- Mar 3, 2017
- 1,747
- 6,598
- 136
We know that Zen5 was in planning months before Zen2 was released in 2019 from an interview early that year, so we can assume at the least that they probably already have Zen6 fairly well in focus now, and with Zen7 on the drawing board regardless of what Intel does.BUT, Intel better get off their duff and make a competing product soon, before AMD turns into the "old" Intel.
And yes, it's still 8 core CCDs for full-fat Zen 5.
Indeed ~100mm2 is too big if going for a 12C CCD going by current AMD's philosophy.Even with this, CCD seems a bit too big for AMD, ~100mm2 is quite big.
Zen2 to Zen 3 is a modest increase in core area, around 15% (~15%+ MTr) (from Locuza annotation of actual die shots) and Zen 3 to Zen 4 is another 18% area (+120% MTr)The core size increase from Zen 2 to Zen 3 was ~35%, and the core logic itself (without L2 and L3) was a ~40% increase.
I'm very certain these core counts are correct, I heard them a few months ago now. I'm also dead certain of who Jim's second "source" is as well - the one that said it was 128c and 192c - and trust me if I were to say who it almost certainly is you'd believe those core counts too.
Anyway at this point I'm also pretty certain about some other info as well, which is making me VERY excited for Zen 5 now.
1. only 8C16T Zen5 CCD means they won't increase core countCan you give us a tiny hint what you are so excited about Uzzi? Like is it about the configuration of the CPU or the IPC or efficiency or release date or what?
Speculation puts it at least moving from 4 wide to 6 wide.The question is how much does IPC increase? For a significant increase of 20%, you need a lot bigger core than Zen4.
Unfortunately there is barely any info in the open to even guess anything. Nothing we hear can be considered reliable. I heard from someone that Zen 5 is a reset of core and SoC architecture at AMD with a broader impact than the original Zen program. But that's about it, no specifics, not even an inkling whether this is remotely true.Can you give us a tiny hint what you are so excited about Uzzi? Like is it about the configuration of the CPU or the IPC or efficiency or release date or what?
Papermaster was talking about Zen 5 during FAD 2022, just that "repipelined front end and wide issue". That's it. 😩Speculation puts it at least moving from 4 wide to 6 wide.
Papermaster seemed to be all but confirm that future cores were going wider in an interview last year that also covered Zen5 and his confidence in its potential - though it doesn't necessarily mean that he was talking about Zen5 when talking about future width of the core.
[A leak] is making me VERY excited for Zen 5 now.
Not much of improvement judging by the rumoured 6000mt/s specs.Curious about the IFOP interconnect more than anything at the moment
These are the pivot points that give us Bulldozer and P4 type products. Fraught with risk.Unfortunately there is barely any info in the open to even guess anything. Nothing we hear can be considered reliable. I heard from someone that Zen 5 is a reset of core and SoC architecture at AMD with a broader impact than the original Zen program. But that's about it, no specifics, not even an inkling whether this is remotely true.
Papermaster was talking about Zen 5 during FAD 2022, just that "repipelined front end and wide issue". That's it. 😩
Yes, kind of frustrating. AMD , Nvidia have Really cranked down on public leaks. Takes some inertia away from the enthusiast crowd - but we aren’t the ones that need to be impressed anymore.Unfortunately there is barely any info in the open to even guess anything. Nothing we hear can be considered reliable. I heard from someone that Zen 5 is a reset of core and SoC architecture at AMD with a broader impact than the original Zen program. But that's about it, no specifics, not even an inkling whether this is remotely true.
Papermaster was talking about Zen 5 during FAD 2022, just that "repipelined front end and wide issue". That's it. 😩
Should be noted that Netburst and Bulldozer had issues with the nodes from two different angles.These are the pivot points that give us Bulldozer and P4 type products. Fraught with risk.
Partly off topic in this thread, but for AMD server remains the clear primary focus. There both the CPU (Genoa, Bergamo, Siena) and GPU (MI250, MI300) development is very promising and improvements seem to appear unabated.Honestly I'm more worried about GPU future with R&D split between 2 different µArchs, AMD still not managing to retake the perf crown from nVidia and now Intel on the scene as a potential contender too.
Partly off topic in this thread, but for AMD server remains the clear primary focus. There both the CPU (Genoa, Bergamo, Siena) and GPU (MI250, MI300) development is very promising and improvements seem to appear unabated.
Consumer space is more mixed. AM5 so far is not the clear upgrade over AM4 it could have been by offering a chip with more than 16 cores. And RDNA3 failed to meet AMD's own prior promises for whatever reason which points to issues in communication and/or execution. These I ascribe to not a lack of capability but a lack of focus or necessity, which I would hope AMD will want to resolve over time.
Zen 4 is still a 4-way machine like Bulldozer (or Conroe!) when the uOP cache doesn't hit. Intel went 5-wide with Sunny in 2019 and 6-wide with Golden in 2021. A new design of 2024 should be wider than Bulldozer, Zen 1, or Conroe. That should bring gains in situation when the uOP cache doesn't provide the necessary backing.What's left is only IPC. If they optimized Zen5 arch heavily, then perf/W should be noticeably better too.
The question is how much does IPC increase? For a significant increase of 20%, you need a lot bigger core than Zen4.
I also expect Zen5 to be wider than Zen4.Zen 4 is still a 4-way machine like Bulldozer (or Conroe!) when the uOP cache doesn't hit. Intel went 5-wide with Sunny in 2019 and 6-wide with Golden in 2021. A new design of 2024 should be wider than Bulldozer, Zen 1, or Conroe. That should bring gains in situation when the uOP cache doesn't provide the necessary backing.
The same applies to the chiplet packaging tech. They are generally still on the 2010 tech having implications for latency and power.
Frequency could go up if AMD lets their processors eat more power. Turin got TDP up to 600W which gives us an indication.
Partly off topic in this thread, but for AMD server remains the clear primary focus. There both the CPU (Genoa, Bergamo, Siena) and GPU (MI250, MI300) development is very promising and improvements seem to appear unabated.
Consumer space is more mixed. AM5 so far is not the clear upgrade over AM4 it could have been by offering a chip with more than 16 cores. And RDNA3 failed to meet AMD's own prior promises for whatever reason which points to issues in communication and/or execution. These I ascribe to not a lack of capability but a lack of focus or necessity, which I would hope AMD will want to resolve over time.
Zen 3, and Zen 4 are designed in mind with this weakness. Zen 4 has >50% larger uOP cache than GLC, while Zen 3 was around the same size as GLC.Zen 4 is still a 4-way machine like Bulldozer (or Conroe!) when the uOP cache doesn't hit. Intel went 5-wide with Sunny in 2019 and 6-wide with Golden in 2021. A new design of 2024 should be wider than Bulldozer, Zen 1, or Conroe. That should bring gains in situation when the uOP cache doesn't provide the necessary backing.
The same applies to the chiplet packaging tech. They are generally still on the 2010 tech having implications for latency and power.
Frequency could go up if AMD lets their processors eat more power. Turin got TDP up to 600W which gives us an indication.
On the client side, before AMD goes to more than 16 cores, IMO, there is not even a point to go from 8 to 16 unless / until there are communications links fast enough to have low core to core latencies, quick access to each other's caches.
Intel can actually hang with AMD in MT workloads now, and while they certainly have an efficiency edge there are plenty of people who don't care about such things.
The kind of people who would want a 24-32 core desktop CPU are those who are probably running rendering software that will gladly scale beyond 16 cores. Core to core latency doesn't matter much in those cases.
There are plenty of workloads that wouldn't care if AMDs solution to offering a 24-core CPU was just adding another chiplet anymore than they would about AMD putting 12 cores on a CCD.
Just don't expect it to be cheap assuming it does exist. AMD would likely try to maintain pricing based on the number of cores similar to the previous generation and use this as an excuse to have a $1,000 CPU. But it's less expensive that Threadripper so some people will buy it.
Probably the biggest problem with all the crazy Zen 5 rumors and speculation...But I don't know how it would scale to EPYC.
You meant this slide below?Not much of improvement judging by the rumoured 6000mt/s specs.
I thought they eventually switch to some kind of active interposer (Zen 6? )
Probably the biggest problem with all the crazy Zen 5 rumors and speculation...
Is how they scale into 8 or even more CCDs while also remaining cost effective
Something interesting though is that there's speculation that Zen 6 would work foremost on cache changes as well as building on Zen 5, and that got me thinking that perhaps crazier chiplet designs might be developed there as cost decreased currently and the near future.
Not that I'll think it will happen, but could AMD create a CPU with three chiplets for AM5?