Why NVIDIA is Terrified of This New AI Chip

Why NVIDIA is Terrified of This New AI Chip

Siraj Raval

1 месяц назад

7,243 Просмотров

Ссылки и html тэги не поддерживаются


Комментарии:

@ps3301
@ps3301 - 08.10.2024 17:44

If people have to pick the ultimate ai chip, Cerebras is better. Gaudi uses hdm that is always slower than sram.

Ответить
@TheHolyEagle33
@TheHolyEagle33 - 08.10.2024 17:59

whats your stance on analog chips? stochastic computing may be the future of computers.

Ответить
@Voidroamer
@Voidroamer - 08.10.2024 18:19

hey, you are still alive! neat to see you around :)

Ответить
@abhayranade5815
@abhayranade5815 - 08.10.2024 18:30

Nice video sir,one question,is any dedicated sdk available to access amd or intel core ultra NPU?

Ответить
@samiloom8565
@samiloom8565 - 08.10.2024 18:30

Your videos are still very good

Ответить
@BrutalStrike2
@BrutalStrike2 - 08.10.2024 18:59

I bet Intel uploading fake benchmarks as usual 😅

Ответить
@ypey1
@ypey1 - 08.10.2024 20:14

Just bought my intel stock

Ответить
@Kanze-f3y
@Kanze-f3y - 08.10.2024 20:32

Very Sad. Mps wakijadili mtu mambo mhumu ya kusaidia Hawa imeachwa.

Ответить
@epokaixyz
@epokaixyz - 08.10.2024 22:15

Consider this your cheat sheet for applying the video's advice:
1. Explore Intel's Gaudi® offerings for faster and cheaper AI model training.
2. Support companies like Intel that promote open ecosystems and avoid vendor lock-in.
3. Consider the versatility of AI hardware like Gaudi® and its ability to handle diverse workloads.
4. Get hands-on experience with Gaudi® through cloud-based solutions like AWS.

Ответить
@blueberryclash8893
@blueberryclash8893 - 09.10.2024 01:28

intel not recalling their faulty chips they might go bankrupt

Ответить
@mvasa2582
@mvasa2582 - 09.10.2024 04:54

HBM2e and PCIe5 are still behind Nvidia. NVLink is far superior when you construct a large number of clusters. Intel has much more to prove for large-scale 80-100 billion parameter > 8K GPU-sized clusters. Infiniband may not be open standards - but it is far superior. We also have SpectrumX for Ethernet compatibility. Infiniband will play a crucial role in Blackwell when you are guaranteed 800Gbps between nodes or between all 36 GPUs in a Rack.

Yes, this is by far a great option for training and probably the best price performance. We need to see more performance benchmarks. I would also love to hear about how you got your workload distribution and potentially CPU saturation working (if you did).

Ответить
@yuvrajkukreja9727
@yuvrajkukreja9727 - 09.10.2024 07:34

yay ! more cheeper ai hardware that breaks the nvedia monopaly !

Ответить
@levelup89204
@levelup89204 - 09.10.2024 08:54

is crazy how people are sleeping on intel

Ответить
@12_MPH_400_metres
@12_MPH_400_metres - 09.10.2024 23:03

i am most mad about TSMC [Taiwan Semiconductor] has probably been getting subsidy from Taiwan for so many years. now TSMC is getting subsidy from the U.S.A. too.

Ответить
@rousabout7578
@rousabout7578 - 10.10.2024 02:56

Pointless comparing the latest Intel chip with NVIDIAs previous generation chip. How does it compare to the new NVIDIA Blackwell chip?

Ответить
@covenant1231
@covenant1231 - 12.10.2024 05:56

This guy is clueless

Ответить
@frankbraker1320
@frankbraker1320 - 12.10.2024 19:42

"I gotta train my AI model." After hitting the stop recording button on his camera, he goes to the other room to sit down with his baby AI and do exercises. What a good father! :)

Ответить
@DerickEhiobu
@DerickEhiobu - 18.10.2024 15:04

As a tech geek, I also see value in this as an investment advice, it's on you Siraj if this stock doesn't peak 😂

Ответить