Dissecting AMD Ryzen - CPU Engineering Discussion, ft. Wendell & AMD Engineer Amit
video description
Date: 2025-06-22
Related videos
Comments and reviews: 20
DwindleFlip
This is what I'm here for. I mean this is just good fun for anyone that really loves this stuff. It's nice to see a discussion/presentation at this level of detail. I could listen to this all day because it's fascinating and educational. Thank you AMD for your incredible team and making them accessible to the public, so that we can truly appreciate the massive amount of work and care that goes into chip design. Most people think of electricity as magic and get on with their daily lives. Computers must seem just as magical to most but for the truly curious, these kinds of conversations can open entirely new worlds to some who may go on to build the next generation of processors. Gamers Nexus AMD Wendell... this one was a home run. Loved it. I wish it never ended :) It's always fun when the AMD guys are on. As a side not, Nvidia may think they are penalizing GN by denying future access to it's engineers, but all it really does is make the rest of us sad that we no longer can watch great conversations from the brilliant minds that contribute to this thing we all love. From Threadrippers, to 9950x3d, and 5090's FEs.... its just fun to meet some of the people behind all of it. AMD has really done a great job by welcoming Steve and all of us into their fascinating world. Hopefully Nvidia will continue to as well.
reply
This is what I'm here for. I mean this is just good fun for anyone that really loves this stuff. It's nice to see a discussion/presentation at this level of detail. I could listen to this all day because it's fascinating and educational. Thank you AMD for your incredible team and making them accessible to the public, so that we can truly appreciate the massive amount of work and care that goes into chip design. Most people think of electricity as magic and get on with their daily lives. Computers must seem just as magical to most but for the truly curious, these kinds of conversations can open entirely new worlds to some who may go on to build the next generation of processors. Gamers Nexus AMD Wendell... this one was a home run. Loved it. I wish it never ended :) It's always fun when the AMD guys are on. As a side not, Nvidia may think they are penalizing GN by denying future access to it's engineers, but all it really does is make the rest of us sad that we no longer can watch great conversations from the brilliant minds that contribute to this thing we all love. From Threadrippers, to 9950x3d, and 5090's FEs.... its just fun to meet some of the people behind all of it. AMD has really done a great job by welcoming Steve and all of us into their fascinating world. Hopefully Nvidia will continue to as well.
reply
EkiToji
With the recent talks of banning export to China of EDA tools it's going to be rather interesting going forward to see how the dominant American companies like Synopsys and Cadence might get supplanted by others although the process could be long and grueling. Even if you've ever done fairly basic circuit analysis with SPICE, some of the bigger implementations like HSPICE and PSPICE are maintained by those companies. I've mostly only worked on Cadence and it's pretty essential for even fairly basic amplifier design let along a modern CPU. TSMC as an example will put out model libraries for their process node so you already know what the transistors are capable of. With those design constraints in mind you can then do your IC layout and circuit simulations and be quite sure it will work doing all of the signoff testing. Without all of those tools everything would go back generations and be likely an order of magnitude more expensive because you'd have to do multiple iterations of very expensive photolithography masks to try hardware implementation and testing. You'd still go through multiple spins but without accurate modeling and testing in software I can't even imagine the struggles when you're dealing with billions of transistors.
reply
With the recent talks of banning export to China of EDA tools it's going to be rather interesting going forward to see how the dominant American companies like Synopsys and Cadence might get supplanted by others although the process could be long and grueling. Even if you've ever done fairly basic circuit analysis with SPICE, some of the bigger implementations like HSPICE and PSPICE are maintained by those companies. I've mostly only worked on Cadence and it's pretty essential for even fairly basic amplifier design let along a modern CPU. TSMC as an example will put out model libraries for their process node so you already know what the transistors are capable of. With those design constraints in mind you can then do your IC layout and circuit simulations and be quite sure it will work doing all of the signoff testing. Without all of those tools everything would go back generations and be likely an order of magnitude more expensive because you'd have to do multiple iterations of very expensive photolithography masks to try hardware implementation and testing. You'd still go through multiple spins but without accurate modeling and testing in software I can't even imagine the struggles when you're dealing with billions of transistors.
reply
BigFarm_ah365
I wonder how many small errors like the wrong mask layer have slipped through and made someone's life a frustrating, living hell. I know my Vega 64 ran incredibly badly for whatever reason, just glitchy with constant stuttering and incredibly bad 1% lows, but it was my first gaming PC(second GPU). I lived with it for far too long and only when I was forced to go back to my Nitro Fury, on community drivers, did it really hit me how bad it had been.
It's amazing to think about how small the components have become. I know I find myself dealing with a world where wires are so thin they can break when soldering and fasteners have gotten to a point where it takes a fair bit of luck for me to operate them. In some ways it would be nice if they progressed to such a scale that I couldn't use my fat, beat up fingers. Like this is the same stuff that people used to solder point to point or with through hole components to a board. When was the exact point where the guy just through down the iron and said there's got to be another way
reply
I wonder how many small errors like the wrong mask layer have slipped through and made someone's life a frustrating, living hell. I know my Vega 64 ran incredibly badly for whatever reason, just glitchy with constant stuttering and incredibly bad 1% lows, but it was my first gaming PC(second GPU). I lived with it for far too long and only when I was forced to go back to my Nitro Fury, on community drivers, did it really hit me how bad it had been.
It's amazing to think about how small the components have become. I know I find myself dealing with a world where wires are so thin they can break when soldering and fasteners have gotten to a point where it takes a fair bit of luck for me to operate them. In some ways it would be nice if they progressed to such a scale that I couldn't use my fat, beat up fingers. Like this is the same stuff that people used to solder point to point or with through hole components to a board. When was the exact point where the guy just through down the iron and said there's got to be another way
reply
Aharpoon24
sufficiently advanced science is indistinguishable from magic. I shall take this rock (silicate), lightning, and dead compressed biomaterial (oil) and make it show you moving pictures and sound. It will control these metal objects that do everything from brewing your drink to sailing the stars.
I would love to see the progression from '60s era warehouse size computers to current era CPUs, how each discovery was made and how those led to greater innovation, smaller size, more processing power, and greater efficiency. How PCB was initially developed and how to put traces, vias and layers to it to make a circuit board, then to mother / daughter boards, how it was discovered that capacitors, resistors, chips etc were needed, how they were designed and developed, how different voltages were needed and why.........
reply
sufficiently advanced science is indistinguishable from magic. I shall take this rock (silicate), lightning, and dead compressed biomaterial (oil) and make it show you moving pictures and sound. It will control these metal objects that do everything from brewing your drink to sailing the stars.
I would love to see the progression from '60s era warehouse size computers to current era CPUs, how each discovery was made and how those led to greater innovation, smaller size, more processing power, and greater efficiency. How PCB was initially developed and how to put traces, vias and layers to it to make a circuit board, then to mother / daughter boards, how it was discovered that capacitors, resistors, chips etc were needed, how they were designed and developed, how different voltages were needed and why.........
reply
someusername121
From the Intel side it's the same vibe. There's tons of automation in design where it's like drag this USB block into your design and then the software is figuring out how to actually make it fit in there, and it's got design rules based on your process and timings and whatnot. 4 layers of validation and simulation and then you get silicon and it's like oh yea there's some critical bug that needs a new stepping to fix it fuuuuuuuuck see you in 3 months because the fab process is so complicated now that's how long it takes to make something....even if you send the fixed design out immediately. In total if something goes on sale, engineering started at least 4 years ahead of that. Designers will be 2 products ahead so Amit here might be working on Zen 7 shit right now.
reply
From the Intel side it's the same vibe. There's tons of automation in design where it's like drag this USB block into your design and then the software is figuring out how to actually make it fit in there, and it's got design rules based on your process and timings and whatnot. 4 layers of validation and simulation and then you get silicon and it's like oh yea there's some critical bug that needs a new stepping to fix it fuuuuuuuuck see you in 3 months because the fab process is so complicated now that's how long it takes to make something....even if you send the fixed design out immediately. In total if something goes on sale, engineering started at least 4 years ahead of that. Designers will be 2 products ahead so Amit here might be working on Zen 7 shit right now.
reply
Marc_Wolfe
I wonder what a really strip down single core version could do as an arduino. The only level shifting being on the output, for 5 volt or at least 3.3 volt tollerance. Single clock domain, no SMT, no RTOS, no MBED OS. Arduino is practically just single core/thread anyway; not much use for cache layers. Maybe give me a 3GHz arduino so I can bit bang 3G SDI video with C code. Maybe even leave some fancy instructions in there for the people doing other impressive graphics things. Would that be the first x86 Arduino Sure people could use it to learn interesting ways to optimize code on PCs.
reply
I wonder what a really strip down single core version could do as an arduino. The only level shifting being on the output, for 5 volt or at least 3.3 volt tollerance. Single clock domain, no SMT, no RTOS, no MBED OS. Arduino is practically just single core/thread anyway; not much use for cache layers. Maybe give me a 3GHz arduino so I can bit bang 3G SDI video with C code. Maybe even leave some fancy instructions in there for the people doing other impressive graphics things. Would that be the first x86 Arduino Sure people could use it to learn interesting ways to optimize code on PCs.
reply
jimtekkit
This is awesome! If I could suggest a topic for a future deep dive, it would be idle power consumption of desktop CPUs. It's such a grey area and very frustrating to make purchasing decisions based on the contradicting advice on forums. And of course power consumption is always measured at full load but rarely at idle. Especially with Ryzen, monolithic APUs behave different to chiplet architectures with an IO die that consumes a constant 20W of SoC power draw. When it's being used as a 24/7 server at home, that overhead matters a lot and yet the reasons behind it are a mystery.
reply
This is awesome! If I could suggest a topic for a future deep dive, it would be idle power consumption of desktop CPUs. It's such a grey area and very frustrating to make purchasing decisions based on the contradicting advice on forums. And of course power consumption is always measured at full load but rarely at idle. Especially with Ryzen, monolithic APUs behave different to chiplet architectures with an IO die that consumes a constant 20W of SoC power draw. When it's being used as a 24/7 server at home, that overhead matters a lot and yet the reasons behind it are a mystery.
reply
NAMOR5000
Nice video, but sad to see how far behind in technology they are. Too complex and should be simplified using multiplexers and harmonic frequencies and optical, light circuits and communications. With AI added and the latest materials and optical usage it would be finally state of the Art and finally enter 25th century age instead of 20th century. Those who are knowledgeable know exactly what I mean and what the correct words and terms should be used. Not gonna do your work for you. Cute non rocket science video, gave me giggles.
Have a great weekend.
reply
Nice video, but sad to see how far behind in technology they are. Too complex and should be simplified using multiplexers and harmonic frequencies and optical, light circuits and communications. With AI added and the latest materials and optical usage it would be finally state of the Art and finally enter 25th century age instead of 20th century. Those who are knowledgeable know exactly what I mean and what the correct words and terms should be used. Not gonna do your work for you. Cute non rocket science video, gave me giggles.
Have a great weekend.
reply
Randrew
Thank you for this talk with Amit! I've done a little logic design and it can be maddening all the things that must be kept track of - timing has always been a biatch. Even though there are macros and better software to help out everywhere, designing logic for something as complex as a modern CPU is out of this world. And from start of design to finished chips, pretty much every engineering discipline is exercised. I am awed by designers like Amit who really are un-sung heroes who enable our modern tech filled lives.
reply
Thank you for this talk with Amit! I've done a little logic design and it can be maddening all the things that must be kept track of - timing has always been a biatch. Even though there are macros and better software to help out everywhere, designing logic for something as complex as a modern CPU is out of this world. And from start of design to finished chips, pretty much every engineering discipline is exercised. I am awed by designers like Amit who really are un-sung heroes who enable our modern tech filled lives.
reply
robertsneddon731
A company I worked for back in the early 2000s did competitor analysis of other people's silicon. They bought in devices with their chips, demounted them and then analysed them using an ion beam microscope to study the circuitry and silicon structures. IIRC they used xenon ions (big and heavy but chemically inert) to scrape the layers of silicon and interconnects off one by one. We assumed the competition was doing the same to our silicon products, very Spy vs. Spy.
reply
A company I worked for back in the early 2000s did competitor analysis of other people's silicon. They bought in devices with their chips, demounted them and then analysed them using an ion beam microscope to study the circuitry and silicon structures. IIRC they used xenon ions (big and heavy but chemically inert) to scrape the layers of silicon and interconnects off one by one. We assumed the competition was doing the same to our silicon products, very Spy vs. Spy.
reply
Djiohn
not related but, after doing last nvidia update, i started a game and the gpu fan went to i hope 100% but i believe somehow it was more than that they made a crackling sound, like something was about to break, never hit altf4 and restart so fast in my life, after the restart without touching anything was all good.. so yeah new nvdia driver tried to kill my rtx5080, just wanted to type this somewhere in case it will die soon after what just happen :D
reply
not related but, after doing last nvidia update, i started a game and the gpu fan went to i hope 100% but i believe somehow it was more than that they made a crackling sound, like something was about to break, never hit altf4 and restart so fast in my life, after the restart without touching anything was all good.. so yeah new nvdia driver tried to kill my rtx5080, just wanted to type this somewhere in case it will die soon after what just happen :D
reply
storm_rl
Awesome video! Amit has to be one of my favourite guests.
I'm a 4th year electrical/computer systems engineering student but we really dont learn this in the Australian curriculum because there is, as far as I know, 0 industry or opportunity to get into microarchitecture in Aus, and no professors at my university are familiar enough with it to teach it on their own. Still, I'm hopeful that I can find my way into something like this one day.
reply
Awesome video! Amit has to be one of my favourite guests.
I'm a 4th year electrical/computer systems engineering student but we really dont learn this in the Australian curriculum because there is, as far as I know, 0 industry or opportunity to get into microarchitecture in Aus, and no professors at my university are familiar enough with it to teach it on their own. Still, I'm hopeful that I can find my way into something like this one day.
reply
Joel-st5uw
The contaminated lead thing is FASCINATING!!! Some time ago I saw a video (maybe The History Guy) that talked about how a film company (Kodak) learned about secret atomic testing before the public knew b/c it was contaminating their products. Also, steel from sunken WW1 and WW2 ships is sought after for the same reason as lead anchors from antiquity (produced prior to atomic testing)
reply
The contaminated lead thing is FASCINATING!!! Some time ago I saw a video (maybe The History Guy) that talked about how a film company (Kodak) learned about secret atomic testing before the public knew b/c it was contaminating their products. Also, steel from sunken WW1 and WW2 ships is sought after for the same reason as lead anchors from antiquity (produced prior to atomic testing)
reply
JoshWalker1
18:52 I know this story well. My dad was the engineer at HP who hypothesized the possibility of it being the solder balls. Though the timeline is a bit off: low-alpha lead had been known and was a thing in certain applications, and even was specified to be used here, but some supply chain thing was sending regular, thus why it wasn't all those chips everywhere, just some
reply
18:52 I know this story well. My dad was the engineer at HP who hypothesized the possibility of it being the solder balls. Though the timeline is a bit off: low-alpha lead had been known and was a thing in certain applications, and even was specified to be used here, but some supply chain thing was sending regular, thus why it wasn't all those chips everywhere, just some
reply
desert_sky_guy
This was amazing. Thank you guys so much for this, it scratches an itch I've had for a long time and very much refined and upscaled my mind's-eye picture of the concept and scope of the idea of a CPU. Also, wow the project-showstopping problem where radioactive lead emitting game over alpha particles in my CPU die was the answer. Mad props for figuring that one out!
reply
This was amazing. Thank you guys so much for this, it scratches an itch I've had for a long time and very much refined and upscaled my mind's-eye picture of the concept and scope of the idea of a CPU. Also, wow the project-showstopping problem where radioactive lead emitting game over alpha particles in my CPU die was the answer. Mad props for figuring that one out!
reply
TankR
As clunky as it may seem to those not into those kinds of stories, I for one would LOVE to hear all the stories he has about fabbing chips and such. Its those tiny little industry things that dont seem to need to go beyond the front door of the shop.....BUT DAMMIT I WANT TO HEAR THEM!!! So, they have at least one set of ears in the outside world for em.....which is nice.....
reply
As clunky as it may seem to those not into those kinds of stories, I for one would LOVE to hear all the stories he has about fabbing chips and such. Its those tiny little industry things that dont seem to need to go beyond the front door of the shop.....BUT DAMMIT I WANT TO HEAR THEM!!! So, they have at least one set of ears in the outside world for em.....which is nice.....
reply
Nevakonaza.
This was fascinating to watch, The guy from AMD is amazing! I'm actually sad that AMD have not made any Sempron line since forever, Sure it was always a Budget line but they were great little CPUs and it would have been amazing to see on modern architecture what could have become of that line,...now we just get Ryzen and they're great but that's it...just Ryzen :/
reply
This was fascinating to watch, The guy from AMD is amazing! I'm actually sad that AMD have not made any Sempron line since forever, Sure it was always a Budget line but they were great little CPUs and it would have been amazing to see on modern architecture what could have become of that line,...now we just get Ryzen and they're great but that's it...just Ryzen :/
reply
gamersnexus
I don't know what half of these terms mean, but it's still insanely interesting. Like with the inductance problem on the old chip, I went, Hey, I understood that!. Also, the radioactive lead, that's something I understood in theory, but didn't know was a thing.
Something in this video for everyone, electronics backgrounds of any level.
reply
I don't know what half of these terms mean, but it's still insanely interesting. Like with the inductance problem on the old chip, I went, Hey, I understood that!. Also, the radioactive lead, that's something I understood in theory, but didn't know was a thing.
Something in this video for everyone, electronics backgrounds of any level.
reply
matatron.
Those die shots are always a total mystery to me. Even with high-res images of 40-year-old CPUs where you can zoom down to the smallest component, I have no idea what all the colors and shapes actually mean, e.g. what constitutes a transistor, a power or data line, which sections are active components rather than just 'blank' space, etc.
reply
Those die shots are always a total mystery to me. Even with high-res images of 40-year-old CPUs where you can zoom down to the smallest component, I have no idea what all the colors and shapes actually mean, e.g. what constitutes a transistor, a power or data line, which sections are active components rather than just 'blank' space, etc.
reply
osgrov
This is the best type of marketing there is, and I'm really grateful we get to hear the real people talk.
Amit seems like a really cool guy! Loved the stories on the old HP chip as well, that's a nice bonus.
Big thanks to everyone involved in making this, and kudos to AMD for letting it happen. More, please!!
reply
This is the best type of marketing there is, and I'm really grateful we get to hear the real people talk.
Amit seems like a really cool guy! Loved the stories on the old HP chip as well, that's a nice bonus.
Big thanks to everyone involved in making this, and kudos to AMD for letting it happen. More, please!!
reply
Add a review, comment
Other channel videos
