Let's say as a software engineer, I have the next 10 years to invest my free time to build a "moat" for my own career by "going down the stack" and integrating specialized hardware with my software, where should I start my learning journey or what skills do you think will become essential?
I'm tempted to go to Apple, Google, FB, and AWS's career page and find such info, but I'm really curious to what you have to say!
Man this is a hard question! I don't really know. I think at a high level every company is trying to build all of this out of sight for developers - and defend their own platform. Your "moat" would be what platform you choose. Right now I would guess your best bet is AWS - they likely have the largest staying power. Its like becoming an iOS developer, you're going to have to marry what platform you want to be on. It would be impressive if you could go down the stack yourself tbh.
Ha, I figured it was worth a shot! But yeah it is no doubt a hard question, especially since it obviously depend on what my current skillset is. Going deep on a platform is certainly interesting; I was wondering if you knew of any ways how software can take advantage of specialized hardware. However, it is very possible all that is abstracted away from the developer.
I just found your newsletter through Hackernews. I've very much your piece on Heterogenous Computation and many others. Whenever I read a good tech writing, it always reminds me of the quote "The future is already here – it's just not evenly distributed", your writing certainly gives me that feeling! Please keep up the great work.
Thank you so much! I read a textbook while I was doing research about semiconductors as an investor - and pretty much was like wow this is a big thing right now.
I was shocked to see how little people cared or knew. Nothing I am writing about is at all that controversial - just not well understood. I kind of felt compelled to write about it. This is exactly what I'm trying to do. Thank you :)
Hey man, I'm curious if you plan to provide an update on how have these compute platforms evolved over the past three years. Very curious to understand better Google's Tensorflow platform given the drama since ChatGPT.
What's your preferred way to track each platform and their capabilities? I'd assume that some of the in-house silicon will be used for internal workloads vs cloud compute for clients (e.g. the video chip from Google).
What do you think about the ARM-NVDIA integration? Will that spook the market and, worse, lock each other in the same way Intel fab and design did?
On spooking the market- many IaaS are looking for RISC-V architectures now. I think NVDA's ambition with CUDA was lauded but I think further integration with ARM made its ambition too overt and the customers realized that NVDA will (or at least can) strong arm them into NVDA's own roadmap.
On locking each other in, what if one of the roadmaps are off? Can NVDA, trying to pursue very specific verticals on the design side, still maintain the relevance of ARM in the long run?
I think the biggest problem is if it goes through first. I don't know what will happen in the end - i think everyone will try to move from ARM but that is a decade long endeavor if not more.
If i had to bet - I would bet on the continued momentum of ARM. Almost always ecosystems and platform momentum and beat out a lot of individual actor's desire to move away.
The difficulties in the deal approval feel just made this whole thing even more futile. NVDA revealed its ambition to lock people in and still doesn't have the thing to lock them in quite yet.
Now ARM China is spooked too. As well as even as the hyperscalers are looking at RISCV.
Don't get me wrong, I think NVDA is hugely strategic and valuable. But I don't know if this is quite the right move. Moving from pre-emption to too greedy.
ARM china I think wasn't.... NVDA related. I believe that whole thing happened before the deal was announced. Hyperscalers have a long way before RISC-V, this reminds me of like ~2010s discussion on ARM vs RISC-V. It took a whole platform shift and a decade before we talk like ARM won everything.
RISC-V cannot compete head to head *yet* on many of the products hyperscalers make. It's going to start in IoT IMO.
Let's say as a software engineer, I have the next 10 years to invest my free time to build a "moat" for my own career by "going down the stack" and integrating specialized hardware with my software, where should I start my learning journey or what skills do you think will become essential?
I'm tempted to go to Apple, Google, FB, and AWS's career page and find such info, but I'm really curious to what you have to say!
Thank you!
Man this is a hard question! I don't really know. I think at a high level every company is trying to build all of this out of sight for developers - and defend their own platform. Your "moat" would be what platform you choose. Right now I would guess your best bet is AWS - they likely have the largest staying power. Its like becoming an iOS developer, you're going to have to marry what platform you want to be on. It would be impressive if you could go down the stack yourself tbh.
Ha, I figured it was worth a shot! But yeah it is no doubt a hard question, especially since it obviously depend on what my current skillset is. Going deep on a platform is certainly interesting; I was wondering if you knew of any ways how software can take advantage of specialized hardware. However, it is very possible all that is abstracted away from the developer.
I just found your newsletter through Hackernews. I've very much your piece on Heterogenous Computation and many others. Whenever I read a good tech writing, it always reminds me of the quote "The future is already here – it's just not evenly distributed", your writing certainly gives me that feeling! Please keep up the great work.
Thank you so much! I read a textbook while I was doing research about semiconductors as an investor - and pretty much was like wow this is a big thing right now.
I was shocked to see how little people cared or knew. Nothing I am writing about is at all that controversial - just not well understood. I kind of felt compelled to write about it. This is exactly what I'm trying to do. Thank you :)
I'd love to know what textbook you're talking about too!
Of course! Would you be able to share the textbook by chance?
Hey man, I'm curious if you plan to provide an update on how have these compute platforms evolved over the past three years. Very curious to understand better Google's Tensorflow platform given the drama since ChatGPT.
What's your preferred way to track each platform and their capabilities? I'd assume that some of the in-house silicon will be used for internal workloads vs cloud compute for clients (e.g. the video chip from Google).
What do you think about the ARM-NVDIA integration? Will that spook the market and, worse, lock each other in the same way Intel fab and design did?
On spooking the market- many IaaS are looking for RISC-V architectures now. I think NVDA's ambition with CUDA was lauded but I think further integration with ARM made its ambition too overt and the customers realized that NVDA will (or at least can) strong arm them into NVDA's own roadmap.
On locking each other in, what if one of the roadmaps are off? Can NVDA, trying to pursue very specific verticals on the design side, still maintain the relevance of ARM in the long run?
I think the biggest problem is if it goes through first. I don't know what will happen in the end - i think everyone will try to move from ARM but that is a decade long endeavor if not more.
If i had to bet - I would bet on the continued momentum of ARM. Almost always ecosystems and platform momentum and beat out a lot of individual actor's desire to move away.
The difficulties in the deal approval feel just made this whole thing even more futile. NVDA revealed its ambition to lock people in and still doesn't have the thing to lock them in quite yet.
Now ARM China is spooked too. As well as even as the hyperscalers are looking at RISCV.
Don't get me wrong, I think NVDA is hugely strategic and valuable. But I don't know if this is quite the right move. Moving from pre-emption to too greedy.
Too greedy is pretty much NVDA's MO lol
ARM china I think wasn't.... NVDA related. I believe that whole thing happened before the deal was announced. Hyperscalers have a long way before RISC-V, this reminds me of like ~2010s discussion on ARM vs RISC-V. It took a whole platform shift and a decade before we talk like ARM won everything.
RISC-V cannot compete head to head *yet* on many of the products hyperscalers make. It's going to start in IoT IMO.
Right- good perspective to have with 2010s