AMD Ryzen 7 5800X3D owners have posted benchmarks that show off what the 3D V-Cache of these CPUs are capable of when operating as a RAM disk.
Per benchmarks shared by the @GPUsAreMagic account, users can see ludicrous speeds of nearly 183,000 MB/s read and over 175,000 MB/s write from AMD’s 3D V-Cache CPUs, although this trick seems to work under CrystalDiskMark only.
CrystalCacheMark pic.twitter.com/MSGyrGkXv8
— Nemez (@GPUsAreMagic) February 13, 2023
CrystalDiskMark with very specific settings running from a RAMDisk (OSFMount) all within the 96MB L3 cache of the 5800X3D.
— Nemez (@GPUsAreMagic) February 13, 2023
Here’s how owners of the Ryzen 7 5800X3D and other 3D V-Cache chips can gawk at these huge numbers:
- Install OSFMount and create a FAT32 formatted ramdisk.
- Run CrystalDiskMark, set the test values to:
- SEQ 256KB, Queue Depth 1, Threads 16
- Set data fill to 0s instead of random.
- Run the test (probably a few times, this is very sensitive to system load).
According to additional benchmarks from other testers, the speeds may even go up to 207,073 MB/s read and 179,309 MB/s write.
“Innovative 3D stacking technology to enable the world’s fastest gaming processors – the AMD Ryzen 7000X3D series – and the world’s highest-performing server processors for technical computing,” reads the marketing for AMD 3D V-Cache technology.


Discussion (5 replies)
Join Discussion →Who the fudge wants a 96mb ramdisk? Really what's the point? I suppose if you're doing some kind of 'disk based' mining that doesn't use capacity but needs raw throughput this would be insane.
greeeat. a run on 3dvcache CPU's now.... sigh...
Doom 1 at 6 trillion fps?
That's cool and all.
Now make it 1TB in size :p
All joking aside, I bet there is an application for the EPYC X variants and their 768MB of L3 cache.
I'm not quite sure what it is yet, but I bet there is one.
I'm just waiting on some hypervisor that lets me make a san disk out of my 3dvcache across several CPU's/systems.
Get a GIG of 3d vcache system cache... sure your CPU's will bottleneck LIKE NO TOMORROW... but you'll have one REALLY fast drive! Two CPU's one all cache!
Maybe you can mine Chia on it? :p
I suspect the reason why CPU's like - for example - the 5800x3d are so much faster than the 5800x despite using the same arch and being clocked 200-400Mhz lower (depending on boost) has everything to do with better utilization of the cores. The cores themselves may be slightly slower due to the lower clock, but they spend much less time idle waiting for data from the relatively slow RAM.
So, will the CPU bottleneck in accessing the L3 cache? Maybe, but the purpose for all that juicy L3 cache in the first place was to eliminate wasteful CPU wait states where the CPU couldn't be fully utilized.