Tweaktown : Nvidia Jetson (Tegra K1)

[altairauditore]

http://www.phoronix.com/scan.php?page=article&item=nvidia_tegrak1_preview&num=1

 

[ams23]

Leaked 7.9" TK1 tablet GFXBench results:

 

[Enigmoid]

Wow looks good. 60 fps on t-rex like Nvidia said

Need to see power but its good that they are hitting targets on a 8" tablet.

 

[ams23]

The R3 Cortex A15 CPU in Tegra K1 can clock as high as 2.3GHz, but using a maximum CPU clock operating frequency of ~ 2.1GHz results in much lower power consumption with only a small reduction in performance, so it is a good choice for a mobile product.

Here are the Geekbench 2.4.3 results for NVIDIA Mocha: http://browser.primatelabs.com/geekbench2/2431774

As a point of comparison, these Geekbench 2.4.3 results for Tegra K1 @ 2.12GHz are ~ 30% higher than some of the fastest results I could find for Snapdragon 800 @ 2.15GHz: http://browser.primatelabs.com/geekbench2/2210469

 

[Sweepr]

Wow looks good. 60 fps on t-rex like Nvidia said

Need to see power but its good that they are hitting targets on a 8" tablet.

Indeed. In comparison, Mullins scores 37 FPS, Snapdragon 801 close to 30 FPS and Bay Trail-T close to 20 FPS (C0 stepping).
I dont think Apple will catch up with A8 and the S805's 40% better shader performance also wont come close. Perhaps Cherry Trail-T has a shot (4x as many EUs as BT-T, new architecture) but thats late 2014.

 

[ams23]

Indeed. In comparison, Mullins scores 37 FPS, Snapdragon 801 close to 30 FPS and Bay Trail-T close to 20 FPS (C0 stepping).

Mullins in a 7.9" tablet would likely be much much slower than 37fps. The 37fps Trex offscreen score that you saw was in an 11.6" tablet using the highest performance Mullins variant and with AC power plugged in. Note that when this Mullins variant is used on AC power, it boosts CPU/GPU clock operating frequencies to boost performance. When used on battery power, it lowers CPU/GPU clock operating frequencies.

 

[jpiniero]

Yeah, I think it's very likely that the power draw is going to be too much for a tablet people might actually buy.

A better comparison is probably the Surface Pro 2 (15 W Core i5 4200U), and that got 63 fps.

 

[jdubs03]

Yeah, I think it's very likely that the power draw is going to be too much for a tablet people might actually buy.

A better comparison is probably the Surface Pro 2 (15 W Core i5 4200U), and that got 63 fps.

Hopefully this will quiet down all of the hecklers saying Nvidia has no shot at gaining market share. I would be very surprised if the TK1 isn't in the next Surface version, Microsoft would be making a big mistake with the choice of the S805 for a Surface 3. The mini fine, use Qualcomm, but the larger version I hope retain Nvidia as the SOC vendor.

There hasn't been anything I've seen that hasn't impressed me in regards to TK1. And we're only talking A15r3. Denver will probably be the choice for the full size Surface 3. To be honest, I'm more interested in the CPU performance of Denver (extrapolating the A15 only goes so far). With that GPU, mixed in with damn good single-thread performance, Denver may be a game-changer. Again, I would be disappointed if they didn't use Denver in the Surface 3, Windows would fly on that thing.

 

[altairauditore]

Looking at GFXBench Manhattan Offscreen results, the current chart topper is the A7 with 13FPS. If the TK! can do 30fps, thats twice as fast!

 

[itsmydamnation]

Mullins in a 7.9" tablet would likely be much much slower than 37fps. The 37fps Trex offscreen score that you saw was in an 11.6" tablet using the highest performance Mullins variant and with AC power plugged in. Note that when this Mullins variant is used on AC power, it boosts CPU/GPU clock operating frequencies to boost performance. When used on battery power, it lowers CPU/GPU clock operating frequencies.

link for this please as i have never seen or heard of that, because you know you are constrained by heat................

 

[caswow]

http://m.ca.ign.com/articles/2014/05/07/microsoft-reportedly-ditching-nvidias-tegra-for-new-surface

Doesnt seem like ms wants to have tegra in its devices anymore

 

[sontin]

So, they are canceling the Surface 2?

 

[note235]

the clock of the a7 is much lower than the k1 non?

 

[North01]

Leaked 7.9" TK1 tablet GFXBench results:

Before anyone asks "why is it only 2x and not close to the ~2.65x shown by NVIDIA?"

NVIDIA had shown us the benchmarks in January.

When the iPad Mini Retina (and other A7 devices) had been benchmarked towards the end of January in Manhattan Offscreen 1080p, they had average scores closer to 667 (10.8 Fps). Based off of this score, the Tegra K1 was offering a score ~2.65x that of the A7.

Sometime around March 10th, the average score for Apple's A7 had changed to 803 (13.0 Fps). I assume there was some sort of software update that boosted the Manhattan score.

 

[ams23]

link for this please as i have never seen or heard of that, because you know you are constrained by heat................

It came from the exact same review where 37fps was measured in the first place (HotHardware IIRC?).

 

[ams23]

Doesnt seem like ms wants to have tegra in its devices anymore

That is a logical fallacy. The 64-bit Tegra K1 variant was already rumored to be powering the next gen Surface 3.

 

[itsmydamnation]

It came from the exact same review where 37fps was measured in the first place (HotHardware IIRC?).

so in otherwords on all accounts you are lying :thumbsup:

http://hothardware.com/Reviews/AMD-...nstream-and-LowPower-2014-APUs-Tested/?page=1

Mullins in a 7.9" tablet would likely be much much slower than 37fps.

first lie, TDP is 4.5 watts SDP 2.8watts

The 37fps Trex offscreen score that you saw was in an 11.6" tablet using the highest performance Mullins variant and with AC power plugged in.

second lie, no where in the reivew is it state its plugged into AC, also see link on your first lie.

Note that when this Mullins variant is used on AC power, it boosts CPU/GPU clock operating frequencies to boost performance

third lie. again see first lie link, 2nd see the power management section See CPu-Z of different load states and how they match AMD specfications

When used on battery power, it lowers CPU/GPU clock operating frequencies.

forth lie, again see all the links above.
:thumbsup:

 

[ams23]

LOL, no no, you are clueless, it was actually from the Techspot review (not HotHardware at all)! My goodness. Remember I said the same review where the data point first came from! Well, serves me right for not recalling the name of the website, lol. Anyway, it is a moot point because the Tegra K1 GPU appears to have truly superior perfomance and perf. per watt compared to any variant of Mullins. And FWIW, unless you can prove otherwise, I stand by my earlier assertion that Mullins would likely need significantly reduced boost clock operating frequencies to work fine in a thin 7.9" tablet.

 

[strata8]

I wonder when NVIDIA will give reviewers hands-on time with a reference platform.

 

[sontin]

A few power numbers from Jetson K1:
http://www.phoronix.com/forums/show...red-To-AMD-AM1-APUs/page5&p=416287#post416287http://www.phoronix.com/forums/show...red-To-AMD-AM1-APUs/page5&p=416287#post416287

 

[altairauditore]

Seems to be pretty good. The board is running additional hardware which also guzzles power.

The System + Fan seems to be using around 5W at load, leaving around 5-6W unaccounted for. At load, the voltage regulator (12v --> 1v , 5v... etc) ) seems to be wasting around 1.8 W by itself. That leaves us around 3.5 - 4W.

Check the Jetson Brief (Pg 13):
http://developer.download.nvidia.com/embedded/jetson/TK1/docs/Jetson_platform_brief_May2014.pdf

Nvidia states that the OS is not tuned for power efficiency, Jetson uses DDR3L and its governors are not for efficiency again.

So according to them Tk1+ RAM = 7W on the Jetson (peak load).
Tk1+RAM = 3.6W on the Jetson (constrained)
Tk1+RAM = 2.6W (Nvidia's Mobile Platform 45% faster than A7)

Seems like TK1 will be a winner if Nvidia manages to sell it.

btw phoronix was using a mouse, keyboard, webcam! + an SD card, HDMI.

 

[strata8]

A few power numbers from Jetson K1:
http://www.phoronix.com/forums/show...red-To-AMD-AM1-APUs/page5&p=416287#post416287

Idle Less System : 0.05A ( 0.61W)

TEST1 : glmark2 -s 1920x1080 --off-screen
Avg. Less Sys : 0.18A ( 2.19W)

TEST2 : CUDA Smoke particle demo
Avg. Less Sys : 0.71A ( 8.26W)

TEST3 : VLC streaming 720p video from NAS GbE
Avg. Less Sys : 0.17A ( 2.01W)

A GPU benchmark using 1/4 the power of a CUDA demo and only 200mW more than 720p video streaming? Something's not right there.

 

[ams23]

NVIDIA's Tegra K1 will power the new Xiaomi MiPad ( http://www.anandtech.com/show/8022/xiaomi-announces-the-mipad-the-first-tegra-k1-device ):

 

[sontin]

29k points in 3DMark is nearly 50% faster than what Mullins gets in the Discovery tablet without the powerplug.

 

[Hitman928]

What's the source on those benches, I don't see them at anand or connecting links?