Discussing Computex, Zen 6 / 7, RDNA 4, and Nvidia's War on Reviewers w/ Level1Techs (Guest Telegrams)
Added 2025-05-23 18:13:10 +0000 UTCAs previously announced, the next episode of Broken Silicon will see Wendell from Level1Techs join us yet again to discuss everything Computex, CPU, and GPU! Wendell is someone who can uniquely speak on more technical subjects, and has a real wealth of knowledge when it comes to all things semiconductors - so don't miss this chance to influence a BS conversation with this much promise. At a minimum, we plan to discuss:
Everything Computex
Nvidia CPUs (Tom knows more...)
Nvidia's War on Reviewers & Truth
RX 9060 XT & RDNA 4
AMD Zen 6 & Zen 7
Intel's Future
You have ~48 hours to submit below!
Last Episode Wendell was on: https://youtu.be/ZK6jAc8LY48?si=1vvzX-EIYOae7wND
Check out Level1Techs Videos: https://www.youtube.com/@Level1Techs/videos
Zen 6 Specs: https://youtu.be/970JyCapx8A?si=HXCgYsM6UPY-IGK1
Zen 6 LP: https://youtu.be/CcXN9xoXwns?si=BAvViKT0a0T0zYYB
Zen 7 Leak: https://youtu.be/bvHrJzB4-MQ?si=m1-iTiSLztuTJQMi
RTX 5080 SUPER 24 GB Leak: https://youtu.be/Z-29AJ5nCyw?si3KQ7qd-TpMdXsC
https://www.theverge.com/pc-gaming/672637/nvidia-rtx-5060-review-meddling-gamersnexus-wake-up-call
Comments
Hi Tom and Wendell. This is a question about Linux binary compatibility and viability for native Linux games. There is a great article from JengaFX, creators of EmberGen, called "The Atrocious State Of Binary Compatibility on Linux and How To Address It". Too keep it short, they state: 1. Containerization (like Flatpack) is not feature ready for pro tools and games 2. That GLIBC (pronounced g-lib-c) is a massive, monolithic system that handles too much. With the GNOME Foundation not caring about breaking compatability with older versions, causing software build years back to stop working. What are your thoughts? Is there any hope for native Linux games to be able to run years after they have been released, without needing to use containerization?
Filip Aničić
2025-05-26 17:28:38 +0000 UTCThis is a fairly long one; Hi Tom and Wendell I came across this fantastic analysis of the 9950x vs the M4Ultra and MAX which confirmed some of my suspicions about bandwidth and compute on some modern CPU's. It compared workloads that are bandwidth limited vs compute limited. It seems the 9950x is quite bandwidth starved for comparison the theoretical bandwidth of the m4 max is equal to the EPYC 9655 which launched at an MSRP of $11852 it makes the launch price of $1999 of the mac studio seem like a bargain even if the bandwidth is shared with the GPU as long as you are running a tasks that stress both. The conclusion he came to which I also agree with is that the Ryzen CPU's (And Intel's but who's counting? lol jk) are WAY overbuilt for their theoretical memory bandwidth we have a GFLOPS/Memory bandwidth of 1946/89.6 for the 9950x and 747/546 for the M4 Max 21.7 vs 1.37 roughly 16x higher. For tasks that have a low operational intensity defined as FLOPS/Byte of DRAM on the low end at about 0.125 to about 1.0 or so in this example, the Apple chip can see up to a 5.6x theoretical advantage and a lot of tasks people do in day to day do might stand to benefit more from the bandwidth than that peak GFLOPS, we need to go all the way to a complexity of about 12.0 FLOPS/BYTE to see the Ryzen match the performance of the m4 Max for example. TLDR: Modern AMD and Intel chips seem to be either very bandwidth constrained for a the amount of peak compute on hand OR have way too much compute for the bandwidth offered depending on your outlook, at least for less complex tasks. Why do you think this is? And will AMD address this issue? Link to the video in question here : https://www.youtube.com/watch?v=dwYaFlnrFgA
MarvoloRiddle
2025-05-26 05:13:23 +0000 UTC
