• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

AMD Zen 5 rumours

So that's him and Mike Clark who hyped up Zen 5.. Poor show
But at least Mike Clark had at least two excuses:
1. Taking a while CPU apart and putting it back together - for a new grounds up design - and not losing any performance is actually a big thing; more so if the new design has plenty of obvious things they can fix for Zen 6 etc.
2. As chief architect, not talking it up is probably a way to get yourself sidelined or fired.
 
These YouTube video titles etc. are so lame.

Way too emotive.
Unfortunately, like newspapers, the use of sensationalistic titles/headlines is what usually gains attention. I don't like it either (though, I enjoyed the video),
but this style of thumbnail, where you have someone reacting to a given topic/with a facial expression, is what is often employed to appeal to our psychology.
Personally, thumbnails like these (this one is mild, IMO) make me cringe, but sometimes they can be funny, though perhaps not for the intended reason.
 
Last edited:
Hardware unboxed said they contacted AMD about the results and they came back saying they are correct and similar to it's own testing within 1-2 percent,

They didn't share further details how they did the testing for the slides other than the end notes
 
Last edited:
No official post mortem yet anywhere AFAIK. And most of the Zen5 speculation posters have moved on the Zen 6 speculation threads.

However, one thing speculation I saw was that Zen 5 is unbalanced. For instance this post over AT:
It's kind of crazy to think that if you're right, the Z5 core is actually over heavy and it's the Z2 era uncore that is mostly/completely holding it back anymore.
So we can guess what the priority will be for Zen 6.

Sort of goes back to:
That won’t help. Fabric bandwidth is the limitation and amd did nothing there.
Okay, there is probably more than just the fabric at play but current speculation (and more some of the more technical reviews have shown some of this) is that Zen 5 is starved.

So while Zen 5 isn't a bad part - just overhyped and overpriced at launch as per AMD's norm* - it is very little for 22 months!

(* But Zen 1, Zen 2, Zen 3, and Zen 4 did actually outperform their predecessors often by a very large margin.)

Surely AMD would have known this years ago, so why not update the rest of the CPU at the same time?

AMD's R&D budget is no longer tiny:
FbJuq4o.png


Okay, lag times for Zen 5 might mean that we can only really consider the 2022 figures but still.

Wonder how any of this affects server?

AMD are now big enough that they really should do dedicated desktop parts. Which some Zen 6 rumours imply - if their mainstream APUs have be penny pinched to have too little cache, then maybe all desktop parts above Ryzen 3 should come with 3D cache? That would also mean that they have to come up with another idea for the IO of their APUs though - yes less PCIe lanes save power but won't sell on desktop.
 
Sort of goes back to:

Okay, there is probably more than just the fabric at play but current speculation (and more some of the more technical reviews have shown some of this) is that Zen 5 is starved.

In the main, not really. Fabric BW was already a choke for zen4 and increasing core performance just means it becomes an even bigger issue.

When x3d arrives, you’ll see a bigger delta on workloads that fit into L3 when comparing 7800x3d vs 9800x3d than when compared to 7700/9700. Then people will act like some miracle AMD pulled off but in reality the IPC in those situations will be realized rather than being bottlenecked.

In short, Zen5 is hit n miss (mainly miss) because AMD took the lazy option on the IOD.
 
Last edited:
Hardware unboxed said they contacted AMD about the results and they came back saying they are correct and similar to it's own testing within 1-2 percent,

They didn't share further details how they did the testing for the slides other than the end notes
Going forwards I won't believe anything AMD says unless it's a live benchmark i.e. something they show on stage during the presentation.
 
Hardware unboxed said they contacted AMD about the results and they came back saying they are correct and similar to it's own testing within 1-2 percent,

They didn't share further details how they did the testing for the slides other than the end notes

Is there a complete disconnect between AMD and AMD the marketing people?

AMD marketing: +17%

HUB review: +2%

AMD: yes +2% is correct.
 
Last edited:
FbJuq4o.png


Okay, lag times for Zen 5 might mean that we can only really consider the 2022 figures but still.

Wonder how any of this affects server?
I can't remember were but it's been mentioned several times AMD has a couple of teams working on CPU architectures. When one finishes it leap frogs the next one to work on a future design. I'm Zen 5's case that would mean the team that worked on Zen 3 largely designed Zen 5 which would put design stage at around 2020/ early 2021.

As for the server we can look at Linux results and there Zen 5 really shines especially in database and coding benchmarks.

 
I can't remember were but it's been mentioned several times AMD has a couple of teams working on CPU architectures. When one finishes it leap frogs the next one to work on a future design. I'm Zen 5's case that would mean the team that worked on Zen 3 largely designed Zen 5 which would put design stage at around 2020/ early 2021.

As for the server we can look at Linux results and there Zen 5 really shines especially in database and coding benchmarks.

Sorry, I should have been clearer.
By server I meant more whether the chiplets + IOD they use for Epyc will not also starve the CPU:
87Hydmb.jpeg

Certainly desktop can use a new IOD - those reviews with DDR5 8000+ barely move anything.
New faster fabrics always have me worrying about power usage (although see https://www.techpowerup.com/325268/silicon-motion-launches-power-efficient-pcie-gen-5-ssd-controller - finally a PCIe 5.0 NVMe controller which reduces power usage).

While the chiplet approach has served AMD well, I now thing that they need to go for a more complex version. Shorter traces = often substantially less power usage.

Really mainstream desktop would be better served by monolith APUs provided it has enough cache and PCIe + other IO.
 
Last edited:
Sorry, I should have been clearer.
By server I meant more whether the chiplets + IOD they use for Epyc will not also starve the CPU:
87Hydmb.jpeg

Certainly desktop can use a new IOD - those reviews with DDR5 8000+ barely move anything.
New faster fabrics always have me worrying about power usage (although see https://www.techpowerup.com/325268/silicon-motion-launches-power-efficient-pcie-gen-5-ssd-controller - finally a PCIe 5.0 NVMe controller which reduces power usage).

While the chiplet approach has served AMD well, I now thing that they need to go for a more complex version. Shorter traces = often substantially less power usage.

Really mainstream desktop would be better served by monolith APUs provided it has enough cache and PCIe + other IO.

What you’re describing is the approach Intel used. Prices will jump from hundreds to thousands and scaling becomes a huge issue.
 
I can't remember were but it's been mentioned several times AMD has a couple of teams working on CPU architectures. When one finishes it leap frogs the next one to work on a future design. I'm Zen 5's case that would mean the team that worked on Zen 3 largely designed Zen 5 which would put design stage at around 2020/ early 2021.
It's not just AMD, Intel, ARM, Nvidia, Apple etc. - they all do that. What we see now on the market has been really designed a years ago and they are currently working on CPU after Zen 6 internally (zen 7?) , as Zen 6 is mostly ready and it's in testing and tweaking phase now. Same with GPUs and other chips that have new version released every few years. Designing such complex chips, then testing, tweaking, fixing etc. takes years.

This is likely also why new GPU comes and people complain it doesn't have something that competition does and they had a year+ to add it - no, they didn't. Chip has been designed already years ago, before they knew what competition will show and went different direction, then couldn't do anything about it and released that they had. It all takes years to see any changes and it's a lot of guessing game (or industrial espionage :p) in between competing companies.

With regards to IOD, if I recall correctly, they did mention a couple of times in interviews (with GN etc.) that it contains also a lot of analogue components and they don't scale at all down anymore, hence they see no benefits of going with a better process for example. They also said keeping it the same lets them cut down cost and increase margins plus move more engineers to CPU cores design instead. It's simply good though in their opinion. Apparently redesigning it and fabric would most likely also require new socket so they will do it only when planning to change socket too - potentially Zen6 or later.
 
Last edited:
This is likely also why new GPU comes and people complain it doesn't have something that competition does and they had a year+ to add it - no, they didn't. Chip has been designed already years ago, before they knew what competition will show and went different direction, then couldn't do anything about it and released that they had. It all takes years to see any changes and it's a lot of guessing game (or industrial espionage :p) in between competing companies.
While that is plausible it does make you wonder how they slipped in all these cpu AI improvements when AI was only just kicking off around 18 months ago.
 
While that is plausible it does make you wonder how they slipped in all these cpu AI improvements when AI was only just kicking off around 18 months ago.
AI acceleration has been available in Arm chips (mobile phones) for many many years now - mostly to process photos but not only. You can bet they had required bits already designed and waiting to be used - when they design these chips they go through multiple iterations and configurations before they settle on one - swapping to another one 2 years or so before release which already is designed and waiting isn't the same as designing it from scratch in that time.
 
Last edited:
Well that was fun watch. Really don't understsand how AMD got to this point. The release is a waste of time and monies and has done nothing for them. What a complete mess all the way through. Genuinely previous gen consistently coming out as the better option.

 
Mainstream desktop is secondary , Zen 5 architecture built in mind for workstation and server chips energy-efficient, consuming less power, lower operating temperatures which give bigger margins then desktop
 
Last edited:
Mainstream desktop is secondary , Zen 5 architecture built in mind for workstation and server chips energy-efficient, consuming less power, lower operating temperatures which give bigger margins then desktop

This is a post launch narrative once the gaming performance and efficiency claims were debunked. Especially given the street price delta.

In the past, it’s not like workstation and server didn’t benefit from the gains so again, it doesn’t hold up.

As of now, 9600x and 9700x are poor products.
 
This is a post launch narrative once the gaming performance and efficiency claims were debunked. Especially given the street price delta.

In the past, it’s not like workstation and server didn’t benefit from the gains so again, it doesn’t hold up.

As of now, 9600x and 9700x are poor products.

We'll see how the higher core models perform for such tasks think reviews are out today

6/8 cores models will remain poor until they drop down in price, could be decent for itx builds
 
Last edited:
Back
Top Bottom