AMD Computex Keynote Address Liveblog: Big Announcements

[btarunr]

We are live from Computex 2024 in Taipei, where AMD CEO Dr Lisa Su takes center-stage to announce what's next from the world of AMD. The company is expected to announce its new "Zen 5" CPU architecture powering new processors across form-factors. In particular, the company is expected to announce processors for the new AI PC Copilot+ standard announced recently by Microsoft. Several more announcements are expected from the world of datacenters and AI GPU acceleration.



01:26 UTC: Setting the tone of things to come is this graphic flashing on the screens. Given the colors, we expect AI announcements both in the enterprise and client segments, but beginning with the enterprise side of it.


01:38 UTC: And we're up, with a TAITRA introduction of AMD.


01:39 UTC: Generative AI is all the rage. It's going to dominate Computex as a running theme.


01:44 UTC: President of the Taiwan Computer Association introduces AI PCs


01:45 UTC: TAITRA introduces Dr Lisa Su, CEO of AMD, to the stage.


01:46 UTC: AMD uniquely position to power the end-to-end of AI.


01:50 UTC: AMD believes in co-innovating with partners to bring the best AI solutions to the market.


01:51 UTC: Behold Zen 5 and XDNA 2


01:51 UTC: Ryzen 9000 powered by Zen 5


01:52 UTC: Zen 5 brings a massive IPC uplift thanks to a dual-pipelined front end an wider instruction window.


01:53 UTC: 16% IPC uplift.


01:53 UTC: Ryzen 9 9950X: 16-core/32-thread, Zen 5, 80 MB total cache, 170 W TDP. Fastest consumer CPU in the world.


01:54 UTC: AVX-512 performance really up 50%. Gaming performance up by up to 23%.


01:55 UTC: Still launching new AM4 processors!


01:55 UTC: Four new processor models, go on sale in July.


01:56 UTC: AMD has been leading AI PCs, first processors with an NPU.


01:56 UTC: 3rd Gen Ryzen AI, Strix Point.


01:57 UTC: Zen 5 + RDNA 3.5 + XDNA 2.


01:57 UTC: New NPU has 50 TOPS on tap.


01:58 UTC: XDNA 2 delivers 50 AI TOPS INT8 performance.


01:59 UTC: XDNA 2 supports the new block FP 16-bit,performance of 8-bit INT, precision of FP16, no quantization needed.


02:01 UTC: Block FP16 is AMD's new NPU secret sauce.


02:02 UTC: Ryzen 300 meets Microsoft Copilot+ AI PC specifications.


02:02 UTC: Microsoft Copilot+


02:05 UTC: Recall and Cocreator powered by the NPU. The reason NPU is key is that it's always on and draws much less power than an iGPU.


02:06 UTC: Copilot+ needs 40 TOPS, Strix Point has 50 TOPS on tap. Also needed are 250 GB of flash storage and 16 GB of RAM.

02:08 UTC: 150+ AI ISVs onboard.


02:10 UTC: HP president and CEO Enrique Lores unveils what's next for AI PCs.


02:12 UTC: Next-Gen Omnibook, powered by Ryzen AI 300 series processor.


02:15 UTC: Stable Diffusion image generation powered by the NPU.


02:15 UTC: Strix Point also packs a powerful 12-core "Zen 5" CPU, and 16 CU iGPU, that can beat the Apple M3 and Core Ultra 185H.


02:17 UTC: Lenovo CEO takes stage.


02:18 UTC: Lenovo to launch Yoga, ThinkPad, and ThinkBook Copilot+ powered by Ryzen AI processors.


02:20 UTC: Lenovo AI Now is an AI assistant powered by the NPU, runs locally.


02:21 UTC: Lenovo Creator Zone is a generative AI tool that runs locally.l


02:22 UTC: Lenovo Yoga powered by Ryzen AI 300


02:24 UTC: Jonney Shih of ASUS takes center stage.


02:25 UTC: AI PC will be one of the most disruptive changes in the PC ecosystem.


02:25 UTC: Zenbook, ProArt, TUF and ROG notebooks powered by Ryzen AI 300 announcing later today.


02:28 UTC: Available from July

02:28 UTC: ASUS Copilot+ PCs pack several first-party generative AI and assistance apps.


02:30 UTC: ASUS has a vast lineup of AMD-based products.


02:32 UTC: More than 100 design wins, notebooks launching in July.


02:32 UTC: AMD is accelerating AI at large scale.


02:35 UTC: AMD holds 33% of the server processor market. Huge.


02:37 UTC: AMD's high core-count servers mean lower rackspace and energy.


02:39 UTC: EPYC Turin: 3 nm chiplets, Zen 5, 192-core/384-thread.


02:40 UTC: Turin is 3.1x faster than the fastest Xeon.


02:41 UTC: Turin has a hybrid CCD design.

02:42 UTC: On track for 2H 2024.


02:42 UTC: A new AI GPU accelerator announcement!

02:42 UTC: MI300X has been AMD's fastest ramping product.


02:43 UTC: MI300X with the latest software backends can be 30% faster than NVIDIA's H100.


02:44 UTC: Stability.ai co-CEO speaks about the collaboration with AMD.


02:46 UTC: SD 3 public access coming June 12!


02:47 UTC: SD 3 with consumer acceleration generation runs faster than the time it takes to type the prompt.


02:50 UTC: SD creative upscaler reconstructs details for photos.


02:52 UTC: SD working on a new video generative AI model.


02:53 UTC: Satya Nadella highlights the deep partnership between AMD and Microsoft.


02:54 UTC: MI300X accelerates GPT-4.


02:55 UTC: CDNA 4 announcing today, MI350X in 2025. CDNA Next in the following year.


02:57 UTC: MI325X packs 288 GB of 6 TB/s memory, and has a distinct advantage over H200. Available in Q4-2024.


02:58 UTC: MI350X based on 3 nm, 288 GB/s FP4 and FP6 data type. Beats NVIDIA Blackwell B200.


03:00 UTC: Networking for AI must be open: Lisa. AMD part of the Ultra Accelerator Link high-performance fabric.


03:01 UTC: UAL 1.0 on track for later this year. AMD part of the UltraEthernet Alliance for scale-out implementations.


03:03 UTC: And that's a wrap. A product-packed keynote address. Just the way we like it!


View at TechPowerUp Main Site

 

[Daven]

Doesn’t look like the live AMD video is working.

Edit: its working now

 

[wolf]

Really hoping AMD start really trying in the Radeon dGPU space, instead of being wish.com Nvidia + VRAM, at Nvidia prices.

 

[Super Firm Tofu]

It's sad that the desktop 9000 series was announced like it was something to just get out of the way so they could get on with the Eh Eye crapola.

 

[Daven]

Massive drop in TDP for the 6, 8 and 12 core Ryzen 9000s. Looking at the 9950X in Cinebench nT, it looks like all cores would run close to 5.7 GHz in order to beat the 14900k by 21% even with the 17% IPC increase in that app.

 

[Denver]

50% faster than the i9 14900k
But 20% faster than 7950X.
How??

As soon as they start talking about AI, I start to fall asleep lol

 

[Noyand]

50% faster than the i9 14900k
But 20% faster than 7950X.
How??
lol

It's apparently really good at AVX 512, and they probably benched it with Intel baseline profile and the latest version of blender probably rely more on AVX512 I guess

 

[Jism]

Massive drop in TDP for the 6, 8 and 12 core Ryzen 9000s. Looking at the 9950X in Cinebench nT, it looks like all cores would run close to 5.7 GHz in order to beat the 14900k by 21% even with the 17% IPC increase in that app.

It don't matter. Innovation is good for us. Meaning more products, better competitive pricing.

 

[Daven]

50% faster than the i9 14900k
But 20% faster than 7950X.
How??

As soon as they start talking about AI, I start to fall asleep lol

Blender is using AVX512 in the benchmark. Both 7950x and 9950x have the right instructions and the 14900k doesn’t.

 

[Chaitanya]

Will be waiting for AI deepcut from Dr. Ian on both AMD and nVidia keynotes.

 

[phanbuey]

Exciting stuff. Can't wait for these to come out.

 

[Chaitanya]

Jonney Shih of ASUS takes center stage.

I remember Shitsus had AI marketting splattered on their motherboard boxes since socket 939 days and now they have new oppurtunity to splatter even more AI shit on their boxes.

 

[leezhiran]

So they will lanunch x3ds as intel next gen's counterpart. Another half year of waiting

 

[xrli]

Turin has 192 Zen 5 cores per socket... Granite Rapids will be fighting a tough battle

 

[phanbuey]

So they will lanunch x3ds as intel next gen's counterpart. Another half year of waiting

Yep... :/

Might be sooner in the unlikely event arrow lake shows up and slaps.

 

[AusWolf]

I wish they dropped all this Copilot AI bullshit, and said a few words about RDNA 4 instead. What a disappointment!

 

[Denver]

It's apparently really good at AVX 512, and they probably benched it with Intel baseline profile and the latest version of blender probably rely more on AVX512 I guess
View attachment 349756

I was thinking it was the effect of leveling the infinity fabric and memories for fair IPC comparison, but this makes more sense.

Too bad they showed the Strix Point compared to MeteorLake and not phoenix (7840/Z1E).