| Intrinsic said: I really believe that by 2020, sony/ms would be able to order M.2 SATA 3 based 1TB SSDs for under $50. |
Once large SSD's hit that price point, then all bets are off.
We are still awhile away from that point though.
... And NAND manufacturers switching over to GDDR6/DDR5 is going to slow that down.
Intrinsic said:
Hehe.... nope I didnt't. And you are actually saying exactly what I was implying too. Yes, you only would need 4 chips for 8GB of DDR4, and that is my point exactly. This is why I keep saying LPDDR4 (as used in phones), cause for that you will only need one chip or should I say one "package" that is the size of one chip of GDDR6 ram. Now if you are looking at a system where you need to have 16GB of DDR4 coupled with say 16GB of GGDR6 then you will need to have a PCB thats big (and complex) enough to accommodate 16 total chips flanking the APU. And this whole thing will be ridiculously expensive too. But using LPDDR4 (like the ones used in phones), you would need only one chip to get 8GB of ram, 2 for 16GB, leaving a lot of room for the GDDR6 chips. And LPDDR4 ram is significantly cheaper too. |
PCB size isn't the issue.
It's the number of traces and thus layers that becomes problematic... And that increases exponentially with the amount of chips you add.
Some DRAM chips need less traces than others as well... So sometimes when you do have twice the number of chips, you don't need as many traces as thus layers.
But it all comes down to price in the end, if they can get high-capacity LPDDR4 cheaper than regular DDR4, then that is probably what they will opt for.
| Intrinsic said: Unless I am mistaken, no... not with LPDDR4. To my understanding its peaking out at 32GB/s. That however may be due to a smaller bus used in mobile phones and could be different in a console if they use a larger bus but I don't know about that so I can't say anything since I couldn't find anywhere that it was used like that. But theoretically I don't see why its not possible. |
You certainly are mistaken. We just haven't seen LPDDR4 implemented in such a way that pushes past 32GB/s of aggregate bandwidth.
It can most certainly go higher.
The more chips you add, the wider the bus you implement... The more bandwidth you have at your disposal.
| Intrinsic said: Which coupled with what your confirmation that any Ram technology can benefit form bandwith increases when paired with other similar ram in a unified bus means that its even better for my theory. Sony/MS could go with two 8GB LPDDR4 chips to get 16GB of OS and CPU Ram with a bandwith of at least 60GB/s or higher if the whole wider bus for LPDDR4 is possible. Meaning they could lock out 8GB of this ram just for the OS and make 8GB available for games coupled with what will now become the at least 8-16GB of dedicated GDDR6 vram. |
It will be interesting to see which way they take things. I doubt Microsoft will go for split pools anyway, they haven't actually done this in a console yet, where-as Sony has done it for the last few generations.
There are Pro's and Con's to it as well, when Microsoft/Sony optimize their software stacks and free up more memory, it's more beneficial to a developer if it's high-speed GDDR6 than DDR4/LPDDR4.
Split pools also means you take a small bandwidth/latency hit when communicating/shuffling data between them as well.
www.youtube.com/@Pemalite
