Semianalysis has a very bullish article on MI300 ramp (in the upcoming quarters):
AMD MI300 Ramp, GPT-4 Performance, ASP & Volumes
Order Volumes From Microsoft, Meta, Oracle, Google, Supermicro/Quanta direct, Amazon AMD’s upcoming MI300 is poised as the only legitimate competitor to Nvidia and Google hardware in LLM inference.…www.semianalysis.com
When it comes to consumer PCs AMD and Intel currently have duopoly cartel going on. That's about to break when others can flood the market with their own ARM Windows machines. There's going to be competition and things can change rather quickly. It would be a really good thing if current situation would change.Yep, dunno why people assume ARM is going to start eating market share all of a sudden.
The reason for previous market moves like mainframes to minis, minis to RISC workstation/servers, RISC workstation/servers to x86 PC/servers is because the replacement was CHEAPER, not because it performed better.
I don't see any evidence that Qualcomm's ARM PCs are going to be sold for less than the equivalent Intel/AMD PC. Knowing Qualcomm they'll price them for hundreds more, and try to claim having built in cellular is an advantage. They'll probably be priced like Apple, with no low end and little midrange. Even if Apple does add a cheaper Mac it will be functionally like the iPhone SE (though it won't be priced nearly that low) and provide a limited midrange but still nothing on the low end.
Custom ARM CPUs from Amazon etc. will likely remain for their internal use and only be sold as a cloud product, so other than maybe limiting the worse excesses of Intel/AMD hyperscale CPU pricing (Intel/AMD have a LOT of room to come down on those prices in response to competition) they won't make much difference.
The problem is MI300 somehow starts (!) competing with a 2022 rival's product (H100) during 2024 when the competition plans a next-gen (3nm B100) in roughly the same time-frame (2024).With the foot in the door, Mi300 will be competing on merit, which will hopefully result in a shift of the market share to AMD, since Mi300 is likely going to outperform H100 in some workloads. And probably also in power efficiency.
WoA with QC exists for quite some time now. It's still to be seen whether now suddenly this (the hype around QC's upcoming SoC, possibly QC's end of supposed exclusivity on WoA) is making an actual difference this time. Sure, things can change, like OS/2, BeOS, Linux desktop or SteamOS replacing Windows in the PC market can technically still happen.When it comes to consumer PCs AMD and Intel currently have duopoly cartel going on. That's about to break when others can flood the market with their own ARM Windows machines. There's going to be competition and things can change rather quickly. It would be a really good thing if current situation would change.
The problem is MI300 somehow starts (!) competing with a 2022 rival's product (H100) during 2024 when the competition plans a next-gen (3nm B100) in roughly the same time-frame (2024).
When it comes to consumer PCs AMD and Intel currently have duopoly cartel going on. That's about to break when others can flood the market with their own ARM Windows machines. There's going to be competition and things can change rather quickly. It would be a really good thing if current situation would change.
Bingo!What incentive does Qualcomm have to majorly undercut Intel and AMD's CPU pricing? Have you SEEN Qualcomm's operating margin? They are not a cutthroat company willing to sacrifice profit to gain market share! If anything they are more greedy than Intel.
Intel earnings report does not break out Gaudi revenue. So, I have no idea what you are referring to.
But in a separate interview, Intel CFO said that Intel hopes Gaudi will achieve its first $100 million revenue sometimes in 2024.
Intel is about as far behind AMD as AMD is behind NVidia in datacenter GPU.
I think you have a misconception about Intel's participation in datacenter AI accelerator market.
NVidia is leading, AMD is far behind NVidia. Intel is part of a pack that is far behind AMD.
That's Gaudi 1.FYI Gaudi 2 is already deployed to the public cloud on AWS:
They are, and you'll see it on December 6th or whenever that AMD event is on.but it is amusing to claim they are far behind AMD
Yes MI200 is only ever used in a tiny little irrelevant system called Frontier.At least the Intel solutions actually exist and can be used for real things.
YES THEY DO.AMD does not have the lead over Intel in this area that you claim they do.
I think you are a bit misinformed.
In the Q2 23 earnings call, Intel mentioned they had a $1 billion potential pipeline for Gaudi 2. In the recent Q3 call, they mentioned the pipeline has doubled to approximately $2 billion. You can see the earnings transcripts for yourself to see this. Now "pipeline" doesn't mean confirmed sales, but your $100M figure is not factual.
AMD does not have the lead over Intel in this area that you claim they do.
The problem AMD is facing is bad advice. They think maintaining margins is better than reducing margins to increase sales and profits.
I'm not sure it is such bad advice, unfortunately for us. Think about how they aquired Xilinx, they used stock. Their high margins propped up the stock price, so they effectively used margins as a wealth multiplier. Such is the state of modern financial markets.The problem AMD is facing is bad advice. They think maintaining margins is better than reducing margins to increase sales and profits.
When times are good, you can charge more. When times are lean, you have to meet the market where it is. I am simply saying what I see. I can remember review sites used to say buy 1st or 2nd generation Zen processors simply for the value and the feature packed motherboards that were less $50-100 less than Intel boards.Without those margins there s no competitive RD, rememeber that they have to do RD on several fronts, CPUs, GPUs and Xilinx FPGAs plus a few related techs.
I don't get it. As am example, I just got a $50 650 motherboard and a $509 7950x and for the last 2 days it been running 100% avx-512 load with no problems. Oh, and $100 60000 cl30 32 gig memory. And your point is ????When times are good, you can charge more. When times are lean, you have to meet the market where it is. I am simply saying what I see. I can remember review sites used to say buy 1st or 2nd generation Zen processors simply for the value and the feature packed motherboards that were less $50-100 less than Intel boards.
If AMD didn't the TDP of the Zen 4 CPU's. They could say our CPU's may not perform quite as well as the Intel stuff, but they use 65-105w. I still defend AMD GPU's. A lot of people think AMD GPU's are a 2nd class product. But 16GB as a minimum and 20GB plus DDR6 Vram and aggressive pricing would go along way for AMD.
The Motherboard prices for Zen 4 are pretty insane. AMD used to have crazy bundle deals on newegg for really good AM4 motherboards and processors.
I will say in closing a RX 7600 with 16GB of Vram for $199 would be a huge shot across the bow at Nvidia. A 6800xt refresh on 5nm would be impressive.
I will still buy AMD hardware. But the affection I had for AMD has diminished over the last few years.
Zen 3 was the best across the board when it was released. That is when you can charge a premium. To gain market share you price according to market conditions, performance and the competition.I'm not sure it is such bad advice, unfortunately for us. Think about how they aquired Xilinx, they used stock. Their high margins propped up the stock price, so they effectively used margins as a wealth multiplier. Such is the state of modern financial markets.
I figured you would wait a few weeks for threadripper so you can get a $2K processor and a $500 motherboard to go with $100 ram.I don't get it. As am example, I just got a $50 650 motherboard and a $509 7950x and for the last 2 days it been running 100% avx-512 load with no problems. Oh, and $100 60000 cl30 32 gig memory. And your point is ????
The problem is MI300 somehow starts (!) competing with a 2022 rival's product (H100) during 2024 when the competition plans a next-gen (3nm B100) in roughly the same time-frame (2024).
In the face of a competitor that is selling product at near-zero profit margin, what else do you expect them to do?The problem AMD is facing is bad advice. They think maintaining margins is better than reducing margins to increase sales and profits.
Intel uses a bunch of PVC in Aurora too. Hardly a compelling argument.Yes MI200 is only ever used in a tiny little irrelevant system called Frontier.
YES THEY DO.
ML is basically an NV and AMD and to a lot lesser degree Cerebras club.
Everyone else is either bad or late to the party.
Just so we are clear, we are comparing projected revenue for AI products for both companies. It is fair if you want to call Intel's $2 billion projection for next year "BS" while trusting the AMD projected numbers but lets just be clear that there is little objectivity in that view. Fortunately, we won't have to wait long to see what actually happens.The only thing that's real is revenue. And Intel / Gaudi does not have it.
The way Intel execs talk is borderline BSing, and no surprise that people (like you and @lightisgood ) are falling for it. Such as Gaudi being 2x to 3x Mi300.
What is known is that Mi300 will have $400m revenue in Q4 2023 and Gaudi will not reach first $100m in revenue until 2024.
AMD Mi300 lead will be in revenue and in performance.
Intel lead is only in BSing.
Aurora is a long delayed meme system, Frontier's been up and running for over a year.Intel uses a bunch of PVC in Aurora too. Hardly a compelling argument.
Yes.
No you just install ROCm, install pytorch and there you go.If you do work in ML/DL and you care about your time and productivity, you use Nvidia. All the standard libraries, frameworks and tools are still very much Nvidia centric and if you bring AMD hardware you still waste lots of time trying to figure out terrible documentation and trying to deal with the joke that is ROCm. I wouldn't bother with using Intel here either, but you're splitting hairs if you think AMD has some big lead here over Intel.
When a company GIVES away products in order to keep the competition from making sales, is that not anti-competitive behavior ? AMD won this suit before when Intel did something similar.Intel Data Center group had an operating margin of 2% and made an operating profit of $0.1B in Q3. The guidance for next quarter is better. I don't think they will be going bankrupt from that...
This is a pretty good outcome for them given how bad Sapphire Rapids is in comparison to AMD offerings.
As far as anti-competitive behavior, can you elaborate more on this?