Raspberry Pi cluster vs Graphic Card

koch44

jr. member

Activity: 54

Merit: 11

Quote from: keychainX on May 13, 2019, 06:41:01 AM

Quote from: koch44 on May 13, 2019, 03:06:56 AM

Quote from: ?? on ??

Quote from: charlie137 on May 10, 2019, 04:29:50 PM

regarding NVIDIA Jetson

Quote

472 GFLOPs of compute performance and consume as little as five watts.

https://www.techrepublic.com/article/raspberry-pi-style-jetson-nano-is-a-powerful-low-cost-ai-computer-from-nvidia/

Specifically 472 GFLOPS on FP16 and 236 GFLOPS on FP32^[1]. For comparison GTX 1060 performances are 68.36 GFLOPS on FP16 and 4.375 TFLOPS on FP32^[2].

Jetson Nano overwhelmingly win on FP16, but only because GTX 1060 focused on gaming which often uses FP32^[3]. So clearly even cluster of Jetson Nano can't beat a GTX 1060

Source :
1. https://devblogs.nvidia.com/jetson-nano-ai-computing/
2. https://www.techpowerup.com/gpu-specs/geforce-gtx-1060-6-gb.c2862
3. https://www.reddit.com/r/gamedev/comments/8rzsye/why_have_most_engines_settled_on_fp32_for/

Can this Jetson Nano run Ubuntu from a USB stick?

Yes, it runs a special version of Ubuntu

look here, this guy just did it https://www.reddit.com/r/JetsonNano/comments/bi5k8l/i_just_updated_my_os_for_the_jetson_nano_to/

/KX

thanks for the info! appreciate it

keychainX

member

Activity: 378

Merit: 53

Telegram @keychainX

Quote from: koch44 on May 13, 2019, 03:06:56 AM

Quote from: ?? on ??

Quote from: charlie137 on May 10, 2019, 04:29:50 PM

regarding NVIDIA Jetson

Quote

472 GFLOPs of compute performance and consume as little as five watts.

https://www.techrepublic.com/article/raspberry-pi-style-jetson-nano-is-a-powerful-low-cost-ai-computer-from-nvidia/

Specifically 472 GFLOPS on FP16 and 236 GFLOPS on FP32^[1]. For comparison GTX 1060 performances are 68.36 GFLOPS on FP16 and 4.375 TFLOPS on FP32^[2].

Jetson Nano overwhelmingly win on FP16, but only because GTX 1060 focused on gaming which often uses FP32^[3]. So clearly even cluster of Jetson Nano can't beat a GTX 1060

Source :
1. https://devblogs.nvidia.com/jetson-nano-ai-computing/
2. https://www.techpowerup.com/gpu-specs/geforce-gtx-1060-6-gb.c2862
3. https://www.reddit.com/r/gamedev/comments/8rzsye/why_have_most_engines_settled_on_fp32_for/

Can this Jetson Nano run Ubuntu from a USB stick?

Yes, it runs a special version of Ubuntu

look here, this guy just did it https://www.reddit.com/r/JetsonNano/comments/bi5k8l/i_just_updated_my_os_for_the_jetson_nano_to/

/KX

koch44

jr. member

Activity: 54

Merit: 11

Quote from: ?? on ??

Quote from: charlie137 on May 10, 2019, 04:29:50 PM

regarding NVIDIA Jetson

Quote

472 GFLOPs of compute performance and consume as little as five watts.

https://www.techrepublic.com/article/raspberry-pi-style-jetson-nano-is-a-powerful-low-cost-ai-computer-from-nvidia/

Specifically 472 GFLOPS on FP16 and 236 GFLOPS on FP32^[1]. For comparison GTX 1060 performances are 68.36 GFLOPS on FP16 and 4.375 TFLOPS on FP32^[2].

Jetson Nano overwhelmingly win on FP16, but only because GTX 1060 focused on gaming which often uses FP32^[3]. So clearly even cluster of Jetson Nano can't beat a GTX 1060

Source :
1. https://devblogs.nvidia.com/jetson-nano-ai-computing/
2. https://www.techpowerup.com/gpu-specs/geforce-gtx-1060-6-gb.c2862
3. https://www.reddit.com/r/gamedev/comments/8rzsye/why_have_most_engines_settled_on_fp32_for/

Can this Jetson Nano run Ubuntu from a USB stick?

odolvlobo

legendary

Activity: 4466

Merit: 3391

I haven't been keeping up with hash rates, but I wouldn't be surprised if a GPU is 50 - 100 times faster than a CPU. Given that an RPi is pretty wimpy, my guess is that it would take 100 - 200 Rpis to match a GPU.

charlie137

full member

Activity: 1204

Merit: 220

(ノಠ益ಠ)ノ

did you consider AWS instead?

regarding NVIDIA Jetson

Quote

472 GFLOPs of compute performance and consume as little as five watts.

https://www.techrepublic.com/article/raspberry-pi-style-jetson-nano-is-a-powerful-low-cost-ai-computer-from-nvidia/

ABCbits

legendary

Activity: 2870

Merit: 7490

Crypto Swap Exchange

Performance of a RPi3 very low, it's only as fast as Pentium III or IV (according to few benchmark i've seen as fast), there's no way it could beat GTX 1060.

But as a side note, NVIDIA release NVIDIA Jeston Nano which have powerful GPU (for barebone PC) with price $99, i don't know cluster of Jetson Nano could beat GTX 1060 for vanity address task, but it definitely far faster than RPi3.

mitsos2017

newbie

Activity: 49

Merit: 0

I would like to know (if possible) if a Raspberry Pi Cluster could compeed a Graphic Card in a vanity address search (as a cheaper alternative).
I'll make an example:
You can buy a 1060 with $250 usd and make a cluster of 5 RPi3 with the same amount.
If the card "finds" a vanity address in 1 hour, would be possible for the cluster to find it earlier?

Topic: Raspberry Pi cluster vs Graphic Card (Read 275 times)