32 core Ryzen for Zen 6?

E

Epyon

[H]ard|Gawd
Joined
Oct 25, 2001
Messages
1,218
If they do in fact up the core count would they still limit pci-e lanes or do you think we would get a slight bump?
 
If AM4 from x370/1700x to x570/5950x is an indication, lanes becoming gen5 instead of 4 seem more likely than adding new lane.
 
Last edited:
16 cores is more than enough for mainstream parts. Not to mention there would have to be a change to core counts per CCD as that's still only 8 unless you're talking Zen4c and 5c cores.
 
I'd agree with prior posts that more lanes is outside mainstream consumer-level territory and probably won't happen anytime soon. Between what's offered by the CPU and the chipset 99%+ of that market is covered. The current PCIe 4 lanes will be upgraded to 5 at some point (particularly the link between the CPU and chipset, which IMHO should have been PCIe 5 to begin with for Zen 4 and current chipsets).

To up the number of lanes they'd have to change to a new, larger socket I'd think (though it may be possible AM5 has enough free pins for a few more lanes). It could happen, but AMD seems better than Intel as far as keeping around CPU sockets.
 
Highly doubtful that you'll get more lanes IMHO. Best case would be a chipset with pcie 5.0 upstream on the x4. Maybe the dual chipset boards would run pcie5.0x4 from the cpu for both of them, you'd lose a cpu m.2, but have a much wider path for data through the chipset.
 
criccio said:
16 cores is more than enough for mainstream parts. Not to mention there would have to be a change to core counts per CCD as that's still only 8 unless you're talking Zen4c and 5c cores.
Click to expand...
There was a time when people said that about 4 cores.

Then 8-cores became available and suddenly tons of applications now can use that power.

Applications are designed around what's available.


However, I would be very pissed if AMD came out with 32-core Ryzen AM5 parts, but didn't adjust the SKUs so we would get

Ryzen 5 9600 6/12
Ryzen 7 9700 8/16
Ryzen 7 9800 8/16
Ryzen 9 9900 12/24
Ryzen 9 9950 16/32
Ryzen 9 9990 24/48
Ryzen 9 9995 32/64

After all, wouldn't want to de-value your silicon!
 
Considering the
KazeoHin said:
There was a time when people said that about 4 cores.

Then 8-cores became available and suddenly tons of applications now can use that power.

Applications are designed around what's available.


However, I would be very pissed if AMD came out with 32-core Ryzen AM5 parts, but didn't adjust the SKUs so we would get

Ryzen 5 9600 6/12
Ryzen 7 9700 8/16
Ryzen 7 9800 8/16
Ryzen 9 9900 12/24
Ryzen 9 9950 16/32
Ryzen 9 9990 24/48
Ryzen 9 9995 32/64

After all, wouldn't want to de-value your silicon!
Click to expand...
It is a fine line between desktop users vs power users. They do not want to nutter their Threadripper / EPYC line for workstations by adding too many cores. Sure they can add cores and keep other features like PCIe lanes low, but again, why cross their product lines and potentially lose out.
 
I'd like to see a single 12 core CCD, I don't know if that's reasonable at all, but they could still sell 6 and 8 core parts with 6 and 4 cores disabled respectively. I don't build my projects locally all that often, but it would be nice to have a 24 core part for builds occasionally. I use a 3900x at work still, having roughly 50 percent faster per core and double the cores would be really, really nice. At the end of the day though, working in any game engine, single threaded workloads are still kind unless you're doing compilation.
 
More than concern over PCIe lanes, wouldn't memory bandwidth really start to become an issue with only dual channel memory for 32 cores?

Zen 4 certainly has much higher IPC even before factoring in speed gains compared to Zen 2 when 16 cores first became available. Zen 5 seems like it's going to be anywhere from 10-30% better still than Zen 4 depending on the application, while accepting only slightly faster memory in first gear. Should Zen 6 deliver further on Zen 5 while also giving a speed gain, then there would need to be a major improvement in memory to keep double the cores fed.
 
Zen5 is surely going to still be 8+8 max. Ryzen 9000 will be announced this coming Monday there has been zero chatter in the rumor mill about anything other 6/8/12/16 configs- if there was going to be a big change in the core configs we would have heard something.

There is a remote but nonzero possibility that there could be a Zen5 refresh in 2025 with an 8+16 chip with the 16core CCD being Zen5c (speculation on my part here, not backed up by extant rumors) but I wouldn't hold your breath.
 
NattyKathy said:
Zen5 is surely going to still be 8+8 max. Ryzen 9000 will be announced this coming Monday there has been zero chatter in the rumor mill about anything other 6/8/12/16 configs- if there was going to be a big change in the core configs we would have heard something.

There is a remote but nonzero possibility that there could be a Zen5 refresh in 2025 with an 8+16 chip with the 16core CCD being Zen5c (speculation on my part here, not backed up by extant rumors) but I wouldn't hold your breath.
Click to expand...
One Zen 5 VCache CCD and one 5c CCD would seem to bring the best of both worlds. Great gaming and also great productivity. Something I would invest in.
 
Teenyman45 said:
More than concern over PCIe lanes, wouldn't memory bandwidth really start to become an issue with only dual channel memory for 32 cores?
Click to expand...

Not for my uses (compilation). Memory latency seems to be more of a factor but is much harder to do anything about.
 
uOpt said:
Only if the software plays ball. Scheduling big/little is still a black art.
Click to expand...

Wnat 12-16 Big homogenous cores on a single CCX withion a single CCD. Zen 6 I sure hope has it.

Cause yeah software scheduling with hybrid still quirks!!
 
Give me a 8 core X3D chiplet and a 16 core “C” chiplet.

This gives me the best performance for latency bound tasks (X3D) and for throughput bound (16 “C” cores).
 
You must log in or register to reply here.
Back
Top