The AMD Execution Thread [2019]

[BRiT]

The 2019 edition of the AMD execution thread.

 

[Alexko]

Well, just to get the conversation started, Lisa Su is set to give Wednesday's CES opening keynote:
https://www.ces.tech/Conference/Keynote-Addresses.aspx

I'm not sure whether it's supposed to be a simple exposition of AMD's vision of computing, or whether there might be announcements. I hope it's the latter, which should give us some clue of what to expect for this tread.

 

[Deleted member 13524]

https://www.anandtech.com/show/13917/amd-earnings-report-q4-fy-2018

AMD's first profitable year since... 2010? 2011? I can't find the info on that.
Though from Ian Cutress' comments, I think they might have pushed forward the hit from GPU excess stock and 2nd-hand sales to achieve that.

It doesn't mean they'll get a bad 2019, but losses during Q1'19 are expected.


Regardless, showing a Q4 YoY growth when pretty most other tech companies took large hits made the company very tempting for investors.


At the same time, AMD is also steadily increasing their R&D budget each quarter.
In Q4 2018 they spent as much as they did in Q2 2010 ($371M).
Though that's still a far cry of how much they were spending between 2007 and 2009 ($430M - $500M), without even adjusting for inflation.

Oh and Intel still spends about 9x more at $3.3 billion.

 

[Deleted member 2197]

Mubadala divests AMD shares

Mubadala, which is owned by the government of Abu Dhabi, tweeted overnight that it sold 34.9 million common shares of AMD after Monday's close. That represents about 3.5% of the common shares outstanding, which had made Mubadala the fourth-largest shareholder, according to FactSet. Mubadala also tweeted that it was converting 75 million warrants into equity shares.

https://www.marketwatch.com/story/a...-mubadala-tweets-about-share-sales-2019-02-05

 

[Arnold Beckenbauer]

Graphics card vendors under inventory pressure

With Nvidia releasing new cards and cutting prices to clear out its previous-generation inventory, AMD is currently under heavy pressure to lower its quotes, while players selling AMD-based cards are also facing strong competition from Nvidia-based ones.

 

[Deleted member 2197]

Intel snatches more AMD staff for its Xe GPU team, Heather Lennon from AMD RTG Digital Marketing

It pretty much started when Raja Koduri from AMD (leader of the AMD RTG team) made a move towards Intel to develop an actual viable high-end GPU series ready for the business and consumer domain.

Ever since that moment, somewhere in the offices of Intel, a huge budget has been opened up to go big on human resources. As a result, a lot of moves in staff between the companies have been made.

Today once again we see somebody from AMD making a move towards Intel, Heather Lennon who has been responsible for AMDs RTG Digital Marketing.

From the looks of it, Koduri and the team are creating a bit of their own 'RTG team' within Intel obviously under a different name. The fact that Intel is serious about Xe and all derivatives on the IGP side that will come from the architecture, it a sure thing. The initial Xe series will cover a broad spectrum of the market, intended for the discrete market, gaming as well as addressing compute, Deep learning and AI markets.

https://www.guru3d.com/news-story/intel-snatches-more-amd-staff-to-add-to-the-xe-gpu-teamheather-lennon-from-amd-rtg-digital-marketing.html

 

[Bondrewd]

Aaaaaand the second USA Exascale machine goes wholly to AMD!
https://www.amd.com/en/press-releas...t-gpus-and-rocm-open-source-software-to-power

 

[Ike Turner]

Aaaaaand the second USA Exascale machine goes wholly to AMD!
https://www.amd.com/en/press-releas...t-gpus-and-rocm-open-source-software-to-power

This will hopefully put some pressure on CUDA and push researchers & institutions to look into using ROCm.

 

[Bondrewd]

This will hopefully put some pressure on CUDA and push researchers & institutions to look into using ROCm.

Between this and OneAPI, I'd say the CUDA monopoly days are ~numbered.

 

[Ike Turner]

Between this and OneAPI, I'd say the CUDA monopoly days are ~numbered.

Not in the DCC space... where they have a literal monopoly.. But this may change if Intel's GPUs are competitive & somewhat successful..

 

[Alexko]

Not in the DCC space... where they have a literal monopoly.. But this may change if Intel's GPUs are competitive & somewhat successful..

What does DCC stand for?

Good and rather unexpected news for AMD, in any case.

 

[Ike Turner]

What does DCC stand for?

Good and rather unexpected news for AMD, in any case.

Digital Content Creation

 

[Alexko]

Digital Content Creation

Ah, of course, thanks. My understanding is that AMD lags behind quite a bit in driver performance and stability for such applications, hence the huge market share gap.

 

[Ike Turner]

Ah, of course, thanks. My understanding is that AMD lags behind quite a bit in driver performance and stability for such applications, hence the huge market share gap.

That's a myth similar to AMD still having shitty gaming drivers when it is in fact the opposite (Nvidia's consumer drivers releases have been abysmal for the past 4 years in terms of QA & stability.
AMD GPUs are on par with Nvidia's in pro apps and lately even more stable. But Nvidia takes the lead when some features are CUDA only (which is the case in many DCC apps).

 

[Alexko]

That's a myth similar to AMD still having shitty gaming drivers when it is in fact the opposite (Nvidia's consumer drivers releases have been abysmal for the past 4 years in terms of QA & stability.
AMD GPUs are on par with Nvidia's in pro apps and lately even more stable. But Nvidia takes the lead when some features are CUDA only (which is the case in many DCC apps).

Are you sure about this? Last I checked a review, AMD's performance was kind of all over the place, with some good, competitive results, and some really bad ones (compared to NVIDIA).

 

[Ike Turner]

It's a toss between the two. Sometimes NV has the lead while other times AMD does (just like on the gaming side). But the real bad ones are often when a certain feature runs on CUDA on NV GPUs and on the CPU on the AMD system and the person doing the benchmark doesn't even take the time to point it out (or simply doesn't have a clue).

 

[Alexko]

It's a toss between the two. Sometimes NV has the lead while other times AMD does (just like on the gaming side. But the real bad ones are often when a certain feature runs on CUDA on NV GPU and on the CPU on the AMD system and the person doing the benchmark doesn't even take the time to point it out (or simply doesn't have a clue).

Are CUDA features that important to users, or is it just market inertia, then?

 

[Jay]

Are CUDA features that important to users, or is it just market inertia, then?

Is important to users as it affects performance.

Think your question should be about devs.

 

[Ike Turner]

Are CUDA features that important to users, or is it just market inertia, then?

Both. Market inertia, heavy evangilsation + moneyhatting from NV + AMD not having the resources are a big part of it. All those features can be done using open standards (OpenCL/Vulkan on Win/Linux or now Metal on Apple who has ditched OpenCL) but nobody bothers now because this market is 95% running on NV GPUs (and NV worked hard to push CUDA while putting little to no effort in their OpenCL drivers...)

 

[Malo]

An example is something like Daz Studio, which I've been playing around in for about 6 months now. It's completely dominated by IRay renderer for PBR, which is provided by Nvidia. Most content produced for it is now IRay-only. There is no ProRender available for Daz.