00:00as interactive as possible so we created vera cpu for the age of ai now inside our system
00:08is used for three different ways the first way of course is vera rubin
00:16for thinking and inside the vera rubin rack there are already two cpus as you know we
00:24are building
00:25and selling millions of vera rubins we have sold millions of grace blackwalls nvidia already
00:33is
00:34one of the largest cpu makers in the world there in the vera rubin rack are two
00:39cpus one
00:42for orchestrating and managing the gpus managing the kv cache dealing with all of the software
00:50that runs in the rack we also have the grace bluefield that is used for security and isolation
00:58the vera compute is used for the harness the orchestration of the ai models tool use accessing
01:08the database and the data servers are right here vera bluefield the fastest storage fastest storage
01:17servers the fastest storage system the world has ever made and the reason why this is so vital
01:23is because agents are accessing memory accessing memory so incredibly fast these systems the storage
01:32server and the cpus are now the critical path of the most expensive part of the data center
01:41this is the most expensive for a good reason the economics the economics of the ai factory is tokens
01:51and the tokens are created here and so of course you want to manufacture and generate as many tokens as
01:59possible this is where you put all of your economics and this has to not be in the way and
02:06so vera cpu
02:07has great pressure on the vera on the cpu architecture which is the reason why we built a
02:13brand new
02:14architecture from the ground up a cpu the world has never seen before we call it vera this
02:21is cpu
02:23for agents all the cpus of the past we built for humans this cpu is built for agents
02:31well there are four
02:32things to keep in mind the four takeaways the first takeaway is that the instructions per clock
02:41of vera has to be incredibly good because we need the latency to be short we need the processing
02:47time
02:48single threaded performance not throughput single threaded performance has to be world class absolutely
02:55the best single threaded performance which is the reason why the ipc the instructions per clock of
03:02vera is so high it's the highest in the world 10 instructions fetch decoded and executed per
03:09clock
03:10number one number two the bandwidth necessary to move data in and out for the cpu has to be
03:19utterly
03:20world class the second thing is bandwidth per core the third is just bandwidth period we're moving
03:29remember i said earlier agentic systems is fundamentally disaggregated and distributed
03:37disaggregated and distributed when computing is disaggregated and distributed networking becomes the
03:45problem therefore we have to move the data around as fast as possible between the cpu cores and between
03:52the cpu and the storage the cpu and the gpu the bandwidth around the system and inside
04:00the cpu core
04:01has to be utterly world class this is the first cpu that's been built a long time that
04:07is literally at
04:08reticle limits with a fabric that connects all of the cpu cores that is speed of light 3.
04:166 terabytes per second
04:19no chiplet tax no chip boundary crossings because we need to have everything because the cpu cores
04:27are talking to each other with extremely high bandwidth they're not rented core per core per core
04:34they're all working together the cross-sectional bandwidth of vera is off the charts
04:40so
04:40so
04:40so
04:40so
04:40so
04:40so
04:40so
04:40so
04:40so
04:40so
04:42so
04:42so
04:42so
04:44so
04:44so
04:44so
04:44so
04:44so
04:44so
04:44so
04:44so
04:44so
04:44so
04:44so
04:44so
04:44so
04:44so
04:44so
留言