Rumors: ZEN3+ Cancelled and ZEN5 going for a BIG.little architecture

I'll be ordering one cpu with 4 bitching fast serial cores and 40 slow-ass power optimized cores. I'am willing to pay 350 euros. It's a good offer - you take it.

 

lmao at the post above

 

Amd should just focus on zen4 early next year. Until alder lake they are still on top as it is.

 

Well you do not need more than 4 full cores for gaming. The rest 40 tiny will "support" background, OS and I/O operations. Still bang for buck.

 

no no-no... 4 bitching fast cores for pig coded games like WoW, ARMA, Flight sims etc. Games which are either not well threaded or otherwise demand high IPC/thread.
but 40 slow, power-optimized cores are simultaneously used executing less occupied threads, and not just background/OS. So you end up with great serial speed, great parallel throughoutput (due to sheer number off cores) and great perf/Watt

 

You do need full 6C/12T today. As some of games already show reduced performance on full 6C/6T. Future for DX12/Vulkan, advanced physics/AI will be 8C/16T.
Someone who looks few years into future will not buy today some chip which is just above edge of showing reduced performance.

But big.LITTLE may be pretty interesting for APUs. That's as long as those small cores are really small. Because saving 10~15% of space per core is not worth the sacrifice.

 

I've been saying this for a while. There's no need for a Zen3+ now that Rocket Lake is out. AMD should focus on their next architecture, much like Intel are doing. RL was nothing but a stop-gap product released in an attempt to narrow AMD's performance lead. Introducing new SKUs when you still haven't fulfilled original demand, and with a CPU line that might only yield insignificant performance gains is a waste of time and silicon better used on your current product nodes.

So... sounds like big-little cores are the future then. I don't know how I feel about that. It sounds like from a pure gaming performance standpoint we might be stagnating for a bit again.

 

BigLittle cores in PC desktop? Oh hell no.

 

Zen3+ can't be cancelled, it was never a thing to begin with.

 

Did Zen 3+ existed outside the rumor mill?

Talking about little.big, ARM approach without SMT is proving to provide a safer enviroment, similar single thread performance, smaller cores and restricted power consumption. So little.big can be a solution to retain multithreading power while increasing single core numbers without making them too much bigger. Intel's approach, mixing hyperthreading with small cores is kind of new.

AMD told there is too soon for them to jump in, but nowadays they don't talk too much about future releases, what I think it's ok.

 

lmao big.little... outside mobile and low power laptops there is no need at all of this.

 

The tittle gave me a chuckle. Its not BIG.little in the sense that is an ARM specific architecture but then its called big.LITTLE. I assume you know all that so +1 for being cute.

 

The rumor is for an APU so its likely its not a desktop chip or if it is its targeted at lower end desktops for business etc.

 

Funny how many people here think they know better about what's best. I for one look forward to x86 CPUs getting this kind of architecture. Worst-case scenario, we'll get cheaper products with lower power consumption. Wow, what a joke of an improvement, right!?
/s
Best-case scenario, the little cores can be overclocked to crazy speeds and yield better performance, because most applications don't need most of the instructions x86 comes with. This is why ARM has been taking off in the server market and yields such amazing performance-per-watt for Apple. Ditch all the baggage and you can get something great. x86 can't afford to ditch everything due to compatibility problems, so, they can just provide simpler cores instead.

 

Look at bright side of things. It is light at end of tunnel for intel.

When you have big.LITTLE, SMT/HT becomes useless. (SMT/HT delivers like 25% extra performance per core on average. LITTLE core may deliver even 50% of big one.)
So, all those lovely side channel attacks are gone in instant.

Then it is not really that bad in terms of transistor efficiency. LITTLE cores may be quite bigger than portion of CPU that enables SMT/HT, but they not only deliver higher performance, their use enables shrinking and simplification of big cores.

I would take 8 big + 8 LITTLE cores over 8C/16T w/ SMT/HT.

Only problem is execution capability. One may think that many LITTLE cores replace need for many big cores. But that's false assumption.
If we have only 4 big cores, we are bound to run into main weakness of LITTLE cores. And that's limited instruction set. When there are threads which need big instructions, they can only run on big cores. And it may happen that many games/applications will suddenly live on mere 4C/4T system even if there are additional 16 LITTLE cores.

This step may be quite interesting. Changes may enable even chiplet with 12/16 big cores and chiplets wit 24/32 LITTLE cores.
- - - -
For laptop APUs that will have RDNA2/3, I would not mind 4 big + 4 or 8 LITTLE. Because chance that CPU will limit GPU there will be small.
But in desktop where GPUs can pull hundreds of fps, having only 4 powerful cores is bound to bite people into arse at some point.

 

SMT is meant to full-load the pipeline (and so having a longer pipeline that really can performs near 100%). big.liitle will never remove its benefits, it is just for idle things. and on desktop we don't need it at all, having 2-3W vs 4-5W is no real benefits, all the rest of devices will consume far more. Also SMT advantage is to remove cache access stalls, while big.little produce cache stalls. We already saw intel experiments on this, and was a disaster in performance terms. On desktop we have better power savings reducing by 5% the screen brightness or just removing some crap software in background like the fraud of antimalware software.

 

I'm actually glad it's never existed/cancelled so I can proceed with Zen 3. I'm not in a rush to buy a new motherboard and RAM so this is a positive for me :S

 

@Alessio1989
aaah, glad to see you crystal ball finally gave you a prediction of that billions of ppl on this planet want.

and please do look up the definition of fraud, as you clearly dont understand its meaning.
i have not been defrauded by any free av/am sw in the past 15y (using the known stuff),
and more trouble with leaks/breaches at yahoo/visa and the like.

 

antimalware software are a fraud. the only real risk is the user. antimalware do the exact things of the software they claim to fight, just with different shades (otherwise you would not pay for them ). And most of their efficiency is outperformed by OS kernel mitigations like ASLR, CFG etc.