Satvik Garimella

Recently @arjunharinath1 and I started building our own version of Nvidia's Blackwell GPU. We built the ALUs and a 4-lane SIMD core in Verilog. Here is a breakdown of how we did it.

@CliffLattner @XanderChin @arjunharinath1 Bet! Will do

@satvikgari @XanderChin @arjunharinath1 If you can actually synthesize this design, DM for internship

Recently @arjunharinath1 and I started building our own version of Nvidia's Blackwell GPU. We built the ALUs and a 4-lane SIMD core in Verilog. Here is a breakdown of how we did it.

@michael_trbo @arjunharinath1 I love you Michael 😘😘😘

@satvikgari @arjunharinath1 Very impressive cant wait to see where this goes ;)

@mikaelhaji @arjunharinath1 🙏🙏🙏🙏

@satvikgari @arjunharinath1 goats

@evanliin @arjunharinath1 Ur goated 😘

@satvikgari @arjunharinath1 goated

@XanderChin @arjunharinath1 😹😹 Xchin my goat

@satvikgari @arjunharinath1 the G in Garimella is for GPU

@sakshambatraa @arjunharinath1 🐐

@satvikgari @arjunharinath1 fireee stuff

@kennykgguo @arjunharinath1 We need to run up another DJ Set tgt

@satvikgari @arjunharinath1 incoming goat + dj

Excited to keep working on this with Arjun . Let us know if you notice any inconsistencies or have any suggestions!

We then wanted to confirm that our 4-lane SIMD worked, so we ran it through a waveform. In the waveform, we can see opcode = 1 being broadcast to all lanes at the same time, and each lane producing its result in parallel (15, 12, 50, and 101). Since all outputs appear simultaneously while using the same instruction, it confirms that the lanes are executing in parallel. Seeing that line up exactly as expected definitely made my cortisol drop.

@michael_trbo @sakshambatraa Sooo tufff

so @sakshambatraa and I are working on re-inventing groq's LPU from scratch this last week we implemented the VXM, the LPU's arithmetic unit here's what we learned

Excited to see where this goes!

for my next adventure, @michael_trbo and I will be working together to build a tinyLPU! for our first checkpoint, we reinvented the MXM: the language processing unit's matrix multiplication engine. here's how we did it

@kennykgguo @sakshambatraa @XanderChin @evanliin @suryasure05 Moondager always out when he’s in London

@sakshambatraa @XanderChin @evanliin @suryasure05 kennyg coming out

from getting inspired by symposium last year to presenting on stage this year! live demoing tiny-tpu was insane. this year has been a wild adventure, and it's only getting started

@kennykgguo @XanderChin @evanliin @suryasure05 U gotta be my song guy lmao we were going crazy at medway

@satvikgari @XanderChin @evanliin @suryasure05 u da real dj here