INIYSA retweeted
turns out if you hire/allocate a bunch of HFT perf nerds to design your inference stack, you get incredible results!
"Based on this framework, we have achieved high-throughput, low-latency large-scale inference services for DeepSeek V4.1 Flash. Currently, the service throughput of a single cluster is 155 trillion tokens/day, and the overall revenue is $38 per card per hour (at off-peak pricing)"
zhuanlan.zhihu.com/p/209556029593…
3
9
221
7,160
Dumb take, apple is the complete exact opposite of Rolex
they do high volume with few models
cutting-edge parts in iphones only make sense if you make millions of them, because that is how you dilute the upfront costs
they do high volume with few models
cutting-edge parts in iphones only make sense if you make millions of them, because that is how you dilute the upfront costs
@LeakerApple Ever heard of a company called Rolex?
1
0
33
1,302
INIYSA retweeted
We benchmarked 3 local LLMs on our lab’s Mac Studio M5 Ultra vs dual DGX Sparks! 🔬
Qwen3.8 Flash Next: 164.8 vs 116.0 tok/s
DeepSeek V4 Flash: 74.4 vs 68.0
GLM-5.3 Flash: 73.1 vs 28.4
Mac is surprisingly strong! 😏
Qwen3.8 Flash Next: 164.8 vs 116.0 tok/s
DeepSeek V4 Flash: 74.4 vs 68.0
GLM-5.3 Flash: 73.1 vs 28.4
Mac is surprisingly strong! 😏
25
16
150
40,642
INIYSA retweeted
Tesla’s flying Roadster is wild, but I’ve got to give it to the Gen Z maker in China who built his own flying skateboard.
Ayu’s inspiration was the New Goblin’s board in Spider-Man 3. His first prototype used five jet engines and caught fire during ground testing, so he switched which he says produce over 660 lb of thrust combined.
Small shifts in his weight could tilt the board. He practiced in a safety harness and used a phone-based assistant that tracked the board’s tilt and gave him spoken cues to adjust his stance.
He says the project took about two months, with more than 10 days spent learning to fly it. Now he’s riding it over rivers and through a canyon near his hometown in Hubei.
@elonmusk, got room for this guy on your team?
Ayu’s inspiration was the New Goblin’s board in Spider-Man 3. His first prototype used five jet engines and caught fire during ground testing, so he switched which he says produce over 660 lb of thrust combined.
Small shifts in his weight could tilt the board. He practiced in a safety harness and used a phone-based assistant that tracked the board’s tilt and gave him spoken cues to adjust his stance.
He says the project took about two months, with more than 10 days spent learning to fly it. Now he’s riding it over rivers and through a canyon near his hometown in Hubei.
@elonmusk, got room for this guy on your team?
Turns out the future does have flying cars
51
156
971
53,947
Apple chips pulling off that performance at just 36W, 15W, and 9W is borderline wizardry. Especially in path tracing renders where Nvidia used to dominate.
can't explain that with just node differences.
And M6 isn't even on the chart, which is gonna be even more efficient
can't explain that with just node differences.
And M6 isn't even on the chart, which is gonna be even more efficient
23
24
695
39,656
When Apple first added HW RT with the M3,
I thought, "Who's gonna render Cycles or Redshift on Apple hardware? Nvidia is way better."
I was dead wrong. Apple made insane progress in just two generations.
Especially with unified memory, being able to handle massive scenes that even $6,000+ Nvidia cards choke on is a huge advantage. Same goes for the N1X though
I thought, "Who's gonna render Cycles or Redshift on Apple hardware? Nvidia is way better."
I was dead wrong. Apple made insane progress in just two generations.
Especially with unified memory, being able to handle massive scenes that even $6,000+ Nvidia cards choke on is a huge advantage. Same goes for the N1X though
0
1
38
2,706
