> I also recall this 3D shell where your desktop was basically like an first-person shooter, where there would be a literal desk with files that you could click on, a media wall that would display your photos and so on.
Sadly nothing about why Pat was forced out of Intel. Panther Lake is quite a good CPU and 14A looks like it'll be a competitive process. IMO, it vindicates the decisions that Pat made a few years ago and wasn't able to see through to completion.
"Designing microprocessors is like playing Russian roulette. You put a gun to your head, pull the trigger, and find out four years later if you blew your brains out." (Robert Palmer, former DEC CEO)
Pat killed Intel’s share price. Should have paid more attention to the balance sheet. I thought he would eventually turn the company around but Intel was priced for bankruptcy.
Stock price has tripled since last August. Hopefully, Intel is really back. They do need a couple of Fab customers.
Was it Pat or Brian? If I recall correctly, it was under Brian when Intel had one of its worst periods of stagnation, when the 10 nm process all the bets were on turned out to be a non-starter, and when Meltdown and Spectre erupted. It's easy to overlook this because Intel had fairly no competition around then, but that doesn't mean the company was in a good shape.
I've always felt like Pat was a scapegoat who was chosen to clean up the mess when the whole place was already up in smoke and the smell was only starting to leak out. I liked his strategy, was disappointed to see him booted out.
BK really destroyed the company and Bob Swan was the finance guy who did not have a vision. Pat was the visionary who saw the value of the fabs but it took a long time to turn things around.
>Bob Swan was the finance guy who did not have a vision
Let's be fair to Bob, he has the vision, but dont know how to execute it because he lacks the technical knowledge.
He was also the one who finally settled the argument I had for 5 years, if Intel were to made 250M Modem for Apple, where is the additional Capex for capacity expansion on their Fabs. The answer, only to be told by Bob in a 2020 interview was they never really planned for it.
>Pat killed Intel’s share price. Should have paid more attention to the balance sheet.
Let's face it. Everything he cut, and products / department he sold were what he wanted to do on day one. He had to force his hand, make the stock price worst and ultimately force the board to allow him to do it.
You could argue he laid all the foundation for today's Intel to thrive.
Yes, I am pointing the fingers at the board. Although words on the street was the board also have their hands tied as they were also beholden to large institutional investors. It is all a Game of Cards.
I don't understand what you mean. Stock price has tripled since last August based on Intel finally having a competitive architecture and a competitive process again, no? At least that in combination with various geopolitical circumstances. Sounds like Pat's decision resulted in Intel's stock price rising?
Much like a president, most of their decisions take years to really be felt. The major changes Pat made weren’t going to change the balance sheet for years.
What exactly has Intel done that’s dramatically different than Pat’s vision that you feel is increasing share price?
Or is it just that they sold their soul to Trump combined with Pat’s choices finally bearing fruit?
I said it at the time and I reiterate it now. Pat had the right strategy and was the right person for the job. His decision are justified. He could have played the politics better with the board and externally and that is why I think he got fired, but he laid the ground work for where they are now.
Sadly nothing about why Pat was forced out of Intel.
My speculation:
* Intel in 2019 had $72 billion in revenue and 110k employees. When he was fired in 2024, Intel had $54 billion in revenue and 124k employees. He also didn't cut dividend until 2024. He nearly put Intel into the grave in 3 years.
* Under Pat, Intel's roadmap was a mess. When you look at AMD's roadmap, it made a ton of sense, predictable, and there are rarely any delays or cancellations. When you looked at an Intel roadmap, expect 30% of them cancelled, 50% delayed by 1-2 quarters, and 30% switched to a different node tech.
* IFS strategy under Pat amounted to nothing. Fab cancellations or on hold. Cancelling 20A altogether. Not being able to woo any notable customers. Not hiring the right people for external customers. There were reports that he didn't know how to run an external fab, which is why the board decided to hire Lip Bu Tan who came from Cadence.
* He missed on AI boom, crypto boom, said AMD was in rear view mirror, politicized TSMC in order to win government grants while still still using TSMC nodes themself.
He was not allowed to cut dividend until 2024. Cutting dividend was the first thing he wanted before he even became CEO. The board said no.
Edit: Would have had ~$20B savings, $10B from early sell off, started the work 3 years earlier. In hindsight the best was probably hire someone to do the dirty work before Pat took over. But again, the board had no idea what they were doing. Or they have their hands tied as well ( Which I dont buy into it much. )
>Intel in 2019 had $72 billion in revenue and 110k employees.
>Under Pat, Intel's roadmap was a mess. When you look at AMD's roadmap, it made a ton of sense, predictable, and there are rarely any delays or cancellations. When you looked at an Intel roadmap, expect 30% of them cancelled, 50% delayed by 1-2 quarters, and 30% switched to a different node tech.
He become CEO in 2021 Feb. He was also the one who sorted out Intel's roadmap from a mess to something comprehensible.
>IFS strategy under Pat amounted to nothing
No person on planet earth could have executed IFS in under 3 years. Especially with little to no cashflow. It was always a long term goal. That is why he has to play the government subsidy card, when he was, again not allow to cut dividend. Edit: And I might just add, getting Intel's IFS to have a separate balance sheet took longer than expected precisely because it would expose certain numbers that could affect stock price. cough Guess who were against it.
>He missed on AI boom, crypto boom
Considering how long the cycle of chip works. I am not sure how he missed any of it.
> Intel in 2019 had $72 billion in revenue and 110k employees. When he was fired in 2024, Intel had $54 billion in revenue and 124k employees. He also didn't cut dividend until 2024. He nearly put Intel into the grave in 3 years.
That's nonsense. Intel's problems were caused by falling behind TSMC, which was not the fault of Pat Gelsinger. Quite the contrary, he did a lot to turn things around, but this takes time, a lot more than three years.
Under Pat, Intel overestimated the covid boom, which nearly put them into bankruptcy. They hired many more employees while their revenue tanked. He went on an IFS spending spree while he lacked execution on actually winning external customers. He didn't stop dividends until 2024 which tells me he lacked awareness of where Intel was as late as 2024.
LBT was hired because he could actually see Intel as what it is today, instead of the Intel dominance that Pat thought he was still operating in.
Sure, if they tried to spend all of their assets on buying back all of the existing stock that would be a sign they might be going bankrupt. Alternately, maybe the market cap isn't a perfect representation of how much a company is worth and it's a reflection of a complex array of factors that aren't easily explained away in favor of a simple narrative.
As others have noted, there was over hiring. I think part of it was driven by Pat’s desire to get governments to give subsidies but in retrospect, it led to too many empty shells being built and too many people getting hired. Intel Foundry eventually got Intel 18A node out and that is a promising node. However, the demand for it is now predicated on the AI boom continuing and TSMC being constrained for capacity. Once that goes away, who knows what happens for demand for non TSMC nodes.
Issuing a dividend is a terrible sign for the market. It's telling investors "we have so much money that we have no idea what to do with, please take some of it yourself and invest it somewhere other than Intel."
Have always felt stock buy-backs are equally terrible.
“We have all this money but we have no long term vision for how to grow the company. We’ll just keep doing what we’ve been doing prop up our share price instead”.
That's better than companies that have poor visions for growing the company where the money gets spent on projects that and don't generate an ROI for investors.
On a broader societal basis, it's often better for companies to stay in their lane, return profits to shareholders, and let shareholders invest in new companies with more growth potential.
For a blue chip that's reliably issued one for years, it's also a major part of the reason to own the stock. Throwing that away will not have a good impact on the value even if it's justified.
But value is meaningless. The ultimate purpose for a producer in a market is to provide infinite value for no cost.
Of course that is a pipe dream, so it should provide the highest value for the lowest cost.
For those who want to argue it should be a balance: consider the opposite position. A producer should provide no value at infinite cost. In this case, everything withers. If party A and party B need each other's products to survive, they can do that when the value is infinite and the cost is zero, but not when the value is zero and the cost is infinite.
The last few decades have shown that giving the finger to the customer and going all in on shareholdermaxxing has nothing but terrible effects and is like sticking a spanner into the wheel of capitalism.
My G4 succumbed to this issue, and I was never able to revive it. I had some important documents and images there that I hadn't yet backed up to cloud that disappeared along with it. Still very sour about that. Other than that I enjoyed the phone, felt the dimensions were perfect and the camera was good for its time. But a defect of that nature is too serious to overlook so that was the last LG phone I ever owned.
(Comparing Q3.5-27B to G4 26B A4B and G4 31B specifically)
I'd assume Q3.5-35B-A3B would performe worse than the Q3.5 deep 27B model, but the cards you pasted above, somehow show that for ELO and TAU2 it's the other way around...
Very impressed by unsloth's team releasing the GGUF so quickly, if that's like the qwen 3.5, I'll wait a few more days in case they make a major update.
Overall great news if it's at parity or slightly better than Qwen 3.5 open weights, hope to see both of these evolve in the sub-32GB-RAM space. Disappointed in Mistral/Ministral being so far behind these US & Chinese models
That Pareto plot doesn't seem include the Gemma 4 models anywhere (not just not at the frontier), likely because pricing wasn't available when the chart was generated. At least, I can't find the Gemma 4 models there. So, not particularly relevant until it is updated for the models released today.
Gemma 4 31B has now wiped out several of those models from the pareto frontier, now that it has pricing. Gemma 4 26B A4B has an Elo, but no pricing, so it still isn't on that chart. The Gemma 4 E2B/E4B models still aren't on the arena at all, but I expect them to move the pareto frontier as well if they're ever added, based on how well they've performed in general.
> Very impressed by unsloth's team releasing the GGUF so quickly, if that's like the qwen 3.5, I'll wait a few more days in case they make a major update.
Same here. I can't wait until mlx-community releases MLX optimized versions of these models as well, but happily running the GGUFs in the meantime!
absolute n00b here is very confused about the many variations; it looks like the Mac optimized MX versions aren’t available in Ollama yet (I mostly use claude code with this)
the benchmarks showing the "old" Chinese qwen models performing basically on par with this fancy new release kinda has me thinking the google models are DOA no? what am I missing?
I just tried with llama.cpp RTX4090 (24GB) GGUF unsloth quant UD_Q4_K_XL
You can probably run them all. G4 31B runs at ~5tok/s , G4 26B A4B runs at ~150 tok/s.
You can run Q3.5-35B-A3B at ~100 tok/s.
I tried G4 26B A4B as a drop-in replacement of Q3.5-35B-A3B for some custom agents and G4 doesn't respect the prompt rules at all. (I added <|think|> in the system prompt as described (but have not spend time checking if the reasoning was effectively on). I'll need to investigate further but it doesn't seem promising.
I also tried G4 26B A4B with images in the webui, and it works quite well.
I have not yet tried the smaller models with audio.
> I'll need to investigate further but it doesn't seem promising.
That's what I meant by "waiting a few days for updates" in my other comment. Qwen 3.5 release, I remember a lot of complaints about: "tool calling isn't working properly" etc.
That was fixed shortly after: there was some template parsing work in llama.cpp. and unsloth pulled out some models and brought back better one for improving something else I can't quite remember, better done Quantization or something...
The model does call tools successfully giving sensible parameters but it seems to not picking the right ones in the right order.
I'll try in a few days. It's great to be able to test it already a few hours after the release. It's the bleeding edge as I had to pull the last from main. And with all the supply chain issues happening everywhere, bleeding edge is always more risky from a security point of view.
There is always also the possibility to fine-tune the model later to make sure it can complete the custom task correctly. But the code for doing some Lora for gemma4 is probably not yet available. The 50% extra speed seems really tempting.
Thank you. I have the same card, and I noticed the same ~100 TPS when I ran Q3.5-35B-A3B. G4 26B A4B running at 150TPS is a 50% performance gain. That's pretty huge.
Reversing the X and Y axis, adding in a few other random models, and dropping all the small Qwens makes this worse than useless as a Qwen 3.5 comparison, it’s actively misleading. If you’re using AI, please don’t rush to copy paste output :/
I transposed the table so that it's readable on mobile devices.
I should have mentioned that the Qwen 3.5 benchmarks were from the Qwen3.5-122B-A10B model card (which includes GPT-5-mini and GPT-OSS-120B); apologies for not including the smaller Qwen 3.5 models.
I have an app I've been working on for 2.5 years and felt kinda stupid making sure llama.cpp worked everywhere, including Android and iOS.
The 0.8B beats every <= 7B model I've used on tool use and can do RAG. Like you could ship it to someone who didn't know AI and it can do all the basics and leave UX intact.
Because TI has a ton of microcontrollers, power management ICs, opamps and so on that doesn't need or is even desirable to produce on smaller processes.
Was it Task Gallery from Microsoft Research?
https://www.microsoft.com/en-us/research/publication/the-tas...
https://dl.acm.org/doi/10.1145/332040.332482
reply