Today, we'll cover the AMD Computex 2024 keynote in three parts. This first part will focus on the desktop and mobile parts. After that we'll have one section dedicated to the AMD EPYC announcement and another to the Instinct GPU announcement. There will likely be some embedded bits that drive the AMD Embedded Plus announcement but we're not going to cover those in this keynote.
This is being done live, so please excuse the typos.
AMD Computex 2024 Keynote AMD Ryzen 9000
AMD is following up AMD Ryzen with a new AM5 CPU, the Zen 5-based AMD Ryzen 9000 series codenamed “Granite Ridge.” AMD said it plans to support AM5 in 2027 and possibly beyond.
AMD says the Zen 5 will be its next-generation performance CPU core after the Zen 4. Zen 4, for reference, also powers AMD EPYC Genova-like CPUs.
AMD Zen 5 offers “up to 2x” performance for specific areas of the core.
The IPC growth for Zen 5 is said to be 16%. We should note that the Geekbench 5.4 AESXTS test is the biggest development ever. This is a previous version of the legacy benchmark, so it feels a little odd that it's included. At the same time, there is a lot of old code out there.
The flagship will be the AMD Ryzen 9 9950X. It will still have a 170W TDP, but will have 16 cores/ 32 threads, a 5.7GHz boost clock (insane) and 80MB L2+L3 cache. This feels like an incremental improvement, but we could be wrong. Hopefully we'll get parts to test when they're released.
here it is.
AMD says it has some big performance gains over the Intel Core i9-14900K.
Here are the first four AMD Ryzen 9000 series SKUs. Since we lose 4MB of L2+L3 cache moving from 16 cores to 12 cores, it feels like 1MB of L2 cache per core. This would mean 64MB of L3 cache on the Ryzen 9 9950X and 9900X.
We'll get pricing closer to launch in July 2024. Next, we have AMD Ryzen AI
AMD Computex 2024 Keynote 3rd Gen AMD Ryzen AI
Of course, AI is the main focus of PC Computex 2024. For that, and for the notebook market, we have the 3rd Gen AMD Ryzen AI.
It uses AMD Zen 5 cores, with an updated RDNA 3.5 GPU IP, and XDNA 2 NPU.
Here are the highlights of the Copilot+ PC section.
This is the part:
Since the NPU is becoming a huge capability, the new XDNA 2 part is much faster than the current AMD Ryzen 8040 series.
AMD says it has more 8-bit TOPS than Qualcomm, Apple and expects Intel to announce with Lunar Lake.
AMD is now discussing data types with XDNA 2.
AMD plans to support block floating point 16 or block FP16 data types for more performance and better accuracy.
AMD says BlockFP16 will have the higher precision of 16-bit data but with the performance of 8-bit data.
The demo shows a static dispersion XL with a single prompt output between INT8 and FP16 and block FP16.
Here's AMD's comparison to other Copilot+ PCs.
Specifically, here, AMD is saying that Qualcomm is doing a great job with its Snapdragon X Elite.
Last words
There's a lot more, but Intel also has a keynote coming up tomorrow, and there's a TON of it. One of the best things here is that the Copilot+ PCs are a huge inflection point. AMD now has three competitors in its performance charts. A few years ago it was AMD vs. Intel. Now the world is more diverse.
Next, let's get to the AMD EPYC announcements!