Monday, June 17th 2019

NVIDIA Brings CUDA to ARM, Enabling New Path to Exascale Supercomputing

NVIDIA today announced its support for Arm CPUs, providing the high performance computing industry a new path to build extremely energy-efficient, AI-enabled exascale supercomputers. NVIDIA is making available to the Arm ecosystem its full stack of AI and HPC software - which accelerates more than 600 HPC applications and all AI frameworks - by year's end. The stack includes all NVIDIA CUDA-X AI and HPC libraries, GPU-accelerated AI frameworks and software development tools such as PGI compilers with OpenACC support and profilers. Once stack optimization is complete, NVIDIA will accelerate all major CPU architectures, including x86, POWER and Arm.

"Supercomputers are the essential instruments of scientific discovery, and achieving exascale supercomputing will dramatically expand the frontier of human knowledge," said Jensen Huang, founder and CEO of NVIDIA. "As traditional compute scaling ends, power will limit all supercomputers. The combination of NVIDIA's CUDA-accelerated computing and Arm's energy-efficient CPU architecture will give the HPC community a boost to exascale."
"Arm is working with our ecosystem to deliver unprecedented compute performance gains and exascale-class capabilities to Arm-based SoCs," said Simon Segars, CEO of Arm. "Collaborating with NVIDIA to bring CUDA acceleration to the Arm architecture is a key milestone for the HPC community, which is already deploying Arm technology to address some of the world's most complex research challenges."

According to the Green500 list released today, NVIDIA powers 22 of the world's 25 most energy-efficient supercomputers.

Key factors making this possible are: the ability of NVIDIA GPU-powered supercomputers to offload heavy processing jobs to more energy-efficient parallel processing CUDA GPUs; NVIDIA's collaboration with Mellanox to optimize processing across entire supercomputing clusters; and NVIDIA's invention of SXM 3D-packaging and NVIDIA NVLink interconnect technology, which allows for extremely dense scale-up nodes.

NVIDIA's support for Arm-based HPC systems builds on more than 10 years of collaboration. NVIDIA uses Arm for several of its system on a chip products available for portable gaming, autonomous vehicles, robotics and embedded AI computing.
Add your own comment

24 Comments on NVIDIA Brings CUDA to ARM, Enabling New Path to Exascale Supercomputing

#2
TheGuruStud
ZoneDymo, post: 4066100, member: 66089"
going super critical
Cars didn't work, so let's try arm HPC!
Posted on Reply
#3
yakk
Since nvidia has no choice in the matter of CPUs, and they have to do it by themselves (nobody would want to partner with them if they can avoid it) I guess they are trying to make do with ARM processors the best they can. Don't see it working out though.
Posted on Reply
#4
john_
Many many years ago I was expecting Nvidia to put all it's bets on the ARM architecture. To have by now full Windows/Linux PCs running on Tegra processors, gaming platforms based on Tegra processors running hi end Nvidia cards, a way to get independent of the X86 architecture. Instead, Nvidia almost killed Tegra, before finding a place in the auto industry for it, let Qualcomm get all the glory of Windows PCs on ARM and only power Switch as a 100% Nvidia system. Nvidia really lacks vision, even considering all their job into creating high performing, high energy efficient GPUs, they really lack vision. Even their efforts in the AI is something that can collapse really fast if better ways than GPUs are used in the near future.
Posted on Reply
#5
Vayra86
john_, post: 4066121, member: 137560"
Many many years ago I was expecting Nvidia to put all it's bets on the ARM architecture. To have by now full Windows/Linux PCs running on Tegra processors, gaming platforms based on Tegra processors running hi end Nvidia cards, a way to get independent of the X86 architecture. Instead, Nvidia almost killed Tegra, before finding a place in the auto industry for it, let Qualcomm get all the glory of Windows PCs on ARM and only power Switch as a 100% Nvidia system. Nvidia really lacks vision, even considering all their job into creating high performing, high energy efficient GPUs, they really lack vision. Even their efforts in the AI is something that can collapse really fast if better ways than GPUs are used in the near future.
They didn't kill Tegra though did they? My impression is they try to find a use for it and fail at every turn.
Posted on Reply
#6
TheGuruStud
Vayra86, post: 4066140, member: 152404"
They didn't kill Tegra though did they? My impression is they try to find a use for it and fail at every turn.
Yep, too power hungry and lacking perf.

Samsung going AMD should have Nvidia very worried lol. Samsung is chipzilla.
Posted on Reply
#7
john_
Vayra86, post: 4066140, member: 152404"
They didn't kill Tegra though did they? My impression is they try to find a use for it and fail at every turn.
Nvidia is only after markets with high profit margins. That's their biggest problem. That's why they can't find a viable, high volume market for a low cost ARM SOC that they will also like.

Only exception was Switch and that's probably because they had a stockpile of Tegra chips, not knowing what to do with them.
Posted on Reply
#8
jabbadap
TheGuruStud, post: 4066103, member: 42692"
Cars didn't work, so let's try arm HPC!
Uhm, they released Cuda for arm, not that they are going to make Tegra ARM HPCs by themselves. Just that ARM servers can use their gpus for computing.

This news piece is just missing those comments from HPC industry, which are on Nvidia's press release. So there are some shift to arm on future HPC projects and Nvidia are just making their gpus available for such a endeavors. In short this has nothing to do with Tegra.
Posted on Reply
#9
TheLostSwede
john_, post: 4066121, member: 137560"
Many many years ago I was expecting Nvidia to put all it's bets on the ARM architecture. To have by now full Windows/Linux PCs running on Tegra processors, gaming platforms based on Tegra processors running hi end Nvidia cards, a way to get independent of the X86 architecture. Instead, Nvidia almost killed Tegra, before finding a place in the auto industry for it, let Qualcomm get all the glory of Windows PCs on ARM and only power Switch as a 100% Nvidia system. Nvidia really lacks vision, even considering all their job into creating high performing, high energy efficient GPUs, they really lack vision. Even their efforts in the AI is something that can collapse really fast if better ways than GPUs are used in the near future.
They got scared when they failed in the mobile space, as they didn't understand their customers, or the market. Throwing money at a problem you don't understand, doesn't solve it.
Posted on Reply
#12
Fluffmeister
In other Nvidia news....

Nvidia's Lead Exceeds Intel's in Cloud

Nvidia holds a more commanding share of dedicated accelerators than Intel holds in overall compute among the top four public cloud services.

Nvidia’s GPUs now account for 97.4% of infrastructure-as-a-service (IaaS) instance types of dedicated accelerators deployed by the top four cloud services. By contrast, Intel’s processors are used in 92.8% of compute instance types, according to one of the first reports from Liftr Cloud Insights’ component tracking service.

https://www.eetimes.com/author.asp?section_id=36&doc_id=1334812&_mc=RSS_EET_EDT#
Posted on Reply
#13
zlobby
Fluffmeister, post: 4066224, member: 101373"
In other Nvidia news....

Nvidia's Lead Exceeds Intel's in Cloud

Nvidia holds a more commanding share of dedicated accelerators than Intel holds in overall compute among the top four public cloud services.

Nvidia’s GPUs now account for 97.4% of infrastructure-as-a-service (IaaS) instance types of dedicated accelerators deployed by the top four cloud services. By contrast, Intel’s processors are used in 92.8% of compute instance types, according to one of the first reports from Liftr Cloud Insights’ component tracking service.

https://www.eetimes.com/author.asp?section_id=36&doc_id=1334812&_mc=RSS_EET_EDT#
Yes, thank you for the bold, 48pt headline. We would have surely missed it if it weren't for the emphasis!
Posted on Reply
#14
Fluffmeister
zlobby, post: 4066232, member: 172939"
Yes, thank you for the bold, 48pt headline. We would have surely missed it if it weren't for the emphasis!
You were yawning away there, I didn't want you to miss it. :laugh:
Posted on Reply
#15
silentbogo
TheGuruStud, post: 4066103, member: 42692"
Cars didn't work, so let's try arm HPC!
It's the same thing they've been doing for the past decade, only now for ARM host CPUs with multi-node support. Same thing they were doing for all other platforms, the only difference is that now they are probably working on a datacenter-oriented all-in-one software package.

yakk, post: 4066115, member: 158293"
Don't see it working out though.
L4T+OpenMPI not working either? That's a big surprise...

Vayra86, post: 4066140, member: 152404"
They didn't kill Tegra though did they? My impression is they try to find a use for it and fail at every turn.
Nope, and I'm certain another gen is on the way.
Not sure where you, guys, get the "failed" part. It was an awful platform for portables, but "surprisingly" a commercial success due to its full-featured GPU. Tegra 3 was in Nexus 7 tablets - the most popular and best-selling tablet of its time, K1 was in Shield tablet and several popular chromebook models (only beaten by Samsung XE303C12 as the only "other" ARM alternative). Tegra X1 was relatively expensive and only made its way into a Shield TV, but X2 on the other hand is currently powering the best-selling console on the market. All of the above does not even include automotive applications (Drive PX/CX etc), and thousands of Jetson dev kits they sourced for educational institutions and sold to hobbyists and CV/AI/Robotics researchers.
Posted on Reply
#16
zlobby
Fluffmeister, post: 4066234, member: 101373"
You were yawning away there, I didn't want you to miss it. :laugh:
Valid point, thank you. I am now enlightened. It appears you headline did grab my attention.
Posted on Reply
#17
jabbadap
silentbogo, post: 4066245, member: 141875"
It's the same thing they've been doing for the past decade, only now for ARM host CPUs with multi-node support. Same thing they were doing for all other platforms, the only difference is that now they are probably working on a datacenter-oriented all-in-one software package.


L4T+OpenMPI not working either? That's a big surprise...


Nope, and I'm certain another gen is on the way.
Not sure where you, guys, get the "failed" part. It was an awful platform for portables, but "surprisingly" a commercial success due to its full-featured GPU. Tegra 3 was in Nexus 7 tablets - the most popular and best-selling tablet of its time, K1 was in Shield tablet and several popular chromebook models (only beaten by Samsung XE303C12 as the only "other" ARM alternative). Tegra X1 was relatively expensive and only made its way into a Shield TV, but X2 on the other hand is currently powering the best-selling console on the market. All of the above does not even include automotive applications (Drive PX/CX etc), and thousands of Jetson dev kits they sourced for educational institutions and sold to hobbyists and CV/AI/Robotics researchers.
It is and it's called Orin. And no Tegra X2(Parker) is the one which lived only in automotive/AI/ML and has now superseded buy Xavier SoC. Nintendo Switch and ShieldTVs uses dead old Tegra X1 chips from the year 2015.
Posted on Reply
#18
silentbogo
jabbadap, post: 4066256, member: 148195"
And no Tegra X2(Parker) is the one which lived only in automotive/AI/ML and has now superseded buy Xavier SoC. Nintendo Switch and ShieldTVs uses dead old Tegra X1 chips from the year 2015.
My bad. Now I remember, it was an X1.
Posted on Reply
#19
jabbadap
silentbogo, post: 4066261, member: 141875"
My bad. Now I remember, it was an X1.
Yeah that 20nm little SoC must have been the most successful for them in the whole Tegra history. Heck the newest product out of it came just this years GTC: Tegra Jetson Nano. Four years after the soc introduction.
Posted on Reply
#20
Camm
Fluffmeister, post: 4066224, member: 101373"
In other Nvidia news....

Nvidia's Lead Exceeds Intel's in Cloud
Haha, marketing at its finest considering GPGPU compute is a subsection of the compute market. Considering the deals AMD signed for supply to Microsoft, Google, and Alibaba though, I can't see that near total dominance lasting for much longer either however.
Posted on Reply
#21
ratirt
I don't know about you but I got a feeling that NV is not struggling. I know their market share is very big but it would seem that NV is kinda left alone in there. Looking for a "hook up" so bad. Like it was always premium and you have had to pay butt load of cash now it is being given away and NV is with it :)
How time change :)
Posted on Reply
#22
Midland Dog
maxwell and by extension pascal both have arm cpus on board
Posted on Reply
#23
renz496
john_, post: 4066121, member: 137560"
Many many years ago I was expecting Nvidia to put all it's bets on the ARM architecture. To have by now full Windows/Linux PCs running on Tegra processors, gaming platforms based on Tegra processors running hi end Nvidia cards, a way to get independent of the X86 architecture. Instead, Nvidia almost killed Tegra, before finding a place in the auto industry for it, let Qualcomm get all the glory of Windows PCs on ARM and only power Switch as a 100% Nvidia system. Nvidia really lacks vision, even considering all their job into creating high performing, high energy efficient GPUs, they really lack vision. Even their efforts in the AI is something that can collapse really fast if better ways than GPUs are used in the near future.
lacks vision? it is important to have vision but it is even more important to keep the company alive than chasing something that will not going to work. windows for ARM nvidia probably one of the best SoC vendor when MS first push that initiatives. i still remember when device with nvidia tegra 3 was the only device you can found at computex being shown from several vendor running windows. other SoC maker at the time can't show anything because their software are not ready. also the reason why MS pick nvidia tegra 3 to power their original surface and tegra 4 for surface 2. back then Qualcomm is quite sour their snapdragon are not chosen by MS for surface and they publically bash surface for being "slower" than device using their snapdragon. nvidia did not kill tegra. if anything when nvidia do something they do it very seriously. the primary issue on mobile is Qualcomm dominance in modem/baseband tech. nvidia is the type that know when it is the time to retreat so it will not going to ruin the whole company. just look at intel. they were so keen on taking a hold on mobile market to the point they are using contra revenue strategy. look where it get them now. failed to make any meaningful impact on mobile and their desktop market are being pressured by AMD.
Posted on Reply
#24
zlobby
renz496, post: 4066760, member: 104135"
lacks vision? it is important to have vision but it is even more important to keep the company alive than chasing something that will not going to work. windows for ARM nvidia probably one of the best SoC vendor when MS first push that initiatives. i still remember when device with nvidia tegra 3 was the only device you can found at computex being shown from several vendor running windows. other SoC maker at the time can't show anything because their software are not ready. also the reason why MS pick nvidia tegra 3 to power their original surface and tegra 4 for surface 2. back then Qualcomm is quite sour their snapdragon are not chosen by MS for surface and they publically bash surface for being "slower" than device using their snapdragon. nvidia did not kill tegra. if anything when nvidia do something they do it very seriously. the primary issue on mobile is Qualcomm dominance in modem/baseband tech. nvidia is the type that know when it is the time to retreat so it will not going to ruin the whole company. just look at intel. they were so keen on taking a hold on mobile market to the point they are using contra revenue strategy. look where it get them now. failed to make any meaningful impact on mobile and their desktop market are being pressured by AMD.
Other SoC makers don't want (yet) to even step in this territory, while nvidia needs this desperately. Can't and don't want to are 2 completely different things.
Posted on Reply
Add your own comment