20nm at that time?

 

[RadiclDreamer]

I would pay about $100 bucks for a shield, tops. I love Nvidia GPUs but they need some "killer apps" and a lower pricetag for the shield to catch on.

 

[raghu78]

@ FearfulSPARTAN: The Xiaomi Mi3 Tegra 4 design win with China Mobile is pretty big news for Tegra 4 for a variety of different reasons. Xiaomi is one of the fastest growing companies in history ($10 billion valuation after only three years in business). China Mobile is the world's largest wireless service provider. And the system performance of the 1.8GHz Tegra 4 variant in this smartphone form factor appears to be very good even compared to the 2.3GHz S800-8974AB variant (even though the latter does have better GPU-centric benchmark performance):

Antutu: 36582 vs. 36026
Quadrant: 18702 vs. 18975
Geekbench2: 4106 vs. 4139

So what is the difference between Tegra 2/3/4 vs. Tegra 5 "Logan" and Tegra 6 "Parker"? The difference is that, with Tegra 2/3/4, there was only one primary design team working with a less-than-modern GPU architecture. So, after Tegra 2 was done, they started Tegra 3, and after Tegra 3 was done, they started Tegra 4 (all the while having strict SoC die size area requirements to keep production cost low, and all the while being unable to heavily leverage existing modern-day GPU assets). Starting with Tegra 5 "Logan" and Tegra 6 "Parker", NVIDIA is designing two modern-day SoC architectures at the same time. This will make a big difference in both performance (with a modern-day GPU feature-set) and time-to-market.

When Tegra 5-powered devices first come to market (which should be as early as Mar-Apr 2014 since the "Logan" SoC started sampling in ~ June 2013), the peak GPU performance of Kepler.M will be unmatched in the ultra mobile space in my opinion. I do not believe that this level of peak GPU performance will be matched in this space until a somewhat mature 20nm fabrication process in the second half of 2014. And right around the corner in 1H 2015 will be Maxwell.M GPU and the circle will turn again in NVIDIA's favor. So instead of being one step behind with respect to GPU performance in the ultra mobile space, they will be one step ahead starting with Tegra 5 "Logan". The writing is on the wall so-to-speak.

Sorry but Logan in mid-2014 is nothing great. Already companies like Mediatek are using Imagination PowerVR Series 6 ‘Rogue’ based GPUs in ARM SOCs. These chips are already in production and will be available in products in Q4 2013.

http://www.imgtec.com/news/Release/index.asp?NewsID=797
http://www.youtube.com/watch?v=Kf_ZGDoEPMo

Apple's upcoming A7 chip will also feature a Rogue based GPU. By Q4 2014 Apple and Qualcomm will have their first 20nm SOC chips shipping. Nvidia is always behind the game in Tegra. Nvidia don't switch to the leading edge node like Apple , Qualcomm and so Logan will fight against 20nm SOC chips for the most part of its 1 year lifecycle. Also Nvidia has never been able to compete on perf/watt when compared to Qualcomm, Apple. With Intel entering the phone market in a big way with Baytrail 22nm SOC chips it gets even more difficult.

 

[lessthanuthought]

And how's AMD doing? Sending minions all over the forums trying to get APU design wins? :hmm:

Infraction issued for thread crapping. Enjoy your vacation.
-- stahlhart

PM sent.
-- stahlhart

 

[ams23]

@ raghu78: I already explained why past Tegra history is a poor basis for future projections. And Kepler.M has already been demonstrated to have 3x higher gaming benchmark perf. per watt than the A6X GPU, so the myth that NVIDIA cannot keep up in terms of GPU perf. per watt has been put to rest. As for Rogue, variants start at ~50 GFLOPS, but multiple clusters are required to get anywhere near the ~ 300-400 GFLOPS throughput that Kepler.M is targeting at the high end. Those higher end Rogue variants will not come to market until closer to end of 2014, and may still have a tough time matching Kepler.M. Sorry, but NVIDIA will be way ahead of everyone else for many months after Kepler.M debuts in early 2014 (particularly at the high end of the spectrum where it can stretch it's legs a bit).

 

[Keysplayr]

Yes, I thought Kepler arch was to enhance or replace the next gen Tegra arch.

http://blog.laptopmag.com/faster-than-a-ps3-nvidia-mobile-keppler-graphics-will-power-next-tegra-cpu

 

[raghu78]

@ raghu78: I already explained why past Tegra history is a poor basis for future projections. And Kepler.M has already been demonstrated to have 3x higher gaming benchmark perf. per watt than the A6X GPU, so the myth that NVIDIA cannot keep up in terms of GPU perf. per watt has been put to rest. As for Rogue, variants start at ~50 GFLOPS, but multiple clusters are required to get anywhere near the ~ 300-400 GFLOPS throughput that Kepler.M is targeting at the high end. Those higher end Rogue variants will not come to market until closer to end of 2014, and may still have a tough time matching Kepler.M. Sorry, but NVIDIA will be way ahead of everyone else for many months after Kepler.M debuts in early 2014 (particularly at the high end of the spectrum where it can stretch it's legs a bit).

Nvidia are always overpromising and underdelivering wrt Tegra. OEM's have dropped Nvidia because Qualcomm provides better perf/watt and LTE integration which is crucial for smartphone market. Even in tablets Qualcomm Snapdragon 800 is killing Tegra 4. There is no single high profile high volume design win for Tegra 4 like a Google NEXUS. Qualcomm replaced nvidia in the recently released Google Nexus 7. One of the biggest reasons for Nvidia to fail in the mobile market is OEMs value perf/watt more than raw perf. Nvidia has always failed to deliver on the former wrt competitors like Qualcomm.

http://www.anandtech.com/show/7169/nvidia-demonstrates-logan-soc-mobile-kepler

As for Logan Nvidia has mentioned 200 GLOPS - 400 GLOPS as the target range performance. To hit even 200 GFLOPS they have to clock the GPU at 500 Mhz. Thats not easy when you want to stay around 2- 2.5W for GPU. Let me say until Nvidia delivers final products I am skeptical of their lofty claims.

 

[Imouto]

Nvidia said that Tegra 4 would beat the crap out of the S800. Guess what happened? It's the same every single time with Nvidia mocking and putting down the competition and being burned later by reality. Mobile OEMs realized this a while ago and won't give Nvidia another design win. I just wonder how many money is Nvidia losing with the Mi3 deal.

Qualcomm is going to update its custom cores, improve their already outstanding GPU and move to 20nm before Nvidia. Tegra is set for another year losing buckets of money.

 

[ams23]

@ raghu78: As I mentioned earlier, the Xiaomi Mi3 design win with China Mobile is pretty big news for Tegra 4. The China Mobile Mi3 variant will sell in the millions of units (!) over the course of it's lifespan. And if you add in worldwide sales for upcoming Asus Transformer Pad Infinity, HP Slatebook x2, and the rest of consumer/embedded/automotive shipments, the end result is not too bad considering that Tegra 4 was delayed and considering that Windows on ARM has not achieved traction.

At the end of the day, Kepler.M has already been measured to have 3x better GFXBench 2.7 perf. per watt than the A6X GPU. That is a measurement, not a promise nor a projection nor a marketing claim. I stand by my words.

 

[Arkadrel]

http://www.xbitlabs.com/news/mobile...Application_Processors_Drop_Dramatically.html

..the company faced declines in sales of Tegra 3 application processors and could not offset the drop with any new revenues from new-generation products. The Tegra SoC business had revenue of $52.6 million, down 49% sequentially and down 70.7% year over year.

Sales drop 70.7% compaired to last year, for the tegra line's.

I dont know about everyone else but loseing almost ~71% sales in a year seems drastic to me.

I think Nvidia are fine now, but their tegra line is dead in the water.
I could see nvidia giveing up on the mobile market space if they cannot manage to get sales to atleast pay for the R&D costs, in the near future.

 

[blackened23]

Exactly. No matter how anyone spins it, Tegra 4 was a failure. They have relatively small design wins and none of those SKUs from Toshiba or Asus will sell well. Oh and then there is Windows RT (excuse me while I laugh). Nobody wants RT. Nvidia doesn't have anything on the scale of the Nexus 7.

That does not bode well since the Tegra division has not made a profit YET. The bottom line is this. Nvidia cannot treat the ARM SOC market like the PC market. Performance is great and all, but nvidia cannot enter the market late even if the product in question performs well and simply expect consumers to line up. The mobile market is very, very different and this will be a cause for some growing pains with nvidia - the Tegra 4 doesn't have any high volume design wins. And that is just unfortunate because nvidia will not be able to recover any of the R+D from T4 and will just further their Tegra related losses.

Tegra 5 has to reverse this trend, it will be a make it or break it type of deal for nvidia's Tegra division - they have to offer the same integration that Qualcomm does. It has to be price competitive. It has to perform well. Most important, it has to be released on time. Having a high performing product that is released late will just ensure that your product gets slaughtered, which is exactly what happened to T4.

 

[ams23]

@ Arkadrel: Tegra has historically had a revenue trough when prior gen Tegra ramps down in anticipation of next gen Tegra ramping up. This is already well understood. With Tegra 4, the situation was exacerbated with the delay in ramp up of Tegra 4. Again, well understood. Tegra revenue will start going up again over the next two fiscal quarters, but will still be well short for the entire fiscal year relative to last year. That said, due to heavily leveraged investments, the actual operating loss should be less this fiscal year vs. prior fiscal year! The expectation is that Tegra will clearly be profitable starting next fiscal year and moving forward due to leveraged investment cost and revenue growth (note that Tegra Automotive alone will reach $400-450 revenue within two fiscal years from now).

The investments that companies such as NVIDIA and Intel have made in low power processors and baseband technology over the last few years will begin to pay off in a big way in the very near future.

 

[rainy]

The investments that companies such as NVIDIA and Intel have made in low power processors and baseband technology over the last few years will begin to pay off in a big way in the very near future.

It's quite funny that you have put both companies in the same context because they're in completely different weight categories.

Imagine, that Nvidia loosing whole GPU market (including professional of course) in a single day: they don't exist anymore as company.

 

[ams23]

Investing in ultra low power processors and baseband technology starting years ago is more about foresight and financial resources than about "weight" category. NVIDIA's strength is GPU and Visual Computing, Intel's strength is CPU Computing, and these will continue to be strengths moving forward (irrespective of the size of each company relative to each other). Anyway, getting rid of Tegra would have relatively little financial impact at this point in time for NVIDIA, but it would be a very short-sided thing to do considering how important mobile computing will be in the future.

 

[Mondozei]

Xiaomi's newest phone, has a Tegra 4 SoC. Xiaomi now has a larger market share than Apple and growing very fast, and they're already thinking of international expansion.

Tegra 5 is due out in Q1 2014. Their supercomputing business does fine.

I think Nvidia is in a stable condition, although their major miss with Tegra 4 was coming out of the gate so late. They won't repeat that with Tegra 5.

 

[Mondozei]

Nvidia said that Tegra 4 would beat the crap out of the S800. Guess what happened? It's the same every single time with Nvidia mocking and putting down the competition and being burned later by reality. Mobile OEMs realized this a while ago and won't give Nvidia another design win. I just wonder how many money is Nvidia losing with the Mi3 deal.

Qualcomm is going to update its custom cores, improve their already outstanding GPU and move to 20nm before Nvidia. Tegra is set for another year losing buckets of money.

The notorious Qualcomm fangirl/boy strikes again. I remember you from the Intel thread. It's almost as if you are paid, but I guess that would be to overestimate you.

 

[blackened23]

The notorious Qualcomm fangirl/boy strikes again. I remember you from the Intel thread. It's almost as if you are paid, but I guess that would be to overestimate you.

When one cannot argue with reason, resort to insults? Come on dude, you can do better than that. The truth of the matter is that there is a basis of reality in what was stated. The Qualcomm snapdragon beat the Tegra 4 soundly and won all of the designs that mattered. All of the designs that will move millions upon millions of units. Maybe Tegra 5 will turn things around, but there is no one with an objective mindset who thinks that T4 was a success.

I've stated this before but NV cannot treat the ARM SOC market like the PC market. You can't charge a premium, have less integration (than qualcomm), and deliver a late product while simultaneously expecting consumers to go along with it. The mobile market doesn't play along like that. NV can get away with that in the PC dGPU market, but things are fairly different in the ARM SOC race. I'm sure NV is well aware of this and they need to properly execute with the Tegra 5 where the T4 failed - hopefully they can turn things around. In fact, I fully expect them to not make the same mistake twice.

 

[Mondozei]

The bottom line is this. Nvidia cannot treat the ARM SOC market like the PC market. Performance is great and all, but nvidia cannot enter the market late even if the product in question performs well and simply expect consumers to line up. The mobile market is very, very different and this will be a cause for some growing pains with nvidia - the Tegra 4 doesn't have any high volume design wins. And that is just unfortunate because nvidia will not be able to recover any of the R+D from T4 and will just further their Tegra related losses.

Tegra 5 has to reverse this trend, it will be a make it or break it type of deal for nvidia's Tegra division - they have to offer the same integration that Qualcomm does. It has to be price competitive. It has to perform well. Most important, it has to be released on time. Having a high performing product that is released late will just ensure that your product gets slaughtered, which is exactly what happened to T4.

You more or less nailed it. T5 will be a make or break moment. T4 wasn't a bad chip, but it wasn't beating S800, it was/is around S600 and with some software optimizations, it should be placed in the space between them. The major problem was that it was simply too late.

T5 in phones will not have 400+ gigaflops/s, more like half that. The clock speed will also be lower than the tablet version to watch the thermal envelope.

But even so, it looks very, very promising and Nvidia is launching it in Q1 2014, so they've learned their lesson on timing.

I really want to see what Qualcomm have up their sleeve, though. We're in that phase right now in the GPU space where we are seeing major increases each generation, sometimes 3x year over year gains. We're seeing that with Intel's integrated graphics, we're seeing that with Tegra. Qualcomm did a GPU which was around 80% faster with the Krait 400 cores. I hope they can do 2x or 3x faster next year.

Will be awesome for consumers and force everyone else to up their game.

 

[Mondozei]

When one cannot argue with reason, resort to insults? Is that where we're at now? The truth of the matter is that there is a basis of reality in what was stated. The Qualcomm snapdragon beat the Tegra 4 soundly and won all of the designs that mattered. All of the designs that will move millions upon millions of units. Maybe Tegra 5 will turn things around, but there is no one with an objective mindset who thinks that T4 was a success.

I've stated this before but NV cannot treat the ARM SOC market like the PC market. You can't charge a premium, have less integration (than qualcomm), and deliver a late product while simultaneously expecting consumers to go along with it. The mobile market doesn't play along like that. NV can get away with that in the PC dGPU market, but things are fairly different in the ARM SOC race. I'm sure NV is well aware of this and they need to properly execute with the Tegra 5 where the T4 failed - hopefully they can turn things around. In fact, I fully expect them to not make the same mistake twice.

See my above reply. I am in agreement with much, though not all, of what you wrote earlier.

Imouto is another case, though. He/she is a dedicated Qualcomm fanboy/girl who isn't really interested in a reasonable discussion but instead hijacking the threads with fanboy/girlism. That was the case with the Bay Trail thread where having a reasonable discussion started to become impossible after a while and I'd hate to see this thread turn in a similar path.

Note, this doesn't mean that I disagree that, say, T4 lost out to S800 and so on, which I already wrote. But there's one thing to judge things on a case-by-case basis and quite another to be reflexsively pro or anti something no matter what.

I'm not going to rule out T5 just because it isn't Qualcomm that makes it. I think it has a decent chance to be competitive and I'm curious to see what Qualcomm will have to offer in return. In short: I want to keep an open mind.

That's the opposite of what fanboys/fangirls do.

Warning issued for member callout and personal attack.
-- stahlhart

 

[Mondozei]

Sorry but Logan in mid-2014 is nothing great. Already companies like Mediatek are using Imagination PowerVR Series 6 ‘Rogue’ based GPUs in ARM SOCs. These chips are already in production and will be available in products in Q4 2013.

http://www.imgtec.com/news/Release/index.asp?NewsID=797
http://www.youtube.com/watch?v=Kf_ZGDoEPMo

Apple's upcoming A7 chip will also feature a Rogue based GPU. By Q4 2014 Apple and Qualcomm will have their first 20nm SOC chips shipping. Nvidia is always behind the game in Tegra. Nvidia don't switch to the leading edge node like Apple , Qualcomm and so Logan will fight against 20nm SOC chips for the most part of its 1 year lifecycle. Also Nvidia has never been able to compete on perf/watt when compared to Qualcomm, Apple. With Intel entering the phone market in a big way with Baytrail 22nm SOC chips it gets even more difficult.

That's interesting, I was going to mention both.

But one thing that should make you a bit cautious about that MediaTek PowerVR 6 Rogue video is this: in their CPU tests, 'competitor C' is clearly an iPad. You can see it on the shape. And the iPad 4 CPU is nothing great.

But then you move on to the GPU test on all of the sudden, competitor C(a.k.a. iPad 4) is removed and you only have A and B left.

I would wait for independent testing before jumping to conclusions. But Rogue does, at least theoretically, look very, very promising and could easily undercut Tegra 5 if it lives up to what's promised.

Mobile GPUs in 2014 are going to be the most interesting space to watch.

 

[ams23]

One of the main reasons for Qualcomm's strength in this space is because of their strength in [4G LTE] baseband processors. This is one of NVIDIA's biggest challenges right now. Keeping up with Qualcomm's CPU/GPU performance is a piece of cake compared to keeping up with their baseband integration and certification. NVIDIA needs to get their Icera i500 4G LTE software-defined modem to market and certified across different wireless carriers to expand their market opportunities. If this baseband strategy comes together with Logan in Tegra 5, then watch out, because new opportunities will arise from there.

 

[AnandThenMan]

"When you don't have a product worth talking about, you talk about the future"

"Never met a Powerpoint presentation I didn't like"

Describes Tegra really well. Nvidia needs a breakthrough to bring Tegra up to competitive speed. Will it happen? I have my doubts, we've been hearing about design wins, leaps in perf/watt and all that for quite a few product cycles now. Instead of closing the gap, Tegra sales are cratering and financial losses are increasing.

Ultra mobile is an incredibly tough market to crack, not even Intel with near unlimited resources has figured it out.

 

[Imouto]

Bay Holy Grail will achieve nothing.

The Tegra 4 story will repeat again with the Tegra 5.

Remember these words and come back in one year. I'm always right. Always.

Now you can can go on with your delusions and hopeless expectatives. Seriously, go on.

 

[ams23]

Imouto trolling yet again about Tegra, go figure Sorry to disappoint you, but Kepler.M perf. per watt has been measured, Logan is already sampling, and it is very obvious that Kepler.M will be a force to be reckoned with since perf. per watt directly translates to performance in a power constrained environment.

Warning issued for callout.
-- stahlhart

 

[AnandThenMan]

^^^ none of that means a damn thing until the product is in peoples hands.