Denver is not dead, he's just sleeping! : Tegra X1 uses A57

[NTMBK]

http://live.theverge.com/nvidia-ces-2015-liveblog/

8 core configuration, no mention of Denver. Sounds like A57+A53 big.LITTLE.

Update: Yes, it's A57 and A53. http://www.marketwired.com/press-re...-X1-Mobile-Super-Chip-NASDAQ-NVDA-1980112.htm

Update 2: NVidia claims that Denver will be back in a future chip:

The reason we did not use Denver on Tegra X1 is that we decided to take a "tick tock" approach to launching our processors. Since TX1 is on a brand new 20nm process we decided to use off the shelf ARM cores. But Tegra K1 was on a well understood [28nm] process and hence we decided to use Denver on it. We still have Denver on our roadmap and will be using it in our future chips as processes mature.

http://arstechnica.com/gadgets/2015/01/nvidia-announces-tegra-x1-soc-with-maxwell-based-gpu/

So my obituary may have been rather premature!

 

[bullzz]

ya was surprised they didnt mention denver. may be they are going the qualcomm way. use A57/53 cores till their cores are ready

 

[sciwizam]

Here's a link with benchmarks. In summary, ~2x Apple's A8X from the iPad Air 2

http://www.hardwarezone.com.sg/feature-preview-nvidia-tegra-x1-benchmark-results

 

[OBLAMA2009]

was denver supposed to be a higher powered/desktop chip? lol k1 showed thats going nowhere

 

[positivedoppler]

Probably not since some chinese firm is making the obox lol.

Very interesting though, it's based on TSMC 20 nm
http://www.anandtech.com/show/8811/nvidia-tegra-x1-preview/2

 

[antihelten]

As far as Denver being "dead", the X1 seemed to be highly geared towards computing in a car environment judging from the keynote.

This would seem to be a scenario where throwing in a ton of cores actually makes sense, given that you constantly have to do image processing from multiple cameras, plus 3 passengers playing around with their own dedicated touchscreens, plus whatever else you would normally do in a car. Thats a lot of fairly demanding tasks (especially the image processing) all running more or less constantly as you are driving.

As such it may just be a case of picking the right tool for this particular task, and there might be other less car focused SKUs using Denver, that are yet to be revealed.

 

[Arachnotronic]

ya was surprised they didnt mention denver. may be they are going the qualcomm way. use A57/53 cores till their cores are ready

Denver is already shipping. That NVIDIA didn't port it over to 20nm for X1 is telling.

 

[NTMBK]

Yup, it's A57. OP updated.

 

[Phynaz]

Honest question, has Nvidia had a successful mobile soc?

 

[NTMBK]

Honest question, has Nvidia had a successful mobile soc?

Hard to say. Their Tegra division reported massive losses every quarter, until they restructured their reports and hid the losses inside another division.

 

[Arachnotronic]

Hard to say. Their Tegra division reported massive losses every quarter, until they restructured their reports and hid the losses inside another division.

Nope, NVIDIA still reports Tegra P&L separately in its SEC filings to my knowledge.

It's still losing money, though the losses are narrowing.

 

[NTMBK]

Nope, NVIDIA still reports Tegra P&L separately in its SEC filings to my knowledge.

It's still losing money, though the losses are narrowing.

Huh, you're right. Not sure what I was thinking of!

 

[jdubs03]

the cpu analysis from the anandtech page is really interesting. basically only the cpu core itself is from arm, everything else is nvidia. plus there is a claim of 1.4x compared to the exynos 5433.

here is another site with benchmarks:

http://hothardware.com/news/hands-on-with-nvidia-tegra-x1-with-benchmarks-and-video

Reps from NVIDIA said they’re taking somewhat of a tick-tock strategy with Tegra, using off-the-shelf CPU cores and a new process node this time around; with the next-gen we may see custom CPU cores again.

in the benchmark we can see more of the details (idk why hardwarezone didn't post the details) for 3d mark; the unlimited physics score is ~15% faster than tk1, so that may establish the range of performance increase from the exynos (based on the physics score: +20%), other cpu tests i imagine it being a similar comparison.

maybe nvidia is waiting for 16FF+ to go for another shot at denver or denver-b. noting the tick-tock nature but a custom core too kind of doesn't make sense, staying at 20nm for tx1's successor would be a bad move. so i think 16nm is in order at least (armv8.1), then the next 16nm core being custom.

the gpu though, those scores are massive. hopefully they can push maxwell with the android tablet vendors and get some more wins. though interestingly, the graphics score isn't that much higher than the core m 5y70.

 

[Arachnotronic]

Huh, you're right. Not sure what I was thinking of!

Intel, perhaps? Which embedded mobile in "Other IA," then broke it out, and is now merging it into PC Client Group?

 

[scannall]

Here's a link with benchmarks. In summary, ~2x Apple's A8X from the iPad Air 2

http://www.hardwarezone.com.sg/feature-preview-nvidia-tegra-x1-benchmark-results

While it may in fact be faster, until a 3rd party does the benchmarks on a shipping device you really have to take those results with a large grain of salt. A reference board, plugged into the wall with no details about clock speed etc, with benchmarks run by NVidia.... Yeah, I think I'll wait on real benchmarks.

 

[jdubs03]

While it may in fact be faster, until a 3rd party does the benchmarks on a shipping device you really have to take those results with a large grain of salt. A reference board, plugged into the wall with no details about clock speed etc, with benchmarks run by NVidia.... Yeah, I think I'll wait on real benchmarks.

yea, but thats the whole thing about reference platforms, intel and qualcomm do it too. we might not see real reviews til june, so its a nice wait.

 

[seitur]

Honest question, has Nvidia had a successful mobile soc?

Tegra 3 possibly? (I think it was "3"?) It had a lot of adoption in many devices. Then Nvidia took forever to release next one that additionally underdelivered massively.

 

[sontin]

Except Tegra 4i everyone was successful.

BTW: The power numbers from TX1 shows why only Apple can compete with nVidia. They have the money to go straight to the next process node even if the price is still to high. Other companies like Qualcomm wont come even close to the GPU performance. TX1 gets more than 60FPS in the offscreen test of Manhatten while the dev plattform of 810 gets 23FPS...

http://www.mobiledroid.co.uk/blog/qualcomm-810-performance-preview/

 

[NTMBK]

Except Tegra 4i everyone was successful.

Oh really, they started actually making a profit?

 

[sontin]

Yes, everyone is making profit.

 

[witeken]

Nvidia doing its usual ridiculous marketing: comparing Tegra Maxwell to Core i7. Should have compared 4.5W Broadwell-Y to 10W Tegra M1. And secondly, since when is it apples to apples to compare 16bit to 32bit?

 

[NTMBK]

Yes, everyone is making profit.

And yet every quarter the Tegra division reports a loss.

 

[sontin]

And yet every quarter the Tegra division reports a loss.

Doesnt change the fact that their gross margins are positive.

 

[scannall]

I'm not sure why they'd go to a smaller, weaker CPU core though. Driver problems? Thermal?

 

[Hans de Vries]

Yet another thread with a wrong title: Denver isn't "dead" at all....

http://arstechnica.com/gadgets/2015/01/nvidia-announces-tegra-x1-soc-with-maxwell-based-gpu/

The reason we did not use Denver on Tegra X1 is that decided to take a "tick tock" approach to launching our processors. Since TX1 is on a brand new 20nm process we decided to use off the shelf ARM cores. But Tegra K1 was on a well understood [28nm] process and hence we decided to use Denver on it. We still have Denver on our roadmap and will be using it in our future chips as processes mature.