Thats not a 64GB/s (or even close) PCIe16 SSD. Thats a standard M2
No it's not, your right. I'm not saying it is.... I'm saying that is the closest thing you would get in the consumer-grade market.
Because they switched to PCI-E 5.0 before any other hardware was ready for it.
No. You can get some x8 AICs for enterprise, like the Gen4 PM1735, but they don't push the full bandwidth that they could. Most of the time they'll be x4 or x4/x4 (dual-port for redundancy) and may use other form factors like ruler (E1.S/E1.L) which will also be x4 or x8 (or x4/x4). E.3 (E.3S/E.3L) can do up to x16 but usually SSDs will be x4 or x8 (with the possibility for dual-port). In the past there have been some wonky/giant AIC SSDs (e.g. HHHL/FHHL ioDrive which could have multiple PCBs) but at the bandwidth levels you're talking, not that I'm aware of, no.
I want to do a new build. My last build was in 2019, and even by todays standards, its still better than 90% of PCs on the market today. That's 4 years. It should be ancient, but it isn't. i guess its a testament to how much slower technology is evolving compared to 20 years ago. I want to do a new build, but I want the specs jump to be significant, or theres no point. I'm hoping the Ryzen 9950X will be 24+ cores instead of 16. I also want a big jump in SSD performance. Right now ,the fastest i have seen is 12GB/s on an x4. I dont know why we cant get 48 on an x16 on one single PCB (not a PCB that allows 4 x4 NVMEs). I mean a single PCB in an x16 slot that can do 48GB/s, that is one single storage device, non-RAID
We are deep into diminishing returns on drive speed, and have been for years.
Ya. I understand what you're saying. But I feel like jumping from 5GB/s to 64GB/s on PCIe 6.0 would make.a big difference again. Maybe not. I don't know
A 10x boost from SATA III to PCIe 4.0 is nearly unnoticeable.
Only in benchmarks and bragging rights.
Big difference in what, exactly?
Ok, but why such diminishing returns? I am very confused. Maybe I need an ELI5 on this one. Help?
Lets put some numbers to this then
Per generation, we've had a 2x increase in performance with Sequential data.
Ok, so if the returns now are so bad, whats the point? Why do we keep manufacturing faster SSDs when its clearly not doing anything?Per generation, we've had a 2x increase in performance with Sequential data.
THis would be if you were copying a big fat 10GB movie from drive A to drive B.
The small 4k data blocks, where the vast majority of our data lives, does not see that.
SSDs major benefit is the near zero access time. This is across ALL SSD types, even the lowly SATA III drives.
1. The rest of the system comes into play. It is NOT just raw sequential drive speed. CPU, RAM, software, etc.
2. That bog number you see is strictly for marketing purposes. Big numbers sell.
Here are two of my drives, in CrystalDiskMark. PCIe 4.0 980 Pro, and SATA III 860 EVO.
Both 1TB.
The numbers you see advertised in in reviews is the top line.
Big difference, right? The 980 Pro is 10x that of the 860.
But that is only the Sequential speed.
The numbers that really count with the vast majority of our data is the two bottom lines. 4k.
Not much difference, is there?
A few months ago, I did a rendering test with Corels VideoStudio.
Rendering video of 1 min, 5 min, 10 min, out to 3 different drives.
The 980 Pro, Intel 660p, and any of my SATA III SSDs.
Zero difference.
(I should redo this and publish the graphs)
The write speed of the drive made NO difference....the rest of the system has FAR FAR more influence over what happens.
And as said above....you can absolutely tell the difference between 10 seconds and 5 seconds.
You can't tell, outside of artificial benchmarks, the difference between 0.1 sec and 0.05 sec.
Even though it is the same 2x difference.
1. New 'standards' keep appearing. PCIe 3.0, 4.0, 5.0, X.0
tl;dr - you're bottlenecked by software. The biggest gains over HDDs were with random latency and that improves slowly for NAND, plus at a certain point you're bottlenecked by the software being unable to make use of more performance. SSDs work through parallelization which requires queue depth (outstanding I/O requests) but most of your daily apps are QD1 and almost all are QD4 at the most. SSDs are just way too fast and the software isn't optimized for it. At least on Windows and until the DirectStorage API is expanded. Bandwidth is related since BW = 1/latency * size which is more realizable with large I/O but who is transferring around 100GB archives every day?
Facebook
Twitter
Instagram