Well I think the WHY is pretty simple. Microsoft's leadership is jealous of all the free (and positive) press Apple got over their move to ARM, and wished someone would talk about Microsoft or Windows in such glowing ways.
Qualcomm isn't going to like that.1. Correct and they called NV to the rescue. Next year it will be QC/AMD for entry/mid range WoA stuff and NV for premium real Apple competitor
130 TOPS/Watt sounds crazy. Is that for INT4?2. Again, NV to the rescue. First with their new VS-Quant INT4 inference DL NPU accelerator that offers ~130TOPS/Watt, way better than anything in the market and second with industry standard Nvidia GPU Tensor/gaming/AI ecosystem.
MS and NV in bed again to fight Apple. That's an interesting battle. Bring in on!
Yep. I put the details in the "Nvidia enters the game. The rise of ARM" topic130 TOPS/Watt sounds crazy. Is that for INT4?
He was talking about AMD's SoC.Sure, everyone should trust you because you know better than Michael Dell and Jensen... or not
Sure, Sound Wave is not.chefe, that semicustom APU isn't 2025
Midrange?and it's not midrange
Yea.Sure, Sound Wave is not.
It's VGH-like!I believe Sound Wave is a LNL-like part?
no that's the same tier as Strix1 (>$1200 SRP devices), just different segment.That is midrange compared to something like Strix Halo/Strix Point.
LEL.1. Correct and they called NV to the rescue. Next year it will be QC/AMD for entry/mid range WoA stuff and NV for premium real Apple competitor
LOOOOOOOOOOOOOOOOOOOOOL2. Again, NV to the rescue. First with their new VS-Quant INT4 inference DL NPU accelerator that offers ~130TOPS/Watt, way better than anything in the market and second with industry standard Nvidia GPU Tensor/gaming/AI ecosystem.
MS and NV in bed again to fight Apple. That's an interesting battle. Bring in on!
The past is one thing but what about now? NV2024 is a totally different beast than NV2015. In fact Nvidia with ARM is doing extremely well currently. Automotive business is on the rise with 5 billion turnover in the pipeline (Thor is a huge success). BlueField-4 DPU has a fast ramp up due to high demand. Grace is a fantastic memory controller for Hopper/Blackwell and Grace2 will be nearly mandatory to push Rubin to its max performance ("memory controller" is a quote from Bill Dally), Switch2 SoC will sell by millions per quarter, next NV ARM SoC will be the shining star of Microsoft 2025 CoPilot+PC line up and the new semi-custom division gains lot of attraction. These are facts, not feelings but you are free to disagree and time will tell...You'd think he would remember the lessons from Denver. The rare occasion that NV's roadmap falls apart after 2 generations and they go back to stock ARM.
Sure enough he was shilling it to high heaven back then.
Yes, the power is truly ughh, outstanding on Cortex-X5.You underestimate the power of
Blackhawk performance is on part with Oryon 2 but that's not the selling point of the NV SoC. GPU Optix/Tensor ecosystem for creators (Adobe suite on ARM optimized for NV + Blender, Maya, Dassault, Siemens, etc) and new NPU for CoPilot+ AI assistant are what MS will push to fight Apple
Òooh.Blackhawk performance is on part with Oryon 2
That's giving new meaning to this article:GPU Optix/Tensor ecosystem for creators (Adobe suite on ARM optimized for NV + Blender, Maya, Dassault, Siemens, etc) and new NPU for CoPilot+ AI assistant are what MS will push to fight Apple.
So is AMD, which was nearly dead in 2015 whereas NV hasn't been that close to catastrophe since at least bumpgate.The past is one thing but what about now? NV2024 is a totally different beast than NV2015.
Canning Atlan sure was an interesting play, we all know that NV has only one relevant BU currently that it is 90% of revenue, though they will certainly need to use that money to diversify their rev streams.In fact Nvidia with ARM is doing extremely well currently. Automotive business is on the rise with 5 billion turnover in the pipeline (Thor is a huge success).
And yet they cannot cram it all on a single package like MI400 will.BlueField-4 DPU has a fast ramp up due to high demand. Grace is a fantastic memory controller for Hopper/Blackwell and Grace2 will be nearly mandatory to push Rubin to its max performance ("memory controller" is a quote from Bill Dally)
We can both agree than Nintendo treats hardware like a necessary evil rather than something to actually push a product, who needs a modern, cheaper, more performant and more efficient SoC anyway? (which both NV and AMD could've provided).Switch2 SoC will sell by millions per quarter
The less said about Microsoft the better.next NV ARM SoC will be the shining star of Microsoft 2025 CoPilot+PC line up and the new semi-custom division gains lot of attraction.
You sure like to ignore other facts but hey, nice to live in a bubble.These are facts, not feelings but you are free to disagree and time will tell...
Should be basically equal by RDNA5 as software is by far the biggest gap here and it takes time.Blackhawk performance is on part with Oryon 2 but that's not the selling point of the NV SoC. GPU Optix/Tensor ecosystem for creators (Adobe suite on ARM optimized for NV + Blender, Maya, Dassault, Siemens, etc) and new NPU for CoPilot+ AI assistant are what MS will push to fight Apple
1. Size comparison then and now:1. So is AMD, which was nearly dead in 2015 whereas NV hasn't been that close to catastrophe since at least bumpgate.
2. Canning Atlan sure was an interesting play, we all know that NV has only one relevant BU currently that it is 90% of revenue, though they will certainly need to use that money to diversify their rev streams.
3. And yet they cannot cram it all on a single package like MI400 will.
4. We can both agree than Nintendo treats hardware like a necessary evil rather than something to actually push a product, who needs a modern, cheaper, more performant and more efficient SoC anyway? (which both NV and AMD could've provided).
5. The less said about Microsoft the better.
6. You sure like to ignore other facts but hey, nice to live in a bubble.
7. Should be basically equal by RDNA5 as software is by far the biggest gap here and it takes time.
The TAM of that market really isn't that big on client, which is why AMD has ignored it until fairly recently.
No?The past is one thing but what about now? NV2024 is a totally different beast than NV2015.
Nvidia's magical wonderland of AI World is doing extremely well.In fact Nvidia with ARM is doing extremely well currently.
Did you write that with the little Jensen Leaflet sitting just above the keyboard to copy?Automotive business is on the rise with 5 billion turnover in the pipeline (Thor is a huge success). BlueField-4 DPU has a fast ramp up due to high demand.
AHAHAHAHAHAHAHAHAHAHAGrace is a fantastic memory controller for Hopper/Blackwell and Grace2 will be nearly mandatory to push Rubin to its max performance ("memory controller" is a quote from Bill Dally)
Ok we are going from Grace shilling to defending a low power Ampere, fabbed on SS 8nm, that's gonna start selling in 2025, as "high volume, thus good".Switch2 SoC will sell by millions per quarter
Of my laughter therapy?next NV ARM SoC will be the shining star
Speaking of that lineup, it is entering Funny territory.of Microsoft 2025 CoPilot+PC line up
Again even AMD fans do not shill the PS5 sales as "attractive".and the new semi-custom division gains lot of attraction.
The only fact I see here is that I used to be a little skeptical of AMD users defending Radeon despite all their blunders, and yet this is a million times worse.These are facts, not feelings but you are free to disagree and time will tell...
NV had better products and a stronger roadmap at the time, AMD was coasting towards Zen with nothing else to show.1. Size comparison then and now:
In 2015, AMD Market Cap was $2.2B with $4B Revenue and $660M net income
in 2015 NVDA market cap was $17.7B with $4.7B Revenue and $663M net income
And now in 2024:
AMD: $269B market cap / ~$20B revenue / less than $3B profit
NVDA: $2.5 trillion market cap / ~$120B revenue / ~$60B profit
Objectively, in 2015 AMD and Nvidia were in the same size bracket. But today, Nvidia is in a totally different league with infinite cash to spend on their expansion. You can't compare the situation then and now
Agreed, automotive has longer timelines for sure.2. Atlan was a not a real new generation update and it had a bad timing so they accelerated Thor, which was the right move
Uhh AMD has made a big song and dance about full stack HW/SW server ecosystem plays, exactly the same concept as NV.3. MI400 adheres to a different approach but they loose badly to Rubin anyway because NV doesn't sell GPUs but systems. In fact, Jensen says NV72L is Blackwell not GB200. The point is these huge LLM don't fit in a single GPU, not even in a single rack. That's why AMD is not competitive to these hyperscalers and cloud customers. Barely a bargain tool to keep NV prices in check
I'm sure the part AMD offered was better than T239 in every way, Nintendo just wanted zero software hassles considering it is seemingly just a faster Switch.4. Totally agree on Nintendo cheaping out their hadware. But you forget to mention that AMD bid for Switch2 and they lost. So AMD was even worst than a slightly custom Orin...
Not the most pleasant statement I guess.6. My bubble is confirmed by the market which is the ultimate truth.
AMD knows they have work to do, it is where they are the furthest behind.7. I don't know RDNA5 except the rumors, but ecosystem is everything and MS knows it. I don't exaggerate if I say that 90% of creators on PC use Nvidia GPUs. And creators are a big part of Apple customers. MS needs to appeal this demographic and they won't achieve it with AMD
I removed the emotional rage from my quote and focus on the interesting points:1. Pretty much nothing has changed in the technicals or leadership lol. Their markets sure have changed, and ARM brings no benefit there...
2. We have now reached the "Grace is fantastic" tier.
Grace, that "Ebin XDD ARM CPU" that is so Ebin-fantastic-amazing that it needs to be taped to Hopper to even have a chance at selling.
A piece of hardware that's the product of years and years of R&D and that has so much "success" that they need to force sell it with their GPUs, have no intention of expanding the business on it(since there's no business), and have to do 3x the marketing to force people to stop asking if they can get Hopper without it because "Grace is so great, we told you so many times".
3. Again even AMD fans do not shill the PS5 sales as "attractive".
And AMD has what, PS5, XBOX, Steam Deck, Rog Ally, Legion Go, and a few others? With only the MSI Claw competing(incredibly poorly cause Intel)? Oh and absolutely nobody but Nintendo bought Orin, of course. Proof of Nvidia's Ebin XDD success in semi-custom.
4. Nothing has changed, the corpo's the same, same qualities, same flaws.
They still go for volume and fat expensive dies instead of being economical or chipletizing.
Same old Nvidia, same old AMD doing the opposite across the pond. Same results too: everyone works with AMD, everyone avoids Nvidia unless they're the only product on the market.
I disagree. ARM brings mostly itself as a meme.I removed the emotional rage from my quote and focus on the interesting points:
1. Like it or not, ARM bring competition and that's already a lot.
Duopolies are pretty much universal in extremely high CAPEX, high R&D businesses.Unless you are happy with 2 players monopolizing the CPU PC space status quo.
No. Effective competition does. And again in ultra high capex/R&D needs, effective competition is next to impossible to enter because you need mad investments for years and a good amount of luck to get a viable result.Competition brings innovation and better value for money.
And ARC has been an excellent example of what I just said. They poured the mad investments, and were one of the very few companies with the clout and skills to get it.The same reason why I welcome intel on the discreet GPU even if they are in a painful journey.
And again, 90% of Nvidia's business is GPU compute and shilling the side thing that allows GPU compute to work is ridiculous.Grace is really a beefy "memory controller" for the GPUs, it was designed this way, and it does it very well. Again, Nvidia is not selling individual silicon but systems. You are looking at the product the wrong way.
"Losing" is a pretty ridiculous term to use here lol.3. And yet AMD lost against a "Ebin XDD" SoC for the Switch2 bizness.
I qualify it as good.So how do you qualify the SoC that AMD offered to Nintendo? Garbage?
Yes that's why Nvidia wins. Not going to argue that.4. The thing we agree is that Nvidia has a different philosophy than AMD. In most of markets they compete, AMD is trying to make money with the difficult exercise of bringing the maximum performance at the lowest cost, yet increasing their margins. On the other side, Nvidia is the performance/industry leader and they know they can charge a premium for their robust solution (ie ecosystem).
Nobody sane is saying that Nvidia can't do chiplets. I'm saying that they're NOT doing it. I.E they still think that it's a thing for another time.But if you think Nvidia don't embrace chiplets, you are very wrong...
Duopolies are pretty much universal in extremely high CAPEX, high R&D businesses.
ATI/NV, Intel/AMD, Samsung/Apple, etc.
Yes that's why Nvidia wins. Not going to argue that.
My problem with your statements is not about NV's general strategy, it's with defending all the ridiculous side gigs and extra "we tried expanding there and there and there", all the arm flailing that Jensen is doing as "proof of Nvidia's success".
You remove all of it, every single last piece of HW, and replace it with a generic QC/AMD solution, it's the same or better.
You keep all of it, but sell it independently from GPUs, and it's getting kicked in the shins to death.
Nvidia has no real portfolio diversity, they have a GPU powerhouse that allows them to shill away an insane diversification of poor, uninteresting products. Calling it "Nvidia sells systems" is beyond charitable, Nvidia sells packages with one thing that's wanted and a bunch of lesser quality stuff that is there because Nvidia would rather sell lower quality stilts to prop up their GPU products than ship their GPUs with better quality products that would give money to competitors.
Jensen doesn't trap anybody. Nvidia ecosystem and their multiple frameworks, APIs, libraries, documentations are praised by the customers. It's the difference between a solution that I can deploy in few weeks vs using the cheaper HW alternative that I must develop everything for, and put into production few months after my competitors...This isn't a detail, it's the centerpiece of the Nvidia insanity: they have a strong full stack compute offering. That's it. The rest is artificially taped to that, glued together, or one way or another force-sold. This is radically different from AMD where you could divest the company back into ATI, AMD and Xilinx, and GPUs, CPUs and FPGAs would each make viable companies. Nvidia is a one trick pony, and has absolutely never grown out of their Compute Hole. Jensen is a Spider in a Leather Jacket, trying to trap you into his ecosystem.
And this conversation is over because I'm gonna get banned again if I respond.Nobody forced these customers to choose Grace CPUs, still that's the reality.
I forgot to reply this part.Uhh AMD has made a big song and dance about full stack HW/SW server ecosystem plays, exactly the same concept as NV.
They are the only other company with a viable roadmap here, but will require some external help. AMD is betting big on making the most comprehensive system-on-package as the basic unit and scaling it up in much the same way.