High Yield
@highyieldYT
Followers
4K
Following
2K
Statuses
1K
Tech Youtuber. Analyzing hardware and chips of all sizes. Everything silicon.
Joined January 2017
@Meito_1 @JumboShrimp787 @Kepler_L2 @CapFrameX I'm not sure it's the main bottleneck. I think it's a combination of the new cache hierarchy (Intel's new 192KB "L1.5$), a lot of ringbus stops and then the added d2d interconnect.
2
0
1
@JumboShrimp787 @Meito_1 @Kepler_L2 @CapFrameX I think Stix Halo will give us a good idea about Zen 6 and I'm sitting on some ARL material which is still under lock from Intel. Would be a interesting video for sure!
0
0
4
@Quarky93 AMD & SKHynix made something amazing with HBM. Sadly it's now reserved for HPC/AI chips only due to limited supply and CoWoS packaging.
1
0
21
As far as I understand it, it's for smaller LLMs, that are not 100% memory/interconnect bandwidth limited. vboost takes power allocation from the memory system (I think the L2$ in particular) and gives it to the Streaming Multiprocessors, which also contain the tensor cores. It's basically shifting power budget on a more fine grain level and let's you "hand tune" the performance. If you don't need the L2$ perf, you can give that power to the SMs.
0
0
18
@pageworkhere AI TOPS are just the marketing name. TOPS are int operations per second and FLOPS are floating point ops per second.
1
0
1
If you have 5 hours of time, there are worse ways to spend them ⬇️
Here's my 5-hour conversation with @dylan522p and @natolambert on DeepSeek, China, OpenAI, NVIDIA, xAI, Google, Anthropic, Meta, Microsoft, TSMC, Stargate, megacluster buildouts, RL, reasoning, and a lot of other topics at the cutting edge of AI. This is was a mind-blowing, super-technical, and fun conversation. Yes, we discuss r1 and o3-mini, but more importantly we look into the future of technology, geopolitics, and humanity in a world that stands on the precipice of a global AI revolution. The first 4 hours are here on X (4 hours is current limit), and the full 5 hours are up everywhere else. Links in comment. Timestamps: 0:00 - Introduction 3:33 - DeepSeek-R1 and DeepSeek-V3 25:07 - Low cost of training 51:25 - DeepSeek compute cluster 58:57 - Export controls on GPUs to China 1:09:16 - AGI timeline 1:18:41 - China's manufacturing capacity 1:26:36 - Cold war with China 1:31:05 - TSMC and Taiwan 1:54:44 - Best GPUs for AI 2:09:36 - Why DeepSeek is so cheap 2:22:55 - Espionage 2:31:57 - Censorship 2:44:52 - Andrej Karpathy and magic of RL 2:55:23 - OpenAI o3-mini vs DeepSeek r1 3:14:31 - NVIDIA 3:18:58 - GPU smuggling 3:25:36 - DeepSeek training on OpenAI data 3:36:04 - AI megaclusters 4:11:26 - Who wins the race to AGI? 4:21:39 - AI agents 4:30:21 - Programming and AI 4:37:49 - Open source 4:47:01 - Stargate 4:54:30 - Future of AI
2
0
25
@alifahrri2 @Apple @techinsightsinc The competition has high-res die shots. It's just the consumers that don't have them anymore. Every competitor has the money to buy the chips and take a look. And they already do that.
1
0
11
@326powah I'm sure the paid services (e.g. TechInsights, etc.) are doing almost every consumer tech. Hobby stuff like me & @FritzchensFritz is always difficult because 1. it's expensive and 2. it does take a lot of time (literal weeks).
0
1
1
@tomwarren That’s true, but it’s a big chunk. Germany, Netherlands and Denmark I’m pretty sure. You know the exact country list?
1
0
7