AMD Gets 💪STACKED💪 With 3D V-Cache Technology – SRAM Cache 3D Die Stacking – Functional Ryzen 5900X Based Prototype Brings >12% Performance Benefits – EOY Production
3 June 2021

Is AMD CEO, Lisa Su natty or not?

Well that I’m not quite so sure about anymore considering she has just announced the biggest technological steroid shot imaginable for AMD Ryzen processors moving forward.

In what has to be without a doubt to be the biggest shock announcement during Computex 2021 entirely, AMD CEO Lisa Su went out on stage to announce what they’re calling “3D V-Cache Technology” or in essence utilizing “X3D” package technology to physically stack SRAM chips placed over the existing L3 cache structure and not over the control logic within the Zen CCD.


This in practice provides a much lower latency additional cache buffer compared to other intricate forms of additional cache such as the SRAM found on the L4 cache of Intel’s long forgotten Broadwell processors.


The capabilities of AMD’s “3D V-Cache Technology” are almost endless, with AMD being capable of stacking up 8 “V-Cache” SRAM stacks of 32MB, though this is highly theoretical the plausibility however of potential future AMD EPYC processors offering total cache counts well into the gigabytes is certainly jaw dropping.

For instance, current generation Ryzen 5000 series processors feature 32 MB of L3 cache per CCD, given that the Ryzen 9 5900X and 5950X make use of two CCD’s for their given 12-core and 16-core configurations, in total they would both feature 64 MB of L3 cache as standard.

Now the benefits of this X3D SRAM stacking is relatively simple, given that they’re placed directly over the existing L3 cache aiding in latency and of course density purposes, AMD are then able to add just a single stack of this SRAM which would either give the Ryzen 5800/5600X a total of 64MB of L3 cache or give the Ryzen 5950/5900X a total of 128MB of L3 cache.

In fact Lisa Su showcased the benefits of their 3D V-Cache Technology by of course introducing a working prototype featuring the stacked SRAM built on the foundation of an existing Zen 3 processor, to be specifically the Ryzen 9 5900X.

This hodge-podge engineering prototype actually features two stacks of “3D V-Cache” for a grand total of 96MB of L3 cache….. per CCD, in total this here Ryzen 9 5900X features a grand total of 192MB of L3 cache.

Simply astonishing.

You may recall that I said that this was a WORKING prototype, because what better way to demonstrate that cache is cash than to compare the working prototype with 192MB of L3 cache and a conventional 5900X with just 64MB than with some comparative gaming performance figures.


AMD have demoed the working prototype in comparison to the conventional Zen 3 equivalent with a demonstration of Gears 5 with both processors locked to a 4GHz frequency, given that this frequency is rather low questions must be raised about whether or not 3D stacking will in any way disrupt thermals in anyway.

The Gears 5 performance comparison between the two processors showcases the importance of a processor’s cache hierarchy as the prototype featuring the “3D V-Cache” technology (based on the 5900X) at a comparable 4GHz locked frequency showcases a stellar 12% performance boost from what would otherwise be an identical Ryzen 9 5900X processor.

Because the segment quickly moves on to showcase the performance improvements from the 3D stack cache design across other various titles utilizing a verity of different API’s such as DOTA 2 on Vulkan which shows a 18% improvement, to Fortnite on DX12 which gives the 3D V-Cache prototype a 17% performance boost, whereas League of Legends provides just a 4% improvement whereas the most noticeable gain is made on Monster Hunter World which is a quite formidable title which offers dodgy performance and much prefers the strength of single core performance.

In Monster Hunter World, the Ryzen 9 5900X prototype showcases an eyewatering 25% performance gain over the conventional Ryzen 5000 equivalent.

So this is all well and truly dandy but you may ask “when can I buy one”? Well obviously considering AMD themselves were utilizing a functioning prototype doesn’t necessarily dictate that we won’t be seeing such technology for quite some time, in actual fact we might be seeing processors built utilizing this “3D V-Cache” technology as soon as the middle of next year.

Because the chiplet design of Ryzen 5000 was already designed in a way to actually accommodate this stacked SRAM without any major design changes, now that’s the sort of planning ahead that the Intel Corporation desperate aught to considerate.

However post AMD’s Computex presentation, AMD confirmed with Dr. Ian Cutress (Anandtech) that 3D V-Cache will actually be in production for Zen 3 products mind you at the end of the year.

This is shockingly surprising and concerning, with mass production being made available at the end of the year I’m honestly at a loss when it comes to the thought of why exactly they’d opt to utilize this technology on existing Zen 3 products rather instead making this a highlight for their next generation Zen 4 processors which would probably arrive sometime around Q2′ 2022.

Personally I don’t believe in there actually being a “Zen 3+”, I don’t exactly see the point in recycling an existing architecture with what would’ve been marginal tweaks and or a frequency bump but however if such a thing were to actually exist I wouldn’t be opposed to its existence were it to come with 3D V-Cache Technology, but the dots just aren’t exactly adding up here.

Production begins at the end of this year, of which by next year we’d be well into the sort of 15+ month gap we’ve typically had between Zen generations, given that we’re almost halfway through the year now if AMD doesn’t announce this barbaric “Zen 3+” shortly I highly doubt it will ever.

AMD can’t exactly at the snap of a finger from the beginning of production already have a mass supply of processors available for sale, that’s not exactly how any of this works.

So those who are excited to see the revolution of 3D V-Cache Technology well before the first quarter of next year, you might want to come to the same realization that Intel fanboys have had for the last 24 months, and get used to disappointment.

For those like myself who actually intended to skip Zen 4 granted that it would be on an entirely new platform and socket along with the launch of what’s to be horrible DDR5 memory, honestly buying the first of anything doesn’t seem like a solid idea if you’re going for longevity.

There will always be a place in my heart for Ryzen 1000 series owners and their 300-series motherboards, but I guess I’m just going to have to throw my plans out of the window now that AMD have resurrected the 5775C.

blog comments powered by Disqus