F5F Stay Refreshed Hardware Desktop Off-Die HBM or L4 cache available on Zen 3 X3D processors?

Off-Die HBM or L4 cache available on Zen 3 X3D processors?

Off-Die HBM or L4 cache available on Zen 3 X3D processors?

S
Say_Say123
Member
120
09-29-2025, 08:11 PM
#1
Zen 3 X3D shows strong potential and even matches entry-level Zen 5 in some areas. AMD could focus on refining memory solutions for Zen 3 instead of pursuing radical changes. Introducing an L4 cache die, akin to Intel’s eDRAM but faster, would be a smart move. Developing a new substrate and compatible CCD using existing technologies could streamline the process. Releasing this in AM5 would bypass AM4 limitations. A more affordable, mature fabrication method delivering comparable or better performance than Zen 5 parts would benefit both AMD and consumers. Unless there are hidden constraints?
S
Say_Say123
09-29-2025, 08:11 PM #1

Zen 3 X3D shows strong potential and even matches entry-level Zen 5 in some areas. AMD could focus on refining memory solutions for Zen 3 instead of pursuing radical changes. Introducing an L4 cache die, akin to Intel’s eDRAM but faster, would be a smart move. Developing a new substrate and compatible CCD using existing technologies could streamline the process. Releasing this in AM5 would bypass AM4 limitations. A more affordable, mature fabrication method delivering comparable or better performance than Zen 5 parts would benefit both AMD and consumers. Unless there are hidden constraints?

A
aoiferox
Member
58
09-30-2025, 04:16 AM
#2
I don't think there is a catch. My former laptop (a MacBook with a 7267U) has 128MB of eDRAM, and that not only acted as a L4 cache for the CPU, but also as a cache for the GPU, like the last-level cache on Lunar Lake. Unfortunately with Intel's ineptitude AMD is run fully by bean counters at this point, so I doubt they will even invest the R&D to try something wacky like this. I hope Arrow Lake is good, because AMD is in their 14nm++++++ phase right now and needs a wakeup call.
A
aoiferox
09-30-2025, 04:16 AM #2

I don't think there is a catch. My former laptop (a MacBook with a 7267U) has 128MB of eDRAM, and that not only acted as a L4 cache for the CPU, but also as a cache for the GPU, like the last-level cache on Lunar Lake. Unfortunately with Intel's ineptitude AMD is run fully by bean counters at this point, so I doubt they will even invest the R&D to try something wacky like this. I hope Arrow Lake is good, because AMD is in their 14nm++++++ phase right now and needs a wakeup call.

A
acekiller990
Member
70
09-30-2025, 09:32 AM
#3
More affordable but still advancing, creating prototypes and testing chips wouldn't be inexpensive. If calculations were off, it could lead to e-waste. Instead of using L4 cache, we might rely more on RAM, and focus on boosting memory controller power so performance jumps from just a bit above 1:1 to something like 2:1—RAM speeds up to match the performance of an AMD infinity fabric in theory.
A
acekiller990
09-30-2025, 09:32 AM #3

More affordable but still advancing, creating prototypes and testing chips wouldn't be inexpensive. If calculations were off, it could lead to e-waste. Instead of using L4 cache, we might rely more on RAM, and focus on boosting memory controller power so performance jumps from just a bit above 1:1 to something like 2:1—RAM speeds up to match the performance of an AMD infinity fabric in theory.