AMD 3D V-Cache RAM Disk Delivers Over 182 GB/s and 175 GB/s Read and Write Speeds

[AleksandarK]

AMD's 3D V-Cache technology utilizes blocks of SRAM stacked on top of the CPU logic die, where CPU cores reside, and allows the processor to access massive pools of cache for applications. However, using this extra level 3 (L3) cache as a RAM disk appears possible, where the L3 SRAM behaves similarly to a storage drive. A big disclaimer here is that this is only possible by exposing the L3 to the CrystalDiskMark benchmark, and no real-world applications can do it in a way that CrystalDiskMark. According to X/Twitter user Nemez (@GPUsAreMagic), the steps to replicate this procedure are: Obtaining an AMD Ryzen CPU with 3D V-Cache, installing OSFMount and creating a FAT32 formatted RAM disk, and running CrystalDiskMark, with values set to values to SEQ 256 KB, Queue Depth 1, Threads 16, and data fill to 0s instead of random.

The results of this experiment? Well, they appear to be rather stunning as the nature of L3 SRAM is that the memory is tiny but very fast and accessible to the CPU, so it can help load data locally before going to the system RAM. With AMD Ryzen 7 5800X3D, the speeds of this RAM disk are over 182 GB/s for reading and over 175 GB/s for writing. In another test, shared by Albert Thomas (@ultrawide219), we managed to see RAM disk based on AMD Ryzen 7800X3D V-Cache, which scores a little less with over 178 GB/s read and over 163 GB/s write speeds. Again, CrystalDiskMark only performed these tests on small allocations varying between 16 MiB and 32 MiB, so no real-world workloads are yet able to utilize this.



View at TechPowerUp Main Site | Source

 

[Space Lynx]

nvme drives are dead. give me 256gb of ram baby!

and a UPS battery backup

 

[A Computer Guy]

That's a shame applications can't access the cache directly and partition off a segment. Not necessarily as a ram disk but it would be nice to allocate and use a chunk of very high speed cache directly for drivers and application acceleration.

 

[Crackong]

3DV EPYC gives you about 768MB of L3 sooooo someone could run windows xp on it

 

[Vya Domus]

That's a shame applications can't access the cache directly and partition off a segment. Not necessarily as a ram disk but it would be nice to allocate a chunk of very high speed cache.

There is no real point in doing that, if you write software that is cache aware it doesn't really make a difference because it will take advantage of it no matter how the cache is configured.

 

[AnotherReader]

While interesting, it's rather likely to be detrimental to most programs. The CPU will make better use of the cache than a user using it as a tiny RAM disk.

 

[A Computer Guy]

3DV EPYC gives you about 768MB of L3 sooooo someone could run windows xp on it

Humm... lets see who can get the fastest BSOD!

 

[Prima.Vera]

4Ks are relatively "slow" ...

 

[Assimilator]

*slow clap*

 

[A Computer Guy]

There is no real point in doing that, if you write software that is cache aware it doesn't really make a difference because it will take advantage of it no matter how the cache is configured.

Extremely high speed memory would be just another tool in the tool box. I'm sure creative minds would be able to optimize some routines to insane levels with that kind of access.

 

[Airisom]

Hmmm maybe replacing a chiplet with L4. 8800X4D hehe

 

[Fouquin]

3DV EPYC gives you about 768MB of L3 sooooo someone could run windows xp on it

With the way XP handles cache residency for stuff like audio processing it'll end up flushing itself into oblivion on accident. The crushing weight of hardware accelerated audio.

 

[DemonicRyzen666]

good way to build a new benchmark to test the increases in generation of 3D-v cache improvements. especially if Intel will have some L4 cache or something similar to 3D-V cache later on.

 

[Denver]

Damn... When I read the title, I was shocked and thought, "What product is AMD making now? A new type of SSD? What a bully wanting to succeed in all the areas that intel failed." But it's just a gimmick.

 

[Kapone33]

This needs to be further explored those RND4K speeds are faster then Optane.

 

[AnotherReader]

This needs to be further explored those RND4K speeds are faster then Optane.

That 4K result is rather low considering it's SRAM; it's probably an artifact of the way the benchmark is written. Even DRAM would have much higher figures than Optane. DRAM and SRAM are very different from non volatile storage. Both have unlimited endurance.

 

[ExcuseMeWtf]

That's cool, but wake me up when they go into hundreds of GBs at least.

 

[Jism]

That's a shame applications can't access the cache directly and partition off a segment. Not necessarily as a ram disk but it would be nice to allocate and use a chunk of very high speed cache directly for drivers and application acceleration.

Did you know drivers get loaded upon boot of Windows? Its a one time thing.

Windows would generally be better, by simply taxing UNUSED ram as cache, as linux does. What good use is there if you have 20GB of unused ram in your system (out of 32GB) ?

Cache on chips is very very expensive. I used primocache for a longer time on my workstation or older notebook. The difference is pretty much huge esp when you have cache hits.

On older computers it works fantastic - pretty much boosting things up significant. On fast or large systems - it's maybe 20% better.

 

[AnotherReader]

Did you know drivers get loaded upon boot of Windows? Its a one time thing.

Windows would generally be better, by simply taxing UNUSED ram as cache, as linux does. What good use is there if you have 20GB of unused ram in your system (out of 32GB) ?

Cache on chips is very very expensive. I used primocache for a longer time on my workstation or older notebook. The difference is pretty much huge esp when you have cache hits.

On older computers it works fantastic - pretty much boosting things up significant. On fast or large systems - it's maybe 20% better.

Windows already uses free RAM as a cache.

 

[FoulOnWhite]

Interesting but that's all.

 

[JasBC]

While interesting, it's rather likely to be detrimental to most programs. The CPU will make better use of the cache than a user using it as a tiny RAM disk.

Well. . . the XBONE kinda had a cache-drive like this that developers utilised all the time it seems to try and mitigate the weaker hardware as compared to the PS4. It's not like these X3D-CPUs are low-end like the XBONE and PS4 were though, they aren't like the mobile CPUs and outdated mid-range GPUs that made up those systems SoCs.

 

[AnotherReader]

Well. . . the XBONE kinda had a cache-drive like this that developers utilised all the time it seems to try and mitigate the weaker hardware as compared to the PS4. It's not like these X3D-CPUs are low-end like the XBONE and PS4 were though, they aren't like the mobile CPUs and outdated mid-range GPUs that made up those systems SoCs.

Yes that was an attempt to mitigate the lower bandwidth of the Xbox One relative to the PS4. While both had 256-bit buses to RAM, PS4 used 5.5 Gbps GDDR5 while the Xbox had to make do with DDR3 at 2133 MT/s. The X3D CPUs don't need any help from programmers as they are more than capable of utilizing the L3 as a normal cache.

 

[Hardware1906]

However, using this extra level 3 (L3) cache as a RAM disk appears possible, where the L3 SRAM behaves similarly to a storage drive

so, the speed of these L3 cache is just fast as RAM
because right now, with quad channel 6000MHz DDR5, you can already reach 192GB/s

 

[Wirko]

The results are close to those obtained by Chips and Cheese, they measured 154 GB/s L3 read bandwidth for the 7950X (not X3D but it doesn't matter). The 12900K reached 105 GB/s.

AMD’s Zen 4, Part 2: Memory Subsystem and Conclusion

Please go through part 1 of our Zen 4 coverage, if you haven’t done so already.

chipsandcheese.com

 

[Jism]

Windows already uses free RAM as a cache.

I think not efficient enough.

Dedicated tools work alot better in my opinion.