Rumour has it that AMD's jumping to 5nm for Zen 3

I feel like this is an unnecessary risk and unlikely. It's also rumoured that the 5000 series will still be on 7nm. So using the 4000 series with 5nm and the 5000 series on 7nm doesn't make sense.
 
I feel like this is an unnecessary risk and unlikely. It's also rumoured that the 5000 series will still be on 7nm. So using the 4000 series with 5nm and the 5000 series on 7nm doesn't make sense.

Well, if the APUs are 7nm, that would make sense for the 5000 series. The APUs tend to be a generation behind.

AMD has already confirmed that Zen 4 will be on a 5nm node. That's on AMD's roadmap.

My gut feeling is that AMD isn't doing this, as they don't need to risk it beat Intel in a lot of markets. That said, with the Huawei situation, AMD has an opportunity to grab onto 5nm earlier to accelerate their roadmap.
 
Intel are already dead in the water in terms of progression.
AMD will be doing this to strengthen their position in the marketplace and grow, they clearly know they cannot get away with Intel style upgrades.
These things will not be cheap mind, AMD do not want to be the "cheap option" anymore.
 
Intel are already dead in the water in terms of progression.
AMD will be doing this to strengthen their position in the marketplace and grow, they clearly know they cannot get away with Intel style upgrades.
These things will not be cheap mind, AMD do not want to be the "cheap option" anymore.

They want to be the only option. So 5nm could be the jackpot of risks if its true and pays off. Enthusiasts are a 1% market for them I guess. Its all about servers, OEMs, etc.
 
Yeah I don't think there's enough evidence to give much weight to this rumour yet, especially given how long ago this change must have been decided. But given AMD's heavy use of HDL for their design(Tbf only Intel really don't use it at every level but for good reasons too) in theory it could have been possible without too many headaches if the design rules didn't vary much, particularly if they just updated the gate node and kept the spacing the same as 7nm(For heat dissipation reasons too, as with Zen+).
 
Last edited:
I would imagine the biggest reason they will do this is rather simply because they can.

There are many other perks to just being on a smaller node and looking cool in front of your mates, though.

Remember, Zen is a chiplet design. That means it is highly successful in yield, because tons of these small chiplets can be sliced from a wafer with minimal waste. The smaller the chiplets? the higher success rate and the lower the cost. After you have factored in the cost of the shrink itself, of course.

If Intel could shrink this easily? you can bet we would have gotten one every single year. Fact is their design SUCKS at it. Yeah they can shrink, but due to the design the clocks fall off of a cliff. That is why they never released their earlier parts, as they could barely muster about 2.8ghz.

Not only are AMD managing to shrink on the regs, but their clock speeds are increasing. And it will get to the point where they ARE faster than Intel on gaming, and Intel will be absolutely up the creek without a paddle.

And from what I am hearing on the ground? that will be Zen 3.

Just to add.

Intel's current tech will not ever make it to 5nm. It simply just does not work. That is why they hired Jim Keller. There is going to come a point soon where Intel are the "new AMD" and suck for a couple of years at least before they come up with something truly new.

They hired him what? two? three? years ago? it took AMD about 5 years to realise what he created. I would imagine it will take Intel about the same, hence the two year quote from a buddy of mine. IE - for two years nothing Intel make will be worth buying (like Piledriver) unless you are an absolute die hard (like all of the people who bought netburst P4s)
 
Last edited:
Don't see this happening at all. They'd need way more time to prep for it for a September release. What I could see happening is a very quick turnaround from the 4000 series to a straight die shrunk 5000 series, both for CPU's and GPU s. I think their GPUs would gain more from snatching an early die shrunk than the CPUs, against their competitors. AMD are far more behind Nvidia than they are Intel in terms of an arms race.
 
Well, if the APUs are 7nm, that would make sense for the 5000 series. The APUs tend to be a generation behind.

AMD has already confirmed that Zen 4 will be on a 5nm node. That's on AMD's roadmap.

My gut feeling is that AMD isn't doing this, as they don't need to risk it beat Intel in a lot of markets. That said, with the Huawei situation, AMD has an opportunity to grab onto 5nm earlier to accelerate their roadmap.

There's really no opportunity there for AMD.

Huawei was only using 15-20% capacity for 7nm. TSMC still are allowed for I believe 90 days to finish their already placed orders. After that there's just more 7nm capacity and I guarantee it's already been filled by anyone using 7nm. This doesn't really allow anything for 5nm. It would still be reserved first and foremost for Apple who is their largest customer.
 
Last edited:
Well, if the APUs are 7nm, that would make sense for the 5000 series. The APUs tend to be a generation behind.

Ahh, yeah, you're right, that makes sense.

AMD has already confirmed that Zen 4 will be on a 5nm node. That's on AMD's roadmap. .

But Zen3 was roadmapped to use 7nm, yet we're now getting rumours it'll use 5nm. Still, if Zen3 APU's use 7nm, the original roadmap would be correct.
 
Don't see this happening at all. They'd need way more time to prep for it for a September release. What I could see happening is a very quick turnaround from the 4000 series to a straight die shrunk 5000 series, both for CPU's and GPU s. I think their GPUs would gain more from snatching an early die shrunk than the CPUs, against their competitors. AMD are far more behind Nvidia than they are Intel in terms of an arms race.
While I agree they'd have to have known about this at the very least a year ago if it's true, I think it's worth noting that these CPU chiplets are tiny compared to GPUs, so would be viable for cheap mass production for CPUs a bit earlier than with larger discrete GPU designs.
 
While I agree they'd have to have known about this at the very least a year ago if it's true, I think it's worth noting that these CPU chiplets are tiny compared to GPUs, so would be viable for cheap mass production for CPUs a bit earlier than with larger discrete GPU designs.
Yeah large chips with low yeilds probably aren't the best for ironing out a new fledgling process. Maybe they'll take the chiplet design to GPUs?
 
I'd expect this gen has already been taped for production. Even if there is virtually no work to be done to migrate to 5nm, there is a tonne of paperwork, testing and bureaucracy involved in these changes, not to mention the testing ect. all of there certifications for meeting different regulations around the world would have to be re-applied for and any copyright protections would have to be updated. I'm not saying that this is not going to happen, just agree that this would have had to be decided a long time ago or add significant delays to the launch. I'd of expected this rumour to have started a while ago too if this was the case.
Maybe will see some refreshed zen 3 on 5nm later on down the road?
 
Yeah large chips with low yeilds probably aren't the best for ironing out a new fledgling process. Maybe they'll take the chiplet design to GPUs?
Definitely one day but we can't pass those kinda bandwidths through interposers without the heat getting insane yet.
 
Back
Top