Did you see him shake his head when the lady introducing him said that AMD's success with Zen was because of him? It's remarkable that a man as accomplished as him, still knows that it takes a team to be successful.
People said Moore's law was dead because Intel almost killed Moore's law by selling basically the same quad cores for a decade. Thanks to Jim Keller's work at AMD Moore's law is back.
Jim Keller is an amazing engineer. Thank you for having him present this. This has been a popular misconception I run around screaming; Moore's law is dead!; as it means the possible end of my career. I'll trust Jim that we still have a few decades left until I need to worry again.
Why would it imply the end of your career? Even if transistor density isn't doubling every two years, it's still vital that it improves with every generation.
well,how amazing,working on such a small scale !!! with such a precission.Nature is allso build from tiny things.so we have distance and proportion or time and structure.for smooth flow we probably dont need the smallest scale we reached. We just have to work with THE magical distance. thats why you find exact measurement in the construction of pyramids,only possible when you know the exact size of the smallest stone use before they started to build.thats what UNDER---STANDING means.Standing underneath the fundament.And thanks for the presentation.Great believe!!!
So I am pursuing soil science, hydrology, climate, etc, things limited by their complexity and current computer models when predicting natural systems. This gives me hope for having the tools to run models currently impossible as we need them in coming climate change scenarios. The atomic scale interactions between these non living systems and their more advanced living partners like plants, animals, etc, and the human component is so advanced. I need the help of these future systems.
Beyond climate reactions, I think computing the altering or creating of soils with small additives to extract compounds via the 4 earth sphere interactions on the microscale that are currently in their infancy (or looking at old soil creation/mining methods). A bottom up model to supply building blocks for the agricultural, chemical and materials industry in natural systems to reduce costs. Especially in desert systems deemed not as valuable to current systems. There is no soil in space but the building blocks are there and are here, speeding up processes are complex.
Thank you for sharing this impressive video! One question here. I doubt the memory latency has gone up that high for real. Even though the cycles have been changed, don't we also have to consider the enhanced memory frequency? Thus the memory latency must stay the same, no?
Excellent question. The frequency improvements on the FSB mean that memory throughput continues to climb, however latency does not directly benefit from that. The signal still has to make its way from the load unit, through the logic in the TLB, to the memory controller, across the traces in the motherboard, to the control circuit on the DRAM which then needs to writeback the current row and fetch the one requested, and then the signal needs to make its way back. Even if the signal could travel at the speed of light, it would take a couple of cycles to cross the motherboard to the DIMMs! Between the distances involved, slew rates of components, and number of components involved, total latency really is in the hundreds of cycles.
At this rate of progress it will take minimum 50 years for those 50x. It has been 6 years since introduction of 14nm, and Intel is still on 14nm+++. Also there are diminishing returns with 3D stacking, costs dont go down and heat is the big problem. 24:49 We also saw poor yields from FINFET`s and thats already 3x included which is in the past now.
Intel doubled down aggressively on Dennard scaling. It worked until it suddenly didn't. And that was the pentium 4 mess where 8 GHz was expected by 2003 at some point and the more conservative athlon 64 curb-stomped intel. It's hard to tease appart if intel had been ahead until the K7 due to process advantage or due to aggressively enabling higher clocks.
If you look at the number of transistors in mainstream Intel processors (10600K for example), then it's clear that Moore's Law doesn't work properly anymore. In 2020 we should see 32x more transistors in a mainstream CPU than in the i7-875K.
@@M.-.D Wow, in one other comment I wrote about performance, so I'm being told the point is transistor count and in this comment I wrote about transistor count, so I'm being told his point is performance. My point is that neither transistor count nor performance of CPUs doubles every 2 years.
Yes, he led the microarchitecture design team. He shook his head because the lady introducing him said that the success of Zen was because of him. That isn't true, it takes a massive team (hundreds to possibly thousands) to design a chip.
@@toutagamon Is that a dig at me? Honestly I toyed with 'is this post racist?' for quite a while and even though i can't see any way that anyone can take offence to it i still almost didn't post it 'just in case'. But then i thought its really sad that i am so worried about second guessing someone taking it the wrong way that i wouldn't share what was meant as a lighthearted, end of the world, movie reference.
Moore's law has apperently been dying/dead for the last 20 to 30 years. Still chugging along for at least another 10 years minimum. I suspect we'll get down to the 3nm to 2nm size before real problems arise with keeping pace with Moore's Law. Will we ever get down to less than 1nm? Maybe, but we'll have to wait and see what the next 15 to 20 years brings.
7:20 "everything that can be invented has been invented. -1899", god i wish in 20-100 years time this is what we would sound like to our future generation🤣
Yes in theory Moore’s law is still alive because it’s physically possible to continue to double transistors for a long time to come but we aren’t successful at engineering doubling a these days so yeah Moore’s law is dead as per the official definition 🤷♂️ This isn’t a debate it’s a fact.
he worked at amd and helped building ryzen.. he then left and started working for intel.. and then left intel.. hmmm kind of a shady guy if you ask me.. genius or not.
Architect of CPUs and his body 💪
Damn bro. Jim lifts.
He intimidating foreal
He got that ripped carrying Intel's trash of 14nm+++++++infinity
we were all thinking the same thing, haha.
How humble and straight to the point he is. Absolutely no bullshit. Amazing guy!!! Thanks for sharing on CZcams.
yeah, you can recognize the true geniuses because they don't call themselves geniuses.
Did you see him shake his head when the lady introducing him said that AMD's success with Zen was because of him? It's remarkable that a man as accomplished as him, still knows that it takes a team to be successful.
@@jonathanschmeh5709 Yes! Thought that every interesting myself upon observation.
People said Moore's law was dead because Intel almost killed Moore's law by selling basically the same quad cores for a decade.
Thanks to Jim Keller's work at AMD Moore's law is back.
@BingBong Boopledoop Jim Keller has nothing to do with "Big Data"
Thank you for posting this online. Jim Keller is an awesome guy who has probably positively impacted technology growth more than he gets credit for.
I don’t know anyone who follows the space who doesn’t see him as one the legends which shaped computers as we know them today.
Sorry Keanu, this is the real deal guy for Cyberpunk 2077.
i want to thank the moderator for the very last question
Yes that was the best part for me
Thank you for live streaming!
More of these talks please!
I understood some of those words
Thanks for this amazing presentation by Jim Keller! Thanks for uploading the full Q&A session, too! :)
Wonderful lecture!
Thank you for sharing. I learned a lot and I have changed my mind because it turns out I was thinking linearly.
Amazing talk. Thank you Jim!
Damn! Incredible human being!
The most brilliant lecture I've seen in years.
Jim is just Cool - - we miss you Jim!
Really enjoyed this talk keep watching it many times along with talks by him.
Reminds me of my comp arch prof
I think that's the best one hour on youtube I've ever spent.
what a legend
Now what? guess I'll watch it again.
"The expectation and mindset sets your direction and your possibilities, and it's important..." 8:14
Amazing perspective.
Damn so humble awesome talented guy really an engineer no suits and stuff just awesome guy
This mentality is the one we need to tackle the challenges of the XXIst century and XXII too!
Big shout out to the audience that day for their questions. Many times the question just out right suck or the people there don't raise their hands.
Jim Keller is an amazing engineer. Thank you for having him present this. This has been a popular misconception I run around screaming; Moore's law is dead!; as it means the possible end of my career. I'll trust Jim that we still have a few decades left until I need to worry again.
Why would it imply the end of your career? Even if transistor density isn't doubling every two years, it's still vital that it improves with every generation.
well,how amazing,working on such a small scale !!! with such a precission.Nature is allso build from tiny things.so we have distance and proportion or time and structure.for smooth flow we probably dont need the smallest scale we reached. We just have to work with THE magical distance. thats why you find exact measurement in the construction of pyramids,only possible when you know the exact size of the smallest stone use before they started to build.thats what UNDER---STANDING means.Standing underneath the fundament.And thanks for the presentation.Great believe!!!
9:00 that's hilarious 😂😂😂😂😂😂😂I love this talk ❤️
I'm tearing 🤣🤣🤣
“If you think you’re going to win the lottery, you’re going to win the lottery”.
Powerhouse.
TLDW version "Scaling is a mindset".
Gym keller
Is this Mark Rippetoe’s genius brother?
Any link to the his microarchitecture talk that he mentions? Or any other video of him on microarchitecture
So I am pursuing soil science, hydrology, climate, etc, things limited by their complexity and current computer models when predicting natural systems. This gives me hope for having the tools to run models currently impossible as we need them in coming climate change scenarios. The atomic scale interactions between these non living systems and their more advanced living partners like plants, animals, etc, and the human component is so advanced. I need the help of these future systems.
Beyond climate reactions, I think computing the altering or creating of soils with small additives to extract compounds via the 4 earth sphere interactions on the microscale that are currently in their infancy (or looking at old soil creation/mining methods). A bottom up model to supply building blocks for the agricultural, chemical and materials industry in natural systems to reduce costs. Especially in desert systems deemed not as valuable to current systems. There is no soil in space but the building blocks are there and are here, speeding up processes are complex.
Good lecture. Thank you, subscribed.
Thank you for sharing this impressive video!
One question here. I doubt the memory latency has gone up that high for real. Even though the cycles have been changed, don't we also have to consider the enhanced memory frequency? Thus the memory latency must stay the same, no?
Excellent question. The frequency improvements on the FSB mean that memory throughput continues to climb, however latency does not directly benefit from that. The signal still has to make its way from the load unit, through the logic in the TLB, to the memory controller, across the traces in the motherboard, to the control circuit on the DRAM which then needs to writeback the current row and fetch the one requested, and then the signal needs to make its way back. Even if the signal could travel at the speed of light, it would take a couple of cycles to cross the motherboard to the DIMMs! Between the distances involved, slew rates of components, and number of components involved, total latency really is in the hundreds of cycles.
Ah yes
43:25 Michael McKean asks a question.
How long before Jim gets cast to a Terminator movie. Guy is jacked
Kudos to Lisa Su's sister
At this rate of progress it will take minimum 50 years for those 50x. It has been 6 years since introduction of 14nm, and Intel is still on 14nm+++. Also there are diminishing returns with 3D stacking, costs dont go down and heat is the big problem. 24:49 We also saw poor yields from FINFET`s and thats already 3x included which is in the past now.
Just stick to your armchair and let the driven optimists do the real work.
Skip to 3:04 to get past the dean's comments.
9:20 i like lees law
Jim Keller is back!
He's probably the one who's going to build the next generation of AI chips.
And now he is, joining the firm he funded.
Intel doubled down aggressively on Dennard scaling. It worked until it suddenly didn't. And that was the pentium 4 mess where 8 GHz was expected by 2003 at some point and the more conservative athlon 64 curb-stomped intel. It's hard to tease appart if intel had been ahead until the K7 due to process advantage or due to aggressively enabling higher clocks.
6:35 table
Do you even lift, bro?
Bro, dost though even hoist?
3:07
Old Comment, now irrelevant:
Introductions: 10:30
Talk start: 13:16
Thanks for the timestamp! I cut off the beginning of silence so the introduction starts much earlier in the video.
Why is he buff 💪 as well?
65 BILLION?!?!!?
If you look at the number of transistors in mainstream Intel processors (10600K for example), then it's clear that Moore's Law doesn't work properly anymore. In 2020 we should see 32x more transistors in a mainstream CPU than in the i7-875K.
His point is performance, not transistor count.
@@M.-.D Wow, in one other comment I wrote about performance, so I'm being told the point is transistor count and in this comment I wrote about transistor count, so I'm being told his point is performance.
My point is that neither transistor count nor performance of CPUs doubles every 2 years.
@@Kynareth6 what is your definition of performance?
@@M.-.D the average improvement in how well various applications perform (let's say a median of 20 applications/games performance)
@@M.-.D or you could even look at one of the popular benchmarks
Captions are not available, - they seem to be auto-generated
It's not really transitioning into speed of calculation. Unless you speak about adding numbers
Imagine a 4D CPU.
So was Jim keller involved in the design of zen architecture? seems like he denies it at 2:30
he didn't do as much as people say that he has done.
lol Gave too much credit to Jim Keller he isnt the one!
Jim Keller was just sitting on his ass for 3 years. He was only the lead architect.. He barely did anything.
@@readeh Paper pusher, I'm sure.
Yes, he led the microarchitecture design team. He shook his head because the lady introducing him said that the success of Zen was because of him. That isn't true, it takes a massive team (hundreds to possibly thousands) to design a chip.
Jim is the white Miles Dyson.
man of culture!
@@toutagamon Is that a dig at me? Honestly I toyed with 'is this post racist?' for quite a while and even though i can't see any way that anyone can take offence to it i still almost didn't post it 'just in case'.
But then i thought its really sad that i am so worried about second guessing someone taking it the wrong way that i wouldn't share what was meant as a lighthearted, end of the world, movie reference.
@@Speak_Out_and_Remove_All_Doubt It was a comment in good spirits mate.
knowyourmeme.com/memes/ah-i-see-youre-a-man-of-culture-as-well
Jim Keller if fucking jacked hahaha
Moore's law has apperently been dying/dead for the last 20 to 30 years. Still chugging along for at least another 10 years minimum. I suspect we'll get down to the 3nm to 2nm size before real problems arise with keeping pace with Moore's Law. Will we ever get down to less than 1nm? Maybe, but we'll have to wait and see what the next 15 to 20 years brings.
7:20
"everything that can be invented has been invented. -1899", god i wish in 20-100 years time this is what we would sound like to our future generation🤣
AI is definitely going to design way better chips than humans like him at some point
1:40 - the company that actually made moor's law dead, lol.
alpha
FinFET = UC BERKELEY EECS
Definitely it's over. It was just an intel marketing slogan 😂
Yes in theory Moore’s law is still alive because it’s physically possible to continue to double transistors for a long time to come but we aren’t successful at engineering doubling a these days so yeah Moore’s law is dead as per the official definition 🤷♂️
This isn’t a debate it’s a fact.
my boy Jimmy is swollen af
he worked at amd and helped building ryzen.. he then left and started working for intel.. and then left intel.. hmmm kind of a shady guy if you ask me.. genius or not.
Some dude beating his meat @39:46 XD