Hacker News Reader

Comment by xiphias2 | Hacker News Reader

This comment is off topic, but while I'm listening to the earnings call, I don't hear about specifically official PyTorch and Tensorflow support for AMD graphic cards. All the questions and answers are generic with buzzwords like AI, doubling down on our software support, but it doesn't give me confidence to change my NVIDIA GPU to an AMD one for the foreseeable future.

I remember the time when Elon Musk said to an analyst that he's asking boring questions to fill in his spreadsheet, and I'm feeling the same thing while listening to the earnings call.

View on HN

My impression is that the accelerated computing side of AMD is receiving far, far too little attention. For example, their flagship GPU is still not officially supported by ROCm (AMD's answer to CUDA) [1]. Imagine the 2080 Ti not being supported by CUDA.

I've become a huge AMD fan, both because of their hardware, and because of their commitment to open source. But while the battles they have one against Intel on the x86 side are impressive, it seems that CUDA is leaving them far behind.

[1] https://github.com/RadeonOpenCompute/ROCm/issues/887

It likely will never have support. AMD chose to bifurcate their GPU designs into compute (CDNA) and games (RDNA) lines, with different architectures. RDNA sheds all the fancy features needed to support modern compute, thus gets more efficient in games that do not use it, but also cannot support the modern compute APIs.

NVIDIA is adding more features, like super-scaling to games, and machine learning models are improving faster than Moore's law. I expect those fancy features, like tensor cores to be a must for 4K gaming in the future.

What's funny is that the same strategy (leaving out specialized instructions from consumer level hardware) that worked extremely well for CPUs won't work for GPUs in my opinion.

If you look at ray tracing hardware (I have it on my RTX 2070 Max-Q card in my laptop), it sucks right now, but it's improving very fast as machine learning algorithms improve.

I just found this:

https://www.tomshardware.com/news/amd-big_navi-rdna2-all-we-...

One thing that I forgot is that AMD can just focus on inferencing hardware (INT16 operations), and leave out tensor cores...so actually you are right, I'll just stay with NVIDIA GPUs.

,,AMD ROCm is validated for GPU compute hardware such as AMD Radeon Instinct GPUs. Other AMD cards may work, however, they are not officially supported at this time.''

It seems like Lisa Su thinks that there's a separate ,,gamer market'' and ,,accelerator market''.

Jensen Huang understands that the same person can like to play games and train machine learning models on the same machine.

I'd love to switch to AMD CPU to have a portable laptop with low resource usage, as I spend most of my time travelling, but as GPUs in the cloud are overpriced (thanks to Jensen with separated pricing for servers), and internet in hotels are unpredictable, I don't want to train models in the cloud.

Anyways, Lisa said that she reads all comments about AMD, so I hope she'll listen :)

Yes I posted another link [1] for earning call but didn't get much traction / upvote. Although I have to admit Acquiring Xilinx is much bigger news.

Judging from watching the Financial News and reading analyst's comment for years. My feeling was that their Job was not to push for hard question or an honest answer. Their job is to push whatever interest they had with the company. So a spin for better long term prospect and downplay risk.

I was happy with the Enterprise results ( +116% YoY ) until I read this

>Revenue was higher year-over-year and quarter-over-quarter due to higher semi-custom product sales and increased EPYC processor sales.

Semi-Custom is definitely PS5 and Xbox.

Basically I still dont see EPYC making enough inroad in the Server Market. And this is worrying, while the Stocks, Reviews, Hype are all going to AMD. No Results so far have shown Intel is hurt or AMD is making big gains in market shares and revenue shares.

The only good part I guess is Ryzen Mobile contribution to Computing and Graphics segment.

[1] https://news.ycombinator.com/item?id=24906314

I can imagine that switching from Intel to AMD takes a bit of time for servers, as supporting 2 architectures at the same time is usually bad news for 99%-ile latencies for web services. At the same time x86 is a mature instruction set (as long as you don't use 512 bit vectorization, where things are getting tricky), so the transition shouldn't be that hard.

The server market moves and replaces slowly. Even when Intel was beating AMD by 30% or more it still took years for AMD percentages to drop.

AMD EPYC on AWS is 10.42% cheaper than Intel per hour across the board for m5 instances (9.6% cheaper for t3 instances). 7nm EPYC saves more than 10% power vs Intel 14nm and per-chip savings from buying AMD are way more than 10% vs similar Intel offerings. Why can Amazon spike prices that much? Because people will pay and still consider it a deal.

AMD's main issue still seems to be available due to do much competition for 7nm and tsmc being reluctant to build new fabs.

Do you have any idea why TSMC is reluctant to build new fabs?

Because outside of very specific cases like solar panels, display and LEDs spinning up new fabs seems to be quite risky the investment costs is huge and it seems that TSMC is very much content to make major bucks with fewer customers that are chasing the latest and greatest node.

The more fabs they have the slower their node progression will be and the cost of each new node these days seems to be almost exponential.

This keeps coming up every time there is "perceived" shortage of capacity.

TSMC dont have capacity problem. They are very much willing to built Fabs if their customers are committed to it. But no company other than Apple has ever done that. AMD could have place a large order over the course of the year. Basically a bet that they will sell that many chips. And TSMC will adjust or built accordingly. That problem is no company is willing to place that bet. What if they dont sell and stuck with big pile of chips?

This is simple Supply Chain Management, it is the same principle in every other industry.

Earnings calls never seem to have any hard questions. There was a projections miss a few quarters ago (very rare for amd), and even then the questions feel more like PR.

Semi-Custom is indeed PS5 and Xbox. This increase was expected of course. It's lower margins than other segments though.

Epyc adoption is indeed slower than I would've liked. But so far it has matched or beat short and long term projections from AMD.

Enterprise is weird. AMD has better price and performance? Let's buy Intel. Meltdown lowers performance? Let's compensate by buying more Intel. Intel is supply constrained? Let's complain, and still buy Intel.

My guess is Intel is still pressuring OEMs to favor Intel. Cloud providers are increasing adoption though, and there have been a few nice HPC wins.

And lets not forget that AMD is selling every chip they can, TSMC production is fully booked.

Earnings calls do have hard questions in them, but you might miss them because they're asked with a big dose of circumlocution.

If you look at the incentives a bit, management gets to decide which analysts can ask questions, so analysts need to stay on management's good side.

Analysts know the company's figures inside and out (often have a 10-tab spreadsheet with an extensive operational model of the company), and are asking questions to tweak key model assumptions.

So analysts ask pointed questions in shared jargon with management. You don't ask 'are you seeing a big sales drop because the crypto bubble blew up?', you say 'can you provide some color on when you excess inventory will clear from the channel?'

Analysts get their answers. Management avoids bad headlines written by casual listeners.

There's an additional layer, which is the analysts know the industry and company very well, so general bad things are already background knowledge (there's no reason to ask about them). If you want to know what the analyst already knows, then pay for their report—they're not reporters fishing for a sound bite.

My investment in AMD is so small that it seems silly to pay for analyst reports. I don't care about short term, and I have enough faith in my own research to keep believing in a brighter long term.

Are there any public resources dissecting an earnings call? Doesn't have to be recent or AMD.

You don't get much information from earnings calls, only how analysts are thinking (usually short term). Product reviews on HN, tech oriented in-detail sites, youtube reviews from gamers, and long term analysts, like Ark Invest, and even github issues/comments are much better long term predictors of success/failure.

Just as an example, I remember reading a lot about AirBnB here at HN when it wasn't even known in Eastern Europe. I suggested him to use it to rent out his luxury apartment that he just bought there, and he was the first one to rent out a luxury apartment in the country (also somebody from AirBnB-s management flew there personally)...he made lots of money from rental fees of course, but also AirBnB got incredibly successful. There are lots of other examples of course, this was just the least controversial that I can write here :)

For earnings calls specifically, Seeking Alpha sometimes has decent discussion.

Take everything you read with a big grain of salt though. You get what you pay for, especially in finance—'free' content usually comes with an agenda.

From a personal point of view of somebody who just upgraded to an AMD-based desktop from an Intel one, the CPUs work great. It's the software support that feels half-baked. Enabling memory integrity blue-screens my computer. I need to update the chipset drivers every month because AMD is still dialing in their scheduler and frequency scaling over a year after release, and the idle power usage is inexcusable.

AMD's performance is fantastic, which is why I was fine getting it for a home gaming desktop, but I'm not sure I'd be willing to pull the trigger on AMD on an enterprise server buildout. Intel still simply (actually or otherwise) feels more reliable.

Reply on news.ycombinator.com