Predict: The Next Generation Console Tech

Status
Not open for further replies.
Aren't most PC games 32-bit apps? If they are, then isn't that what's limiting memory usage of those games?

For the same reason, I don't think it's very realistic of Proelite to expect a 32-bit CPU in the next PS, even if it has 'only' 4 GB or RAM.

The original leak on Pastebin had the Orbis cpu at 32-bit, and some comments made by gaffers with connections seem to indicate such.

I am very confident that it's the case.
 
The original leak on Pastebin had the Orbis cpu at 32-bit, and some comments made by gaffers with connections seem to indicate such.

I am very confident that it's the case.

Why go out of your way to have a custom 32 bit core when you could have a less custom 64 bit core? Bus widths? Die area? TDP?
 
Yea, that pastebin rumor did have too much "rights" to be wrong. If you post on pastebin as a guest you can't edit it so it pretty much confirms he had "Durango" and "Orbis" codenames first (4-5 months earlier actually). He also mentioned 8 core CPU in Durango, but rated at 1.2TF (although that HAS to be APU), but he also mentioned GPU so its slightly bogus.

PS4 he also got right pretty much, 4 core CPU and 2GB.

On the second thought, he is probably not "jobless student" who got in possession of target specs, maybe developer.
 
You can use 32bit pointers on a 64bit CPU, forfeiting the big address space but gaining performance such as less space taken by pointers in L1.
This exists under Linux as "x32"
http://en.wikipedia.org/wiki/X32_ABI

So if the PS4 has a low amount of memory such as 2GB, or 3GB gddr5 on 192bit (that configuration could be pretty decent), using that mode would make sense ; it's like a free performance increase.

With 4GB memory it's not as good as you either need to have a bit of useless memory (like XP's 3.25GB problem or the 640KB problem) or do bank switching.
 
No need to limit the memory to 4GB just because it's 32bit, linux can easily do up to 64GB on the 32bit kernel with PAE. But I love the idea, L1 cache could possibly be more efficient. Is there a statistic somewhere about the percentage amount of space being taken by addresses versus opcode in the instruction L1 cache? For a typical game? That would give a fair idea if staying in 32 bit is worthwhile. You get that much more from the instruction cache, can't be a bad thing if it comes almost for free.
 
You need address space for I/O to the devices (such as storage, networking, chipset functions), I would have said for talking to the graphics card but hopefully there's single address space for CPU and GPU.

Virtual memory?, it could well be used on consoles. in good old times, consoles didn't run an OS but modern consoles do and in a usual OS, processes have virtual memory. But that maybe is a separate issue.

No need to limit the memory to 4GB just because it's 32bit, linux can easily do up to 64GB on the 32bit kernel with PAE. But I love the idea, L1 cache could possibly be more efficient. Is there a statistic somewhere about the percentage amount of space being taken by addresses versus opcode in the instruction L1 cache? For a typical game? That would give a fair idea if staying in 32 bit is worthwhile. You get that much more from the instruction cache, can't be a bad thing if it comes almost for free.

32 bit Windows can do PAE too (> 4GB only enabled on server editions), and before the commodore 128, spectrum 128 etc. did something similar (simpler and low level). But doing this is not free. So for 4GB or more I'd expect 64bit (keeping things simple as well) and for something like 3GB + edram 32bit is more usable.
 
Latest expectations for Orbis and Durango adjusted to rumors.

Orbis
4 core 32-bit processor at 3.0 ghz+
4GB GDDR5 ram
1152 SPU at 800 mhz
+DSPs

Durango
8 core x64 jaguar at 1.6 ghz
8GB of ram + Edram
1024 SPU at 800mhz
+DSPs

sogood.gif

what about the 8-16 gb of ram in ps4 dev kits - it could have 8 gb ddr4 ! :oops:
 
its still almost 2x ddr3 if they use good quality, and much cheaper in the future.

tho with the graphic car using ddr5 they would have to have two separate banks..

DDR4 won't be twice as fast as DDR3 until maybe 2015. Launching in 2013 the fastest a console with DDR4 might see is 2.5GHz, which is achievable with DD3 but probably not likely. Maybe 2-2.4GHz for either DDR3 and DDR4 is what I envision. DDR3 would be a very stupid move despite the performance being the same due to costs increasing in the future.
 
which option is better for nxt gen consoles - fast small EDRAM + lots of slow ddr3 ram; or very fast ddr5 ram only ?
 
Last edited by a moderator:
which option is better for nxt gen consoles - fast small EDRAM + lots of slow ddr3 ram; or very fast ddr5 ram only ?

I believe fast bandwidth in consoles is a must, developers can do a lot of fancy things with fast bandwidth.
XBOX360 (VS PS3 released a year later) or PS2 (VS GC and XBOX1 released 2 years later) clearly demonstrated the advantages of such an approach (fast small memory + bigger slower RAM) with this solution developers would get the best of both worlds (bandwidth for specific tasks + big size of RAM for most tasks).

I think ps3 developers have 2 main problems : the limited bandwidth of the RSX + Insufficient size of RAM pool.
for xbox360, developers are struggling with only the second problem (they need more RAM).

in conclusion : different memory types for different tasks is a huge advantage for consoles. the only problem for next gen is how to give access to all these types of memory for all the processors ? thats a huge challenge....
 
I think ps3 developers have 2 main problems : the limited bandwidth of the RSX + Insufficient size of RAM pool.
for xbox360, developers are struggling with only the second problem (they need more RAM)..

RSX has much more bandwidth to local RAM then 360 does and it can access and use the main XDR RAM so RSX can access just as much memory as 360 can ( Minus memory the OS consumes )
 
Status
Not open for further replies.
Back
Top