Predict: The Next Generation Console Tech

Status
Not open for further replies.
I think it's odd how so many are suggesting separated pools after all the thanks MS got for unified pool last gen
 
I think it's odd how so many are suggesting separated pools after all the thanks MS got for unified pool last gen
I think it has to do with the _amount_ of memory. Last gen, Sony was pretty much slumming it by providing only 240MB or so of graphics memory. Now, where we're talking gigabytes, having a split pool is exactly what PCs have, and can provide ample memory for both graphics and logic.

So it's basically safe to say that there are advantages and disadvantages to both, and having much larger pools in a split configuration mitigates it's disadvantages a lot.
 
Ok, Sony managed to stack few memory chips on top of Vita SoC.

If we take some moderately accurate SoC chip size that can hold quadcore cpu and 1800gflops gpu section, how much surface is there atop of it for memory chips? How many chips can there be stacked, Vita style?
 
Ok, Sony managed to stack few memory chips on top of Vita SoC.

If we take some moderately accurate SoC chip size that can hold quadcore cpu and 1800gflops gpu section, how much surface is there atop of it for memory chips? How many chips can there be stacked, Vita style?

I was thinking about the same thing right after I made this post about about Sony & Ram .
 
I think it's odd how so many are suggesting separated pools after all the thanks MS got for unified pool last gen

I have replied to this statement a couple times already. What do you specifically take issue with my response that you continue to throw it out? Honestly curious.
 
So it's basically safe to say that there are advantages and disadvantages to both, and having much larger pools in a split configuration mitigates it's disadvantages a lot.

Certainly, just so long as you *do* have larger pools of memory! The PS3 vs the Xbox 360 is a good example of how to use split memory pools to your disadvantage though (and not just size - BW for CPU access to GPU memory is something awful like 16 MB/s isn't it?).
 
Certainly, just so long as you *do* have larger pools of memory! The PS3 vs the Xbox 360 is a good example of how to use split memory pools to your disadvantage though (and not just size - BW for CPU access to GPU memory is something awful like 16 MB/s isn't it?).

From PS3devwiki.com

Speed, Bandwidth, and Latency

System bandwith (theoretical maximum):
Cell to/from 256MB XDR : 25.6 GB/s
Cell to RSX (IOIFO): 20GB/s (practical : 15.8GB/s @ packetsize 128B)
Cell from RSX (IOIFI) : 15GB/s (practical : 11.9GB/s @ packetsize 128B)
RSX to/from 256MB GDDR3 : 20.8GB/s (@ 650MHz)
Because of the aforementioned layout of the communication path between the different chips, and the latency and bandwidth differences between the various components, there are different access speeds depending on the direction of the access in relation to the source and destination. The following is a chart showing the speed of reads and writes to the GDDR3 and XDR memory from the viewpoint of the Cell and RSX. Note that these are measured speeds (rather than calculated speeds) and they should be worse if RSX and GDDR3 access are involved because these figures were measured when the RSX was clocked at 550Mhz and the GDDR3 memory was clocked at 700Mhz. The shipped PS3 has the RSX clocked in at 500Mhz (front and back end, although the pixel shaders run separately inside at 550Mhz). In addition, the GDDR3 memory was also clocked lower at 650Mhz.
speed table
Processor 256MB XDR 256MB GDDR3
Cell Read 16.8GB/s 16MB/s (15.6MB/s @ 650MHz)
Cell Write 24.9GB/s 4GB/s
RSX Read 15.5GB/s 22.4GB/s (20.8GB/s @ 650MHz)
RSX Write 10.6GB/s 22.4GB/s (20.8GB/s @ 650MHz)

Because of the VERY slow Cell Read speed from the 256MB GDDR3 memory, it is more efficient for the Cell to work in XDR and then have the RSX pull data from XDR and write to GDDR3 for output to the HDMI display. This is why extra texture lookup instructions were included in the RSX to allow loading data from XDR memory (as opposed to the local GDDR3 memory).
 
why cant an APU have two different types of memory within one unified address space. the highest sig bits determine what memory type, by default cpu based workloads are stored in DDR3 and GPU in GDDR5

expose it to the programers so they can control where data is stored for whatever reason.

its pretty funny you guys are all doom and gloom over a couple of gigs of ram yet seem content with jaguar cpu rumors. im willing to bet a 4 core steamroller APU with 1 gig of ram swill beat the pants off a 8core jaguar APU with 8gig of ram.

without a stupidly large texture footprint what are you going to use it all for? the more memory you have the faster you need to read and write to it if you want to take advantage of it.
 
why cant an APU have two different types of memory within one unified address space. the highest sig bits determine what memory type, by default cpu based workloads are stored in DDR3 and GPU in GDDR5

expose it to the programers so they can control where data is stored for whatever reason.

its pretty funny you guys are all doom and gloom over a couple of gigs of ram yet seem content with jaguar cpu rumors. im willing to bet a 4 core steamroller APU with 1 gig of ram swill beat the pants off a 8core jaguar APU with 8gig of ram.

without a stupidly large texture footprint what are you going to use it all for? the more memory you have the faster you need to read and write to it if you want to take advantage of it.

They can but I'm sure Sony would want the GPU & CPU to work together since that's one of the reasons for having them both on the same chip & if they are working together the GPU would have to drop down to the speed of the memory that the CPU is using so I'm sure they would want for the CPU to have fast Ram too like how they used XDR for the main memory in the PS3.
 
I have replied to this statement a couple times already. What do you specifically take issue with my response that you continue to throw it out? Honestly curious.

Note the "yes, I'm drunk", if it's been anywhere near weekends I've most likely been drunk and thus forgotten it. If it's been during week, I've just missed it.

Gonna search tomorrow when I'm sober
 
They can but I'm sure Sony would want the GPU & CPU to work together since that's one of the reasons for having them both on the same chip & if they are working together the GPU would have to drop down to the speed of the memory that the CPU is using so I'm sure they would want for the CPU to have fast Ram too like how they used XDR for the main memory in the PS3.

first remember the context, the issue is getting 4gig of GDDR5 and then also the cost. so now people are talking 2+2 or 4+2 etc

Memory throughput hasn't been a big bottleneck for low core count CPU's for ages now (latency is).
why would the GPU have to drop down to CPU memory speed (what does that even mean)?

CPU workloads care about latency what has lower latency DDR3 or GDDR5? I cant find a really good answer, it looks like GDDR5 read/write at 256bits and DDR3 at 128bits so non linear access to memory might be better with DDR3 but its running at higher base frequencies then DDR3.

seeing a comparison of random read / random write and sequential read/write at 32/64/128/256 bit word lengths would be very interesting to see.

but even if GDDR5 has better latency and throughput then DDR3 it doesn't matter for what i was talking about because its still just a form of memory QOS. you put data in the location that best matches its usages requirements. Even if you put some of your GPU data on the DDR3 it doesn't mean your throughput drops, infact it likely increases.


im not saying this is a good idea, because the reality is your now at a 256bit bus and assuming cost isn't the issue you just go 4gig GDDR5 ( again assuming its all round better memory).

my point was just because its two different memory types doesn't mean it needs to be two different "pools"

Because not everyone is happy with on rails shooters and bad Xbox ports of amazing games that chug because the game runs out of memory for storing world variables?

i didn't know thats all the PS3 has to offer??? oh wait.........
So maybe thats the developers fault for not considering the systems they where developing for.
 
i didn't know thats all the PS3 has to offer??? oh wait.........
So maybe thats the developers fault for not considering the systems they where developing for.

Well when you put it like that, I guess there's no need at all for Cell to be able to access half of the platform's memory.

Memory is clearly overrated. Only lazydevs and hacks benefit from being able to access it all at will, especially on the (usually) worse performing "port target" platform that has both less available and less total memory.
 
first remember the context, the issue is getting 4gig of GDDR5 and then also the cost. so now people are talking 2+2 or 4+2 etc

Memory throughput hasn't been a big bottleneck for low core count CPU's for ages now (latency is).
why would the GPU have to drop down to CPU memory speed (what does that even mean)?


CPU workloads care about latency what has lower latency DDR3 or GDDR5? I cant find a really good answer, it looks like GDDR5 read/write at 256bits and DDR3 at 128bits so non linear access to memory might be better with DDR3 but its running at higher base frequencies then DDR3.

seeing a comparison of random read / random write and sequential read/write at 32/64/128/256 bit word lengths would be very interesting to see.

but even if GDDR5 has better latency and throughput then DDR3 it doesn't matter for what i was talking about because its still just a form of memory QOS. you put data in the location that best matches its usages requirements. Even if you put some of your GPU data on the DDR3 it doesn't mean your throughput drops, infact it likely increases.


im not saying this is a good idea, because the reality is your now at a 256bit bus and assuming cost isn't the issue you just go 4gig GDDR5 ( again assuming its all round better memory).

my point was just because its two different memory types doesn't mean it needs to be two different "pools"



i didn't know thats all the PS3 has to offer??? oh wait.........
So maybe thats the developers fault for not considering the systems they where developing for.

I think if they are running code in parallel on the CPU & GPU they will have to be using the same bandwidth so I'm pretty sure they wouldn't want the GPU to have to use slower bandwidth just to work with the CPU so why not have fast Ram for both the CPU & GPU?

PS3 used XDR for it's CPU memory so I think it would be a step back to use DDR3 for the PS4 CPU.



Speaking of XDR I remember seeing something about newer AMD high end GPUs using XDR2 could that be what Sony is planing to use in the PS4 & the reason they are looking at 2GB of Ram for the PS4?




XDR™2 Memory Architecture

The XDR™2 memory architecture is the world's fastest memory system solution capable of providing more than twice the peak bandwidth per device when compared to a GDDR5-based system. Further, the XDR 2 memory architecture delivers this performance at 30% lower power than GDDR5 at equivalent bandwidth.

Designed for scalability, power efficiency and manufacturability, the XDR 2 architecture is a complete memory solution ideally suited for high-performance gaming, graphics and multi-core compute applications. Each XDR 2 DRAM can deliver up to 80GB/s of peak bandwidth from a single, 4-byte-wide, 20Gbps XDR 2 DRAM device. With this capability, systems can achieve memory bandwidth of over 500GB/s on a single SoC.

Capable of data rates up to 20Gbps, the XDR 2 architecture is part of the award-winning family of XDR products. With backwards compatibility to XDR DRAM and single-ended industry-standard memories, the XDR 2 architecture is part of a continuously compatible roadmap, offering a path for both performance upgrades and system cost reductions.



The XDR 2 memory architecture is the first to incorporate innovations from Rambus' Terabyte Bandwidth Initiative along with other key Rambus innovations including:

32X Data Rate enables high data rates (up to 20Gbps) at lower system clock and on-chip bus interface speeds.

Fully Differential Memory Architecture (FDMA) improves signal integrity, reduces power and enables the highest memory performance available.

Enhanced FlexPhase™ enables high data rates, simplifies layout and eliminates trace length matching.

FlexLink™ C/A reduces system costs and controller pin-count while providing scalable capacity and flexible access granularity.

FlexMode™ interface technology enables support of both differential and single-ended memories in a single SoC package design with no extra pins.

Micro-threading increases transfer efficiency on micro-threaded workloads while reducing power consumption.
 
PS3 used XDR for it's CPU memory so I think it would be a step back to use DDR3 for the PS4 CPU.
DDR3 is what CPUs use these days, so that's what it'll be. It's not a "step back". DDR3 can easily match or exceed the XDR memory bus of the PS3 today, and DDR3 chips are cheap, mass-produced commodity products while XDR remains extremely fringe (nevermind XDR2 which doesn't even exist except on paper, and maybe prototype form from what I understand.)
 
DDR3 is what CPUs use these days, so that's what it'll be. It's not a "step back". DDR3 can easily match or exceed the XDR memory bus of the PS3 today, and DDR3 chips are cheap, mass-produced commodity products while XDR remains extremely fringe (nevermind XDR2 which doesn't even exist except on paper, and maybe prototype form from what I understand.)

If they are going with stacked memory, isn't completely irrelevant what is mass-produced and what not?
 
Status
Not open for further replies.
Back
Top