With its Radeon graphics technology, AMD is a powerhouse

Topic

New Reply

SILICON By Brian Livingston At today’s breakneck pace of technological change, the semiconductor industry’s product cycles run faster than a one-legge
[See the full post at: With its Radeon graphics technology, AMD is a powerhouse]

Reply | Quote

Replies

What about drivers? According to the experts some years ago, the AMD drivers were quite buggy; Nvidia did a much better job. What’s the current standing?

Reply | Quote

The current consensus in the tech world still seems to be that Nvidia has better drivers. I have a 5700xt and the latest Radeon WHQL driver (22.5.1) crashed my browser on occasion and I also lost some settings. I rolled back to 22.4.1 because that driver seemed more stable.  This has happened with more than one driver in the past as well.  It’s a little irritating but not enough to stop me from buying another Radeon GPU in the future if I feel it has good price/performance value.

Reply | Quote

In Windows, perhaps. In Linux, it’s a different story, and AMD’s support of Linux undoubtedly has played at least some role in their recent success (which is an incredible thing, as just a few years ago they were so far behind that it seemed they were headed for oblivion).

AMD and Intel have open-sourced their Linux drivers. Intel, for all the talk of it being a villain, is one of the leading contributors to Linux at many levels, and their Linux support for their hardware is second to none. Intel pays a lot of developers to write code that they give away to the community, and I certainly keep that in mind when it is time to decide on hardware purchases.

AMD got on board and open-sourced their current graphics drivers years ago, and now they use the same userspace framework that Intel uses for its integrated GPUs (Mesa). The open-source Nouveau nVidia drivers also use Mesa, but nVidia’s refusal to offer any useful information has resulted in those Nouveau drivers being very slow and lacking support for many features.

For users of nVidia hardware, like me, the proprietary nVidia driver is the only real choice, and their lack of support for a lot of the Linux standards (like VAAPI for video acceleration, which everyone supports except them) has been a big pain in the butt. They have not put as much effort into developing their Linux drivers as they have the Windows drivers, and with those drivers being closed, those things can’t be added by anyone outside of nVidia.

It should not be a surprise, with that in mind, that Steam chose to use AMD hardware for their Steam Deck portable console, which is a Linux PC internally that uses Proton to play Windows games. I don’t know if nVidia was interested in being the provider for that project or not, but it was just one more thing throwing market share to AMD. Linux has not been a major force in PC gaming historically, but the Steam Deck shows how rapidly this is changing.

Many Windows-oriented tech prognosticators have opined that Linux is the future of PC gaming, and AMD stands poised to benefit from this. Now that the performance per watt advantage of nVidia is no longer there, and that AMD’s best GPUs are nipping at the heels of nVidia’s very best (for far less cost), there is nothing holding the market back from a shift to AMD as the standard if their support for Linux begins to matter in a big way.

I hope nVidia is coming around. They announced recently that they are open-sourcing the kernel modules for their driver for newer nVidia architectures (Turing and Ampere, presently) starting with driver 515, currently in beta. This will benefit the Nouveau project, as reading how nVidia does things within their kernel modules will allow Nouveau devs to learn things about how nVidia cards work that they would not have otherwise known. Hopefully it will be a sign of a new nVidia that, like Microsoft, will embrace open-source. That could prove beneficial to Windows users too, as improvements to the nVidia drivers from the community may be something that they would want to incorporate into the Windows drivers. They may even wish to consolidate some of the code base between the Linux and Windows drivers, saving themselves some money.

If not, I will certainly be looking at AMD’s offerings as time goes forward. I just bought a new gaming laptop that has yet to be delivered (scheduled for later today!), and it’s another nVidia, even though I would have preferred AMD for the reasons above. Most gaming laptops use nVidia, which means if you happen to find a really great deal, it most probably will be on an nVidia-powered unit, just on the numbers. For the price I paid, there was nothing close on the AMD side that would even come close in performance terms. I’m hopeful about nVidia, and I hope I won’t be kicking myself for not paying more and getting the AMD option.

Dell XPS 13/9310, i5-1135G7/16GB, KDE Neon 6.2
XPG Xenia 15, i7-9750H/32GB & GTX1660ti, Kubuntu 24.04
Acer Swift Go 14, i5-1335U/16GB, Kubuntu 24.04 (and Win 11)

1 user thanked author for this post.

Simon_Weel

Reply | Quote

Brian Livingstone: Yes, GPUs are amazing: superfast!
But you pretty much left out, addressing it only indirectly, one use of GPUs of not negligible importance:

First it was “Titan”

https://en.wikipedia.org/wiki/Titan_(supercomputer)

Then:

“Solve the World’s Greatest Challenges with Supercomputing”

https://www.nvidia.com/en-us/industries/supercomputing/

And so it goes: Teraflops? Nah! Gigaflops! What??? Petaflops. No way: Exaflops! …

“Nvidia reveals H100 GPU for AI and teases ‘world’s fastest AI supercomputer’ ”

https://www.theverge.com/2022/3/22/22989182/nvidia-ai-hopper-architecture-h100-gpu-eos-supercomputer

Excerpt :
(my emphasis)

“The supercomputer, named Eos, will be built using the Hopper architecture and contain some 4,600 H100 GPUs to offer 18.4 exaflops(*) of “AI performance.” The system will be used for Nvidia’s internal research only, and the company said it would be online in a few months’ time. ”

(*) 1 exaflop = 1 followed by 18 zeroes of flops: one million of millions of millions  floating point operations per second.

Ex-Windows user (Win. 98, XP, 7); since mid-2017 using also macOS. Presently on Monterey 12.15 & sometimes running also Linux (Mint).

MacBook Pro circa mid-2015, 15" display, with 16GB 1600 GHz DDR3 RAM, 1 TB SSD, a Haswell architecture Intel CPU with 4 Cores and 8 Threads model i7-4870HQ @ 2.50GHz.
Intel Iris Pro GPU with Built-in Bus, VRAM 1.5 GB, Display 2880 x 1800 Retina, 24-Bit color.
macOS Monterey; browsers: Waterfox "Current", Vivaldi and (now and then) Chrome; security apps. Intego AV

Reply | Quote

AMD’s Frontier supercomputer debuts as world’s fastest, breaking exascale barrier

May 30, 2022
The Frontier supercomputer at the Department of Energy’s Oak Ridge National Laboratory earned the top ranking today as the world’s fastest on the 59th TOP500 list, with 1.1 exaflops of performance. The system is the first to achieve an unprecedented level of computing performance known as exascale, a threshold of a quintillion calculations per second.

Frontier features a theoretical peak performance of 2 exaflops, or two quintillion calculations per second, making it ten times more powerful than ORNL’s Summit system. The system leverages ORNL’s extensive expertise in accelerated computing and will enable scientists to develop critically needed technologies for the country’s energy, economic and national security, helping researchers address problems of national importance that were impossible to solve just five years ago…

World’s First Exascale Supercomputer Powered by AMD EPYC™ Processors and AMD Instinct™ Accelerators

Reply | Quote

So it is happening already: the exaflop computing era has dawned.

As the Oak Ridge National Lab statement quoted above by Alex makes clear, this is not just about playing games and implementing ever more sophisticated and faster kinds of AI.

As to what the 1st exaflop supercomputer “Frontier” is made of:

Because of all the “cool” tech jargon + PR in the AMD announcement, I’ve looked elsewhere for an explanation of those crucial components of this supercomputer, and found that AMD EPYC processors are CPU’s with lots of cores, that the AMD Instict are GPUs, and that I have not a single clue of what “Accelerators” means here:

https://en.wikipedia.org/wiki/Epyc

https://www.tomshardware.com/news/amd-radeon-instinct-miopen-deep-learning,33170.html

Ex-Windows user (Win. 98, XP, 7); since mid-2017 using also macOS. Presently on Monterey 12.15 & sometimes running also Linux (Mint).

MacBook Pro circa mid-2015, 15" display, with 16GB 1600 GHz DDR3 RAM, 1 TB SSD, a Haswell architecture Intel CPU with 4 Cores and 8 Threads model i7-4870HQ @ 2.50GHz.
Intel Iris Pro GPU with Built-in Bus, VRAM 1.5 GB, Display 2880 x 1800 Retina, 24-Bit color.
macOS Monterey; browsers: Waterfox "Current", Vivaldi and (now and then) Chrome; security apps. Intego AV

Reply | Quote