TR Forums

So this is a PCIe 4x card that has one M2 slot for Optane, one M2 slot for NVMe SSD and a plain jane SATA 3 port. The Optane caches MFU data from the NVMe SSD and the NVMe SSD caches the MFU data from the Sata storage device (SSD or large capacity HDD). The card is OS-agnostic and just works, without requiring any software or driver installation and just presents the Sata storage device to the OS. This way, it can work with either Intel or AMD PCs or even Macs. I would definitely make this if I had the funding but since I don't, I invite any enterprising engineer to take my idea and change the world. Forever.

Sounds like a really expensive solution in search of a problem to me.

just brew it! wrote:

Sounds like a really expensive solution in search of a problem to me.

What is so expensive in it? Just two M2 slots and a SATA port and a controller chip or even an FPGA to do the caching part. A company like Asus could have it on the market in less than 6 months. The problem it will solve is speeding up hard disk accesses, especially for 4TB or larger game drives. It could even be used to speed up databases for small businesses that can't afford 512GB RAM. This way, they could make do with 128GB RAM and let the rest be cached on the accelerator card.

Plus the expense of an Optane device, an NVMe device, and a SSD?

And all the extra overhead of moving stuff around will slow things down. Assuming you get the caching algorithms exactly right maybe you'll see some marginal gains? It could even end up being a net loss.

It sounds like the kind of thing someone comes up with believing it'll make things easier when all it really does is add additional complexity that serves no real purpose.

I'm sure I'm not the only one here who has had to clean up code like that before just to be left wonder what on Earth they were thinking.

Be positive, guys. What if your boss came up with that idea? You don't wanna risk getting fired for being pessimistic now, do you?

You'd rather waste their time and resources doing something you know is bound for failure than to speak up at the start, I guess?

We're not talking about self driving cars, of some other field with lots of open problems here. It's a problem of basic computer architecture, and that hasn't changed much in quite some time.

Redocbew wrote:

It's a problem of basic computer architecture, and that hasn't changed much in quite some time.

Does it do any good acting like a 20th century physicist telling yourself that there is no more physics left to do?

There's plenty left to do. The field doesn't suck. It's just your idea that sucks.

My idea for caching data sucks yet there are three levels of cache in a modern multicore CPU and significant die space is dedicated to that. Try getting 16GB or 32GB of data faster to the CPU at Optane speeds and it's a bad idea? Whatever happened to removing bottlenecks?

Igor_Kavinski wrote:

Be positive, guys. What if your boss came up with that idea? You don't wanna risk getting fired for being pessimistic now, do you?

I'd probably nod and smile, then start looking for another job, because it would be obvious I'm working for an idiot.

just brew it! wrote:

Igor_Kavinski wrote:

Be positive, guys. What if your boss came up with that idea? You don't wanna risk getting fired for being pessimistic now, do you?

I'd probably nod and smile, then start looking for another job, because it would be obvious I'm working for an idiot.

Even if the idiot paid you a **** ton of money to do his idiotic project?

Igor_Kavinski wrote:

My idea for caching data sucks yet there are three levels of cache in a modern multicore CPU and significant die space is dedicated to that. Try getting 16GB or 32GB of data faster to the CPU at Optane speeds and it's a bad idea? Whatever happened to removing bottlenecks?

The fact that you're talking about "optane speeds" as if it's better than just loading everything into memory means you've bought into the marketing and aren't thinking about this technically, so yes that sucks.

Redocbew wrote:

Igor_Kavinski wrote:

My idea for caching data sucks yet there are three levels of cache in a modern multicore CPU and significant die space is dedicated to that. Try getting 16GB or 32GB of data faster to the CPU at Optane speeds and it's a bad idea? Whatever happened to removing bottlenecks?

The fact that you're talking about "optane speeds" as if it's better than just loading everything into memory means you've bought into the marketing and aren't thinking about this technically, so yes that sucks.

You can have 118GB Optane in the M2 slot for less than $200. 128GB RAM would cost a lot more, if it is even supported by the mobo.

M2 slots operate over PCIe which has less bandwidth than main memory. They're not interchangeable, and regardless of what you call it once you start working with storage through a filesystem, then you're adding an additional layer of complexity which is also going to slow things down.

Cost, or speed. Pick one.

Igor_Kavinski wrote:

just brew it! wrote:

Igor_Kavinski wrote:

I'd probably nod and smile, then start looking for another job, because it would be obvious I'm working for an idiot.

Even if the idiot paid you a **** ton of money to do his idiotic project?

A company with management that dumb probably won't be around long.

just brew it! wrote:

Igor_Kavinski wrote:

just brew it! wrote:

Even if the idiot paid you a **** ton of money to do his idiotic project?

A company with management that dumb probably won't be around long.

You evaded the question. Will you take the money? It's not like he would sue you if his impractical idea didn't work.

Igor_Kavinski wrote:

just brew it! wrote:

Igor_Kavinski wrote:

A company with management that dumb probably won't be around long.

You evaded the question. Will you take the money? It's not like he would sue you if his impractical idea didn't work.

Sure, I'd take the money in the meantime. But I would also be looking for a viable exit strategy. Getting paid big bucks in the short term is all for naught if you end up out of work when the project craters; and the stress of not knowing where your next paycheck will come from (or when...) means you can't enjoy the time off.

What is the point of this tangent anyway? It is nearly as silly as the original concept.

just brew it! wrote:

What is the point of this tangent anyway? It is nearly as silly as the original concept.

I just want to see this happen. There are much more niche gadgets/devices in the world. One more won't hurt. How about a kickstarter? Would you give it your all if people believed in you?

Igor_Kavinski wrote:

Would you give it your all if people believed in you?

People can believe in me all they want. If I think it is a stupid idea it is going to be impossible to get enthusiastic about it.

The only people who are going to be enthusiastic will be the people who don't realize it's a bad idea.

https://www.amazon.ae/AHCIM-2-Adapter-C ... 07T4P4GBY/

Looks like someone already implemented it, minus the caching part.

Igor_Kavinski wrote:

https://www.amazon.ae/AHCIM-2-Adapter-Converter-Bracket%EF%BC%88SSD-included%EF%BC%89/dp/B07T4P4GBY/

Looks like someone already implemented it, minus the caching part.

Getting the caching (firmware/drivers) right is the hard part. And even if you do that, I doubt it will provide a meaningful boost for your envisioned use case.

That thing is just an adapter, not an "accelerator".

Gaming performance is still mostly bounded by GPU and CPU. I/O is still not a tier-1 factor. Even if the idea works you are still not optimizing the right area.

Flying Fox wrote:

Gaming performance is still mostly bounded by GPU and CPU. I/O is still not a tier-1 factor.

Isn't that exactly what Microsoft and Sony optimized in their new consoles?

Did you just invent a caching raid controller?

I have an idea lets put a lot of ram on the card too, that's even faster than the other memory types. Also lets expand the interface to PCIe 16x.

Igor_Kavinski wrote:

Flying Fox wrote:

Gaming performance is still mostly bounded by GPU and CPU. I/O is still not a tier-1 factor.

Isn't that exactly what Microsoft and Sony optimized in their new consoles?

You mean improve load times but getting better hardware? Did you know you can optimize a PS4 or Xbox One by putting in an SSD?

tfp wrote:

Igor_Kavinski wrote:

Flying Fox wrote:

Gaming performance is still mostly bounded by GPU and CPU. I/O is still not a tier-1 factor.

Isn't that exactly what Microsoft and Sony optimized in their new consoles?

You mean improve load times but getting better hardware? Did you know you can optimize a PS4 or Xbox One by putting in an SSD?

I/O performance improvement in the PS4 and XB1 is limited by the SATA interface.

tfp wrote:

Did you just invent a caching raid controller?

I have an idea lets put a lot of ram on the card too, that's even faster than the other memory types. Also lets expand the interface to PCIe 16x.

My idea has nothing to do with RAID. But yeah, now that you mention it, there is a possible market for a PCIe card that accepts different types of old server RAM that can be bought for pretty good prices. Would be a nice way to cache boost your PC's performance even better and cheaper than Optane.

Igor_Kavinski wrote:

tfp wrote:

Igor_Kavinski wrote:

Isn't that exactly what Microsoft and Sony optimized in their new consoles?

You mean improve load times but getting better hardware? Did you know you can optimize a PS4 or Xbox One by putting in an SSD?

I/O performance improvement in the PS4 and XB1 is limited by the SATA interface.

They are also limited by a spiny drive. Load times of PC games on a SATA interface are significantly improved by moving to SSD from a standard HD. For large loads an SSD will be able to fill the bus more fully and longer than a standard HD.

Are there fast interfaces sure but it requires a new MB or if they had PCIe slots a new card and new drives. The changes they are doing are not a revolution for disk access and are no different then building an pretty standard desktop PC today.

Igor_Kavinski wrote:

tfp wrote:

Did you just invent a caching raid controller?

I have an idea lets put a lot of ram on the card too, that's even faster than the other memory types. Also lets expand the interface to PCIe 16x.

My idea has nothing to do with RAID. But yeah, now that you mention it, there is a possible market for a PCIe card that accepts different types of old server RAM that can be bought for pretty good prices. Would be a nice way to cache boost your PC's performance even better and cheaper than Optane.

I'm talking about the hardware on a RAID card. It pretty much does what you are talking about but with ram. Ram has power failure issues but that's what batteries are for. Maybe they will use Optane or some sort of fast persistent ram as back up for power failure for ram in the future. However with SSDs and their speed I just haven't paid attention to this kind of setup because the cost doesn't justify the performance unless you are in a server environment.

If I remember right, you can just stick one drive on a card and use the cards caching but it's not really all that useful. In that case just get a harddrive with more built in ram for caching. It's been a while so I don't know how much ram RAID cards are pushing now.