A Quick Test Of NVIDIA's "Carmel" CPU Performance

[phoronix]

Phoronix: A Quick Test Of NVIDIA's "Carmel" CPU Performance

NVIDIA's Tegra Xavier SoC is becoming more widely available now that the Jetson Xavier Development Kit has begun shipping. Besides this latest SoC being an exciting design with its Volta-based GPU and having a Tensor Processing Unit / Deep Learning Accelerator, it's exciting on the CPU side as well with NVIDIA's custom-designed ARMv8 "Carmel" CPU cores.

http://www.phoronix.com/vr.php?view=26891

[vegabook]

I have no clue why Nvidia wouldn't let the Shield TV run Linux.

[Michael]

I have no clue why Nvidia wouldn't let the Shield TV run Linux.

It can, or at least previously could, that's how I benchmarked the Tegra X1 originally: https://www.phoronix.com/vr.php?view=21936 Was given remote access to a SHIELD TV running Ubuntu, but not sure of how that was done or if it still would work.

[willmore]

I have no clue why Nvidia wouldn't let the Shield TV run Linux.

Or why we haven't seen an updated one in years. Or a tablet.....

[starshipeleven]

I have no clue why Nvidia wouldn't let the Shield TV run Linux.

They are control freaks. Considering that most of their businness is based on tightly controlling who can have what feature depending on the price he paid, it's not surprising.

[starshipeleven]

Or why we haven't seen an updated one in years. Or a tablet.....

low sales. TV and tablet markets are extremely saturated markets.

[RussianNeuroMancer]

Is there any previous Tegra board that stable with nouveau?

[ldesnogu]

First of all sorry for not directly using OpenBenchmarking, I need to learn it first :-)

On my TX2 I get this:

$ ./c-ray-mt -t 6 -s 3840x2160 -r 16 -i sphfract -o output.pp
c-ray-mt v1.1
Rendering took: 386 seconds (386646 milliseconds)

That's with default compiler (5.4.0) and with only -O3. The trick? Run this:

sudo nvpmodel -m 0

This will turn on both clusters, the 4 Cortex-A57 and the 2 Denver at max frequency. By default the 2 Denver cores are disabled.

If I get back to default mode:

$ sudo nvpmodel -m 3
$ ./c-ray-mt -t 4 -s 3840x2160 -r 16 -i sphfract -o output.pp
c-ray-mt v1.1
Rendering took: 568 seconds (568519 milliseconds)
$ ./c-ray-mt -t 16 -s 3840x2160 -r 16 -i sphfract -o output.pp
c-ray-mt v1.1
Rendering took: 539 seconds (539265 milliseconds)

[ldesnogu]

For primesieve:

$ sudo nvpmodel -m 3
$ ./primesieve 1e12
Sieve size = 32 KiB
Threads = 4
100%
Seconds: 297.812
Primes: 37607912018
sudo nvpmodel -m 0
./primesieve 1e12
Sieve size = 32 KiB
Threads = 6
100%
Seconds: 163.736
Primes: 37607912018

This beats the Xavier board.

[Stefem]

First of all sorry for not directly using OpenBenchmarking, I need to learn it first :-)

On my TX2 I get this:



That's with default compiler (5.4.0) and with only -O3. The trick? Run this:



This will turn on both clusters, the 4 Cortex-A57 and the 2 Denver at max frequency. By default the 2 Denver cores are disabled.

If I get back to default mode:

Really interesting numbers