• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

Zen 3: 32MB cache(current) vs 64MB

Jan 1, 2021
743 (2.20/day)
System Name BirthdayBuild
Processor i5-4690K @ 4.3GHz
Motherboard Asus Z97-A Motherboard
Cooling Deepcool Gammaxx 400v2 Tower CPU Cooler
Memory 32GB: 2x8GB G.Skill Ares 1600MHz 9-9-9-24, 2x8GB Kingston HX318C10F/8 1866MHz 10-11-10-30
Video Card(s) NVIDIA RTX 3060Ti Founder's Edition
Storage 500GB 970 Evo Plus NVMe, 120GB 750 Evo, 2x500GB HDDs
Display(s) Dell S2240L Full HD IPS Monitor
Case Corsair Carbide Series 500R White
Audio Device(s) Levi's Bluetooth Headphones
Power Supply Seasonic X-650 Gold PSU (SS-650KM3)
Mouse Logitech Wired Mouse
Keyboard Dell KB213 Wired Keyboard
Software Windows 7 x64
What are your thoughts on the announcement by AMD? Why increase the cache while keeping the CPUs the same?

I think it might prove to be a losing move over a whole new generation by Intel(Alder Lake), which is also getting a die shrink.

I was also going to say "why do it now, and not at Zen 3 launch?", but I think the answer to that is vertical stacking; it must have been difficult/impossible to put that much amount of cache on the original versions of Zen 3 processors. But still, why go with only that? Why not launch Zen 4? It seems like a compromise/stepping stone to me.
Jan 27, 2015
1,111 (0.44/day)
System Name Legion
Processor i9-10850K
Motherboard Asus Prime Z490M Plus
Cooling Arctic Liquid Freezer 2 240mm AIO
Memory G.Skill Ripjaws V 32GB (2 x 16GB) DDR4-3200 F4-3200C16D-32GVK
Video Card(s) EVGA GeForce RTX 2060 KO Ultra
Storage Team Group MP33 Pro 2TB M.2 2280 PCIe NVMe 3.0 x4 + WD Blue 1TB SATA SSD
Display(s) Acer K272HUL 1440p / 34" MSI MAG341CQ 3440x1440
Case Lian Li 205M
Power Supply PowerSpec 650W 80+ Bronze Semi-Modular PS 650BSM
Mouse Logitech MX Anywhere 25
Keyboard Logitech MX Keys
Software Lots
I've always looked at Zen 3 as mostly being Zen 2 with a re-engineered cache.

That said, cache is very important in latency sensitive use cases and doubly so where it has to take the additional load due to the CCD architecture AMD uses.

In the real world that performance is, IMO, all that matters.

However, Zen 2 / 3 and now 4 are not all new microarchitectures. You really don't get those every couple of years, even though the media and fan bases will say it is all new.

Again though, none of that minutia matters if they can get the performance out.

+10-15% in gaming from a CPU is pretty significant and indicates they have shaved a lot of latency off. That may not translate into other types of applications (productivity) where latency isn't a big issue. Also need a very fast very expensive GPU to make use of that in a game.


Staff member
May 14, 2004
23,285 (3.63/day)
Processor Core i7-8700K
Memory 32 GB
Video Card(s) RTX 3080
Display(s) 30" 2560x1600 + 19" 1280x1024
Software Windows 10 64-bit
Apr 30, 2011
2,052 (0.53/day)
Processor AMD Ryzen 5 2600X@95W
Motherboard MSI B450 Tomahawk MAX
Cooling Deepcool Gammaxx 400 Black
Memory 2*8GB PATRIOT PVS416G373C7K@3333MT_C16
Video Card(s) Sapphire Radeon RX 5700 Pulse 8GB
Storage Sandisk SSD 128GB, Kingston A2000 NVMe 1TB, Samsung F1 1TB, Hitachi HUS724040ALE640 4TB
Display(s) AOC 27G2U/BK IPS 144Hz
Audio Device(s) Realtek 7.1 onboard
Power Supply Zalman Z550
Mouse Sharkoon SHARK Force Black
Keyboard Trust GXT280
Software Win 7 sp1 64bit/Win 10 pro 64bit
Benchmark Scores CB R15 64bit: single core 173p, multicore 1306p
3D cache will get only in 12 and 16-core CPUs with 2 CCXs. The 8-core or less CPUs don't need so much cache, or the increase in performance is too small for the higher cost and power draw. My 5c.
Apr 24, 2020
1,379 (2.33/day)
What are your thoughts on the announcement by AMD? Why increase the cache while keeping the CPUs the same?

AMD seems to be on a core / uncore cadance, similar to Intel's old "tick tock" cycle.

AMD makes changes to the core (Zen 1), then the uncore / caches (Zen2), then the core (Zen3), then the uncore / caches (Zen3+SRAM stacking)

Zen3 actually had a pretty drastic change to the uncore though, lol. But the idea is to minimize the changes you do from each generation, to minimize the amounts of mistakes that may occur in any given generation. These are pretty complicated machines: you'd hate to be debugging two performance issues simultaneously if something bad happened.