AMD’s Zen 5 is a missed opportunity in messaging

Masterclass in how to submarine good tech

AMD Ryzen LogoLast week AMD did their ‘Tech Day’ and it was anything but. SemiAccurate is shocked at how badly AMD has regressed in it’s messaging.

We keep telling companies that when they don’t supply the information needed to do the job, the story becomes about them. AMD badly face planted this time with their endless repeated AI b*llsh*t interspersed with the occasional tidbit about the products. It was one of the worst briefs we have been to in years, and to top it off, everyone pretended to no have a hand in the awfulness nor to know who caused it.

After many ‘chats’ about the lack of tech at Tech Day, AMD belatedly said they would do a follow up brief about the tech at a later date. That was Monday and it made matters worse. Why? First what was meant to be a two day deep dive was less than an hour, question time included. That was a joke but AMD made it worse with their intentional hamstringing of the ability to ask questions afterwards and write.

What do we mean? Between the brief and the embargo lift there was less than 48 hours. Between going over the materials again, figuring out what you want to ask about, and getting a response, that tends to be more than the allotted time before the embargo. This is intentional and asinine behavior but AMD keeps doing it. No one is willing to say who to talk to in order to change things much less do the right thing and fix their sh*t behavior. So the story becomes about them, not the product.

All this said, the follow up briefing was a taste of what the Tech Day should have been so we will talk a little about that, but not cover the rest of the BS. Intel does things right, AMD has been told repeatedly for years what to do but they refuse for transparently wrong reasons. How bad was it you ask? There was a one hour session on partners and the GPU, and no we can’t explain how those two things were related, they aren’t. That said during that hour, the GPU WAS NEVER MENTIONED.

AI was. Again and again and again, in the worst most cursory way. Every session ignored the purported topic and literally repeated the same awful AI talk. It will make your life better you know, somehow, someday. And did we mention the noxious rebranding of consumer Ryzen parts to Ryzen AI? Heaven forbid an AMD product reveal with the same brand as last time, that was expected, but the pandering to the stupid has jumped the proverbial shark. Even Intel isn’t this bad.

So what about these chips? We will start out by saying that they aren’t that bad looking but the lower end Ryzen APUs, we won’t use the AI crap names, are not something we would deploy, anywhere, ever, due to the hardware security backdoor that is Pluton. If you think the latest Crowdstrike ‘whoopsie’ was bad, imagine it baked into hardware in the near future. That said last gen the non-APU Ryzens did not have this backdoor, no clue if this generation is similarly safe. Either way, do we need to tell you why a user inaccessible, unblockable, hardware backdoor to your system that is controlled by a third party is a bad thing? Again?

So back to the hardware. AMD put out an AI pro…. die die die. AMD put out a new core called Zen 5. It is the core between the last gen Zen 4 and the upcoming Zen 6, quite the coincidence there, eh? Is it better? Yup. Significantly so? Looks like. Will it beat Intel on performance per Watt in mobile applications? They say so but history says wait for the half step when the power functions are debugged. That said since anything mobile will be the APU, just don’t, Intel doesn’t have this serious security issue.

AMD Zen 4 vs Zen 5

AMD’s Zen 4 vs Zen 5 cores

There is a lot of cool stuff in the Zen 5 core but at this point we would rather chew our hands off than lose two days of sleep trying in vain to write it up and failing because of lack of time. AMD did this by design, again, so we aren’t going to ruin our lives to make up for that crap.

AMD Zen 5 layout

No Time To Explain This Bit

OK we still do care a little at this point, and we do mean a little. The above bits are for the full core, some variants cut bits off here and there, fuse the odd bit down, and inflict other indignities mainly on the data paths. More on that below but not much more.

More Cut Down Than Shown

There are a few interesting points here. First is that the APU version of the CCDs have their small cores’ L3 cut in half. This isn’t a big deal but it should be noted. What is more problematic is that the APUs have their big core AVX-512 units cut down from the full 512b width to a 2-pass 256b unit. The non-APU versions have the full hardware 512b wide unit but, well it may be cut down in software to 256b functionality. There was a lot of confused messaging surrounding this change and the changes to the datapaths it caused on the different iterations. Normally we would try and clear up this confusion but less than 48 hours from brief to embargo lift meant by the time we read through everything and realized that the data wasn’t in the deck, there wasn’t time. Thanks AMD!

ADM Strix Point SOC diagram

Strix Point – Backdoor Not Shown

The above *BASIC* information wasn’t ANYWHERE in the day and a half of ‘Tech Day’. It was only our bitching that forced a follow up that we got this bare minimum information. We still don’t know how fast the USB ports are but we do note that the PCIe port count regressed. And that there is a glaring lack of PCIe5, come on AMD, get with 2022.

AMD RDNA 35 architecture

AMD’s Tweaked GPU Architecture

This was the first time the GPU architecture was mentioned. Again, no time to figure out the details, hope someone else bothers.

AMD XDNA2 architecture

Half a Versal More or Less

The NPU is a little more interesting and AMD, sit down for this, actually gave a good talk about it at Tech Day. That said if you imagine the Xilinx Versal architecture, chop the bottom bits off, and stick it on an AMD CPU, then squint a little, you have the new AMD NPU architecture. It looks to be a solid piece of hardware with nothing useful to do. That is Microsoft’s fault, not AMD’s but AMD isn’t helping the issue with their awful messaging. This doesn’t diminish the hardware though, maybe someday it will have a use. Don’t hold your breath.

AMD Granite Ridge diagram

Granite Ridge Not Rapids

Granite Ridge is the non-APU version of the Zen 5 consumer device family and it has some of the goodies but not all. No USB4 here, this is a good thing, but there is PCIe5, another good thing. The AVX-512 unit is full width but we can’t say if it isn’t fused down to 256b. AMD will probably write us an email to tell us but we don’t care at this point and we won’t care if and when it comes.

So there you have it, and we didn’t bore you to, and then well past the point of pushing a pen into your ear until the hurting stops with painfully dumb and repeated AI speeches. By repeated we mean the same pandering, low-level talk literally a half dozen times in the guise of technical talks. Did we mention how bad it was? If not, let us just say AMD is now the worst of the major companies in terms of messaging, and no a belated brief long after the fact does not make up for it, especially since AMD is destroying it’s reputation with the short embargo times. We would have loved to write this up correctly but AMD made sure that could not happen. As we said, this story is now mainly about them.S|A

The following two tabs change content below.

Charlie Demerjian

Roving engine of chaos and snide remarks at SemiAccurate
Charlie Demerjian is the founder of Stone Arch Networking Services and SemiAccurate.com. SemiAccurate.com is a technology news site; addressing hardware design, software selection, customization, securing and maintenance, with over one million views per month. He is a technologist and analyst specializing in semiconductors, system and network architecture. As head writer of SemiAccurate.com, he regularly advises writers, analysts, and industry executives on technical matters and long lead industry trends. Charlie is also available through Guidepoint and Mosaic. FullyAccurate