By: Aaron Spink (aaronspink.delete@this.notearthlink.net), April 18, 2019 1:32 pm
Room: Moderated Discussions
anon (spam.delete.delete.delete@this.this.this.spam.com) on April 18, 2019 12:07 pm wrote:
> Yes, we all agree that any NAND will get stupidly high sequential
> read speeds, higher than the controller and interfae can handle.
> This does not make the downsides of QLC disappear.
>
For this application, I'm not convinced that QLC downsides are at all relevant.
> Are you sure they'll be able to get 1 TB QLC for 20$ a year or so before launch? Judging by current
> prices something would have to go horribly wrong for NAND manufacturers to offer such a deal.
>
On a multi-year capacity contract, its possible. Manufacturers are close to transitioning to smaller processes and higher stack counts which will push pricing down. Current market pricing would be ~$35-40 which should drop with the next node and stack size.
> I would've said 250-500 GB NAND + 4 TB HDD. Do the math with that and I've also used
> higher prices for QLC. I'll believe in 20$ per TB when I see it. If your premise is
> that the price of NAND will be lower than any HDD of the same capacity then the workload
> doesn't even factor into it anymore. I'm just not willing to buy your premise.
>
HDD only get a cost advantage when they are pushing max capacity. 14 TB+ drives will have a cost advantage but that advantage doesn't scale down to realistic sizes for a console. When you start getting into 4TB drives, you are pretty much already at or below the cost floor for HDDs. And QLC/OLC are coming on pretty fast with significant ramifications.
> Also datacenter workloads are quite different. You'd expect all data to be accessed. Unless someone
> comes up with a >500 GB game and I really doubt that the working set will never exceed that size and
> a 500 GB cache will work quite well. Realistically I'd expect much less. Apart from a few outliers it's
> mostly around 50 GB now, so 100-150, maybe 200 for a few sound reasonable. That's 3-5 games fully cached
> on 500 GB. Seems good enough to me. Again, that's quite different from a datacenter workload.
>
I'm talking about consumer workloads with caching, not datacenter workloads. It does a pretty bad job.
And there are multiple games in the ~100GB range at the moment with many more coming. I just don't see them doing less than 1TB of flash storage nor do I foresee them spending the complexity or resources for caching. If they support HDDs at all, it will be over USB.
> Yes, we all agree that any NAND will get stupidly high sequential
> read speeds, higher than the controller and interfae can handle.
> This does not make the downsides of QLC disappear.
>
For this application, I'm not convinced that QLC downsides are at all relevant.
> Are you sure they'll be able to get 1 TB QLC for 20$ a year or so before launch? Judging by current
> prices something would have to go horribly wrong for NAND manufacturers to offer such a deal.
>
On a multi-year capacity contract, its possible. Manufacturers are close to transitioning to smaller processes and higher stack counts which will push pricing down. Current market pricing would be ~$35-40 which should drop with the next node and stack size.
> I would've said 250-500 GB NAND + 4 TB HDD. Do the math with that and I've also used
> higher prices for QLC. I'll believe in 20$ per TB when I see it. If your premise is
> that the price of NAND will be lower than any HDD of the same capacity then the workload
> doesn't even factor into it anymore. I'm just not willing to buy your premise.
>
HDD only get a cost advantage when they are pushing max capacity. 14 TB+ drives will have a cost advantage but that advantage doesn't scale down to realistic sizes for a console. When you start getting into 4TB drives, you are pretty much already at or below the cost floor for HDDs. And QLC/OLC are coming on pretty fast with significant ramifications.
> Also datacenter workloads are quite different. You'd expect all data to be accessed. Unless someone
> comes up with a >500 GB game and I really doubt that the working set will never exceed that size and
> a 500 GB cache will work quite well. Realistically I'd expect much less. Apart from a few outliers it's
> mostly around 50 GB now, so 100-150, maybe 200 for a few sound reasonable. That's 3-5 games fully cached
> on 500 GB. Seems good enough to me. Again, that's quite different from a datacenter workload.
>
I'm talking about consumer workloads with caching, not datacenter workloads. It does a pretty bad job.
And there are multiple games in the ~100GB range at the moment with many more coming. I just don't see them doing less than 1TB of flash storage nor do I foresee them spending the complexity or resources for caching. If they support HDDs at all, it will be over USB.