Nvidia vs. Intel: Gaudi 3's Reality Check

[@YelloWabbit]

Im a sucker for underdogs and I think Intel will bounce back. Don’t forget Intel will have its own chip manufacturing plant in the near future, in the U.S. The closest company that has a chance to compete with NVIDIA in this space is Intel, and we should be thankful really. Otherwise you’ll have a monopoly on your hands a lot sooner than you think considering how fast technology is progressing with AI.

[@diederik6975]

I think AMD is already competing with MI300X - and already have a market share

[@SY-zv2cz]

Even if Intels Gaudi 3 were to be faster than H100 in the largest AI models, people are still going to buy H100 for the higher end market. The question is not whether Intel or AMD is capable of designing a GPU that beats Nvidia in raw performance, but the actual question is, will Intel and AMD's programming models be accepted by the AI industry as a replacement for Nvidia's proprietary CUDA ecosystem. AI companies dont want to shift their programming model at this stage, coz they are competing with each other on TTM and for them paying extra for the H100 makes sure they dont have to deal with a lot of software porting at this early stage. I think Intel is playing a slightly better game than AMD here by targeting the mid-lower end of the market with a decent GPU rather than competing Nvidia in absolute performance. AMD is gunning for the top spot in performance which they can achieve to some extend. But still those large AI customers will only choose AMD's GPUs for the mid-lower spectrum where they dont mind shifting to RoCm or OneAPI based models. That said, once AI implementations start progressing towards maturity, AI companies may start taking the risk of porting part of their implementations to non-CUDA ecosystem. That might take at least a couple of years.

[@rondobrondo]

or we get open CUDA

[@dapdapstuff]

If you watch the whole presentation, Pat explained that industry is moving to open framework such PyTorch and a few lines of code can switch between NVIDIA to other hardware (Intel)

[@КОМБИНАТ-т1ъ]

When the market is boomin' and promises massive profits, there will be demand for anything that calculates. Just like it was when cryptos were on rise. Too bad it turned to be just a financial bubble

[@iokwong1871]

@@rondobrondo Yeah good luck with....that LOL

[@christopherdennis6785]

Honda has engineered a formidable engine that has been the cornerstone of Red Bull Racing’s dominance in Formula One. This partnership has seen Red Bull outperform rivals, including the storied Ferrari team.

[@SA-ht5fw]

Yeah but how many streets do you know where you are allowed to drive at 130mph?

[@jamescole3152]

It seems you may be missing the most important facts. Intel will be making chips for everyone. Just like TSMC. So Nvidia may be using Intel to make their chips. Remember Nvidia does not make chips. Nvidia does not make chips. So look to the agreements between companies. If you want the real scoop.

[@one_rich]

you're also missing an important fact, TSMC is so far ahead of intel in chip fabrication that Intel is outsourcing its next generation chips to them just so they can remain competitive with AMD.

[@iokwong1871]

@@one_rich Yeah... about that. sure AMD is doing great, but Isn't like they are years ahead, Intel is able to go head to head with AMD with their 7-10nm chip, where AMD is using 3-5nm.... But yeah, Intel did have some dumb CEO for some years. that I agree.

[@juniorjunior8494]

I agree with most of what you said, but you basically did the same thing you're criticizing, Intel posted benchmarks on specific scenarios. NVIDIA's blackwell claims are also very specific and cannot be taken at face value, but you just repeated their claims and stopped there. Please read the fine print and you will see how they come up with such multipliers of performance, that's assuming you understand hardware architectures and how they work

[@iokwong1871]

I am no Intel fan boy, but for "reallity check" Intel is still about 63% of the Server CPU Market..... and AMD only have 36%. And yes, AMD is catching up.... but it is still only half of Intel......... You mean it sounds like Intel will do nothing and just sit there. LOL. Sure the future maybe a GPU market for server. But just like anything tech. The table can all round 180. Just like Apple suddenly have the most powerful Home Computer CPU, well and GPU soon enough... As good as Nvidia's RTX line up, if not better.... And let's not forget that Intel is the only Major player who have thier own Manufacturing plant. I mean really, I personally think we are at the best era ever, where all three major player is doing great. and none of them are perfect. But it is so much better then the day where there's only Intel, or Nvidia.

[@thepunish3r735]

Couldn’t have said it better myself… and for Intel to be spending more money than any other chip maker to be more competitive says alot .. I love people like him saying Intel is trash, I’m buying up Intel stock while it’s cheap

[@probablybored13]

I think that cost/performance here is an important point. For large deep learning models GPU is needed for inference as well as training, and many businesses will be using pre-trained models that may or may not require significant fine-tuning. Likewise, its not always "bigger is better" for models. 13B parameter models can be generalisable enough to be really useful for RAG applications, text classification, entity resolution, summarisation etc. If you don't need something as large as 70B parameters to satisfy your business problem, why would you serve a larger model? If businesses can run SOTA models off the shelf for a lower cost it could become very popular. The challenge will be more around the software side ensuring that there is compatibility with popular frameworks like PyTorch which look like they've been addressed with a GPU migration toolkit for CUDA based models. Early days but I'm cautiously optimistic

[@criticalloves]

Makes sense although Intel felt back somehow thinking us would be on their back and globalization shows other companies would make the same product better efficiently and cheaper

[@MARKXHWANG]

They have these claims since Audi 2, but no one uses it. Lol

[@gk9670]

You finish by saying “ dive into the detail”. Did you think about why Intel is losing market share in data center? It’s because the hyperscalers don’t want to pay Intel. They are playing AMD vs Intel. Also ARM is not a competition. Most of the applications are written for x86. So ARM is a niche. Likewise, not many workloads prefer AMD. Intel is gold standard. These are the details you must know. But these aside, I like a different point of view.

[@EquityEmpireResearch]

Heard the same thing when M1 was released by Apple. People said the software applications would run better on x86.

[@gk9670]

@@EquityEmpireResearch apple is quite different as everything is consumer application. Enterprises are dinosaurs.

[@КОМБИНАТ-т1ъ]

@@EquityEmpireResearch heavy compute that benefits from complex instructions runs better on advanced x86. Some tasks cannot be threaded so you need those advanced instructions

[@juniorjunior8494]

@@EquityEmpireResearch Apple controls every detail of their platform from silicon to every line of code and development kits, that level of control alone can allow insane optimizations. I've always said this, the ISA architecture has nothing to do with performance or efficiency, its the silicon architecture and what the goals of the designer is that drives both. x86 can be as efficient as ARM if that is the goal. Likewise ARM can be as performant as x86 if that is the goal. You can't compare any M chip to x86 because they are fundamentally so different because one chip is targeting a narrow product portfolio with very specific design goals, while the other needs to be flexible and run on just about any device on the planet

[@FoundingFathersUSA]

I’m confused. Intel hasn’t yet released a GPU, only an accelerator, so why are they being compared with H100? Sorry if I’m missing something.

[@one_rich]

H100's are a cluster of GPUs with boat loads of VRAM, Intel has had integrated GPU/CPU since 2010 and released it's ARC line of standalone GPUs last year. While they may call it an AI accelerator, It's basically just a purpose built cluster of GPUs. GPU's are just math accelerators.

[@tringuyen7519]

Intel Gaudi 3 is made on TSMC’s 5nm process. Intel doesn’t even use its own foundry business for its state of the art GPU!

[@EquityEmpireResearch]

True

[@thepunish3r735]

Correct , they are spending billions on their foundries as we speak .. Ohio foundry opens in 2027 ….they are not building a target .. this takes a long ass time .. besides they want to make partnerships with tsmc ..ask me in 2030 if they are biulding their ai accelerators in their foundries 🤷

[@jmtake85]

if u know anything about semiconductor just tell us what method dose tsmc make chip for Nadia, and at what year tsmc must upgrade lithography machine to complete with Intel

[@billmancos9820]

Thanks, he just made me a subscriber, because I wanted to know the truth, especially about has been, Intel. You are really appreciated!

[@Manicsar1]

I get that the media are not experts in the fields they cover, and like with KZbin, the flashier the headline, the more clicks it gets. While it is true that Gaudi 3 will not supplant Nvidia's H100, it is still a good chip for A.I. workloads if you want to save money and need it sooner rather than later. TSMC didn't surpass Intel until 2017 regarding node and market capitalization. It wasn't until last year that TSMC surpassed Intel in revenue. AMD brought in 5.5 billion last quarter, and Intel 12.72 billion. Things at Intel aren't nearly as bad as the media would have you believe. All you are doing is copying every other media outlet and finding ways to justify the current stock price. Intel is the best semi play imo.

[@molacool]

More videos like this, please An overview of amazon, google and Microsoft’s chips. Are they a threat to NVidia ?

[@MARKXHWANG]

Intel had outrageous claim for 10 years and never deliver a single thing

[@Sitharii]

so true what you've said . I'll copy/paste the comment i made few days ago when i first heard those claims : """"About those Intel announcements : They aren't showing comparisons against *nVIDIA Blackwell* ,not even against *nVIDIA H200* ,but against ….. *nVIDIA H100* !! (*H200 can give around + *40% boost* in certain tests Vs H100 like the LargeLanguageModel benchmark for Llama2 70B ,not to even mention Blackwell's leap in performance !! ) How nice , “useful” and most importantly .. “up-to date comparisons” 🥳from Intel !! It’s like releasing their Battlemage consumer GPUs and show us numbers against …. nVIDIA's Turing or Ampere lineup(3-6 years older) !!!"""""

[@Seekee]

i advise you to pray for Intel to get the fight to Nvidia ....Otherwise you will pay with your sweet and time Nvidia sell you The dream video card you want or whatever they sell . Im 50 ...i payd all my childhood and adult life ......have a nice life and read what Nvidia Business practice are .( remember EVGA?) that's the tip of the iceberg .READ stop believing you tube......what's your qualification to tell me anything about technology and chips .....did you benchmark the Intel or Nvidia? im afraid im right you did not ......have nice life !

[@anoniem9518]

so.... because Intel didn't tout the chips performance on the larger parameter model and Meta is bringing out lama 3 as a successor of lama 2, Intel is doomed. It seems "Bleeding Edge" is promoting its short positions on Intel.

[@politeboy92]

Amazing content

[@arnebovarne7759]

Litt for mange store ord og for lite fakta.

[@akaJustMo]

I think you have a good point about current state of AI lead by Nvidia but you missed the big picture. Intel is coming to this market strong and has proven to the world they can adapt and come up with new architecture in relatively short time despite their size. They are shaking up the GPU market against AMD mid-low tier. Their next step is very well known. It does not matter who has the strongest chip if they cannot deliver!! Intel shiped in Q4 alone 50M processos while AMD, Apple, and Nvidia combined shipped 19.8M in the same quantity including GPUs. Not everyone buys a Ferrari or Lamborghini.. the market needs credible enough processors and they want them NOW! Nvidia manufacturing cannot catch up to the demand. So, remove your tech hat, wear an investor hat and bet who will sell more. I don't care about share and marketing value. Nvidia better not be another Tesla - dancing around until the true large manufacturers caught up. If they can't ship enough they are doomed.

[@EquityEmpireResearch]

Intel shipped 50M processors but lost money. It's about gross margin, profits. Nvidia is cranking out 70%+ margins on a $25k chip. So of course they can ship far less of them and generate more profits.

[@akaJustMo]

@@EquityEmpireResearch What are you talking about? First of all gross margin means nothing.. net profit is what matters. Also Intel did NOT report negative margins or net profit. Finally, while Intel shares are cheaper, they have $0.74 dividends per share while Nvidia has $0.16 per share. So yeah, even if the share value drops in the short term you still make more money from Intel shares every quarter from dividends. That's the financial fact of 2023. No one doubts that others are taking market share from Intel. This happens when you're in the top (the only way is to stay or down). Others will eat from your market if you don't play it right. You made it sound like Intel is history while they are still more than 70% of the market, providing better dividends to share holders, and have strong silicone strategies that are transforming new CPUs (read about upcoming architectures Intel announced - not just the AI ones)

[@hanglailai9952]

You need to learn about semi technology before you make any comments. NVDA does not have any new method to make chips, and the software can be overcome easily.

[@EquityEmpireResearch]

Yeah, I’ve heard this for about a decade. Meanwhile Nvidia is worth $2T. Next.

[@dapdapstuff]

@TheBleedingEdgeNewsletter, it appears you may be reluctant to consider alternative perspectives. @hanglailai9952 raises a valid argument. Intel Gaudi3 outperforms H100 in specific scenarios. To draw from your analogy between Ferrari and Honda, for businesses that running around local neighborhood, opting for Honda over the pricier Ferrari would be better.

[@EquityEmpireResearch]

The headlines were Intel bringing the fight to Nvidia. Not Intel filling a small niche.

[@PaulJohnBeatles]

@@EquityEmpireResearch exactly

[@PaulJohnBeatles]

"NVDA does not have any new method to make chips, and the software can be overcome easily." When someone writes a line like this......sad.

[@helenlee6137]

I really listen to you!

[@EquityEmpireResearch]

Thank you!

[@jmtake85]

you see piggy! tsmc can't print out N3 On time.! no blackwell In Q4!! and same time intel take IBM As gaudi 3 customer.

[@EquityEmpireResearch]

spoken like someone who's down big on INTC

[@jmtake85]

@@EquityEmpireResearch sorry piggy I sold Intel at 47 and buy back in at 19.50 show us u p&l I can show u mine 🐷

[@jmtake85]

@@EquityEmpireResearch sorry piggy I sold Intel at 47 and buy back in 19.50 still in profit shows us u p&l I can show mine 🐷

[@Diamond_Hanz]

This guy is a cry baby

[@ziaulislam87]

If intel gaudi cant perform then its dead company

[@Running_injuries]

it's sad that you're a fan of fake chip companies such as AMD/NVVIDIA. you gonna recommend Taiwan semiconductor & Samsung stock too?

[@jmtake85]

funny how people know nothing about semiconductor fab make chip video

[@a0705717]

hey Colin, i wrote you an email, since you said you were open to an editor.

[@EquityEmpireResearch]

Appreciate it!

[@a0705717]

@@EquityEmpireResearch it went to the email adress for the Investor channel, since that's the only email I could find. plus an IG dm on your IG, hope we can connect

[@jimmmaaay1]

thanks colin

[@EquityEmpireResearch]

Thank you!

[@jmtake85]

sorry piggy I sold Intel at 47 and buy back in 19.50 still in profit shows us u p&l I can show mine 🐷 and stop deleted my comments🐷🐷🐷

[@billykotsos4642]

youre hilarious