I don't think we will see 32GB of memory..
Middle Generation Console Upgrade Discussion [Scorpio, 4Pro]
- Status
64 GB for "true" next-gen please, even if I have to wait 4 more years with it that's OK
16x RAM last-gen became 8 GB, so I hope for at least 8x again, but personally I'm guessing 32 GB for PS5/XboxTwo
Can't go by previous gens leaps anymore. I'd say it's going to be 12GB/16GB for Scorpio. Rumors have the PS4 neo still at 8GB, the average for gpu's now is 8GB. So I think Microsoft will target memory so there is 8-12Gb available for games...which probably means 16GB overall.
AMD has stated at various points that its up to the semi-custom division's customer to decide when their product is to be announced or discussed.
as others have said the intel nuc is 15watts and it has no fan. You can include a fan and up the power consumption a bit.An AppleTV is a <2Watt device. Use some common sense here
The xbox one was about 110w while playing a game with the Kinect and drive going. So they might be able to do something simlar
rockaman
Regular
I mean for the real next-gen, not this fake "next-gen" ^^
Rikimaru
Veteran
Why most people do not think about cost of such specs.
32gigs might be out of the question but we could see a 16 gig or even a 24gig set up. 8 gigs for the cpu of ddr 4 and 16 gigs for the GPU in the form of HBM2 .
The ps4/xbox one has access to 4-5 gigs. So even giving the gpu 8 gigs to work with would be a major step up and giving it 8 gigs of HBM 2 would be great. Throw in 8 gigs for the CPU of ddr 4 and you would have a big leap over todays consoles
we've seen split memory before in consoles. .
Spilt memory tends to be the cheaper option, hence it getting used time and again in devices with them having multiple RAM pools. Split memory has the advantage of no contention.
D
Deleted member 11852
Guest
I think split memory pools are only really cheaper in fringe cases, such as where you need a small amount of very [expensive] fast RAM and a large amount of [cheaper] slow RAM but where the overall RAM capability makes the costs and engineering and manufacturing a dual-bus system cheaper. Split memory pools generally mean a more complex board and this is a cost that you live with for the lifetime of a system and which is difficult to cost reduce. Components get cheaper but board manufacture generally doesn't by large margins.
I was thinking of any splut memory, but I suppose you mean main storage and not working RAM. Certainly very few devices focussed on performance have operated from a single pool of memory. PC has split pool, PS3 did, XB360 did (eDRAM pool), XB1 does, PS2 did, GC did, etc. A single pool of RAM providing adeqaute storage and bandwidth is hard to clearly hard to come by. Butif one doesn't mean working RAM, then yes, a single pool of memory with one RAM type is the preferred option.
I guess we never really determined the number of savings in terms of CPU cycle times for having to just do pointer direction vs a copy. I'm assuming most of it is one way unless we are talking about GPGPU. Does it out weigh bandwidth contention issues?
But doesn't single pool suffer from having to rely on HDD streaming? Unless you fit the whole level into what's remaining, then perhaps you don't have to worry about it.
Sent from my iPhone using Tapatalk
But doesn't single pool suffer from having to rely on HDD streaming? Unless you fit the whole level into what's remaining, then perhaps you don't have to worry about it.
Sent from my iPhone using Tapatalk
HDD streaming is an issue no matter what RAM topology you have if you don't have enough capacity to fit the whole level. And split pool tends to increase RAM requirement per level for duplicate data. eg. Load assets into RAM on PC and copy across to VRAM for GPU, or load assets into RAM on XB1 and copy across to eSRAM as needed. The ideal solution is crazy amounts of RAM at crazy BW and low latency that it doesn't matter about contention or limits, but that's not possible so we have to chop our storage up into more specific quantities for more specific solutions. I think the latest Fallout DLC shows the problem PS4 has with unified RAM in drawing those fog particles (although we don't really look to Bethesda to provide optimised solutions for their engine...).
D
Deleted member 11852
Guest
I do mean working RAM. Few consumer devices use split memory because these are not engineered primarily for performance but are balanced to hit a cost target. PS3 used split memory not because it was preferable but because there was no choice. Cell couldn't work with GDDR3 and RSX couldn't work with XDR. And don't forget that many PCs, for quite a few years, come with an integrated GPU which uses a segment of RAM for the graphics buffer. Thanks to the explosion of laptops these types of system probably outnumber traditional PCs which have DDR for the CPU and GDDR for the GPU. But even the ones that aren't are geared for performance and because the GPU sits on the wrong side of the slow ass PCI local bus. PCI Express rev.4 has a mere 31.5Gb/sec bandwidth in x16 mode, which is ten times slower than the memory bandwidth of a Geforce 980Ti with GDDR5 and a quarter of a mid-range Geforce 960. PCIe rev.4 doesn't exist in consumer space so everybody is still using PCI rev.3 which is half as fast!
Adding a second bus, wiring additional lanes, extra clocks to support the traditional north/south bridge architecture has never been cheaper - all things being equal.
In terms of streaming and having data onhand for the GPU and not being choked by the PCI bus, there is a better solution which is for the GPU to have two pools of RAM, high-speed GDDR for actual GPU operations and a pool of slower DDR as a buffer so that the PCI bus is never a bottleneck. GPU's are needing increasing amounts of GDDR5/X not because the GPU needs that frame to frame but because games are putting a lot of read-ahead buffered data in there as well.
The PCI bus really does need a kick up the arse.
Last edited by a moderator:
- Status
