Support NeoGAF

[IHaveCandy]

Somebody correct me if I'm wrong. People are pointing the BD drive as a bottleneck for the GDDR5 ram speeds, but isn't the benefit of 176 GBps shown with how fast the GPU and CPU can access memory? I thought this is how it works:

1. BD/HD drive feeds data to RAM at its own speed.

2. That data sits in RAM, ready to be used whenever the CPU or GPU needs it.

3. When the GPU needs to pull a ton of assets that have already been loaded into RAM, this is where that 176 GBps bandwidth comes into play. The GPU and CPU have more immediate access to content already loaded into RAM and can process it faster.

 

[Gorgon]

not sure whats there really extra except for eSRAM but then again, with 8 GB of GDDR5, PS4 mobo is pretty complex... plus there is an extra cpu that runs OS.

then again who really knows right now... I think they are targeting either more advanced kinect or cheaper box at the end but we wont know for sure.

Is this confirmed at all? I've heard this thrown around here and there but I didn't see anything solid.

 

[Perkel]

I edited my post instead of replying.

 

[IHaveCandy]

Is this confirmed at all? I've heard this thrown around here and there but I didn't see anything solid.

There is an extra CPU that runs the video encoding and background downloading/installing. It sounds like the OS itself runs on the main CPU.

 

[teh_pwn]

BDROM is only a bottleneck for your RAM if you RAM pool is so small it has to continually flush and reload from disc.

RAM capacity got big several years ago on PCs and one of the optimizations in OS software design has been to keep a cache of recently opened apps.

Once the game assets are initially loaded, you'll only be accessing your BDROM for videos and audio.

RAM latency does not matter (relative between GDDR and DDR) for video games. GDDR is specifically a type of memory for video applications, including games. I feel like a doctor in a pseudo science health thread.

 

[Gorgon]

There is an extra CPU that runs the video encoding and background downloading/installing. It sounds like the OS itself runs on the main CPU.

Thanks. That was exactly my impression too. The OS is running on the 8 core CPU, as expected. I really wouldn't expect otherwise, but since it was being thrown around...

 

[PjotrStroganov]

BDROM is only a bottleneck for your RAM if you RAM pool is so small it has to continually flush and reload from disc.

RAM capacity got big several years ago on PCs and one of the optimizations in OS software design has been to keep a cache of recently opened apps.

Once the game assets are initially loaded, you'll only be accessing your BDROM for videos and audio.

RAM latency does not matter (relative between GDDR and DDR) for video games. GDDR is specifically a type of memory for video applications, including games. I feel like a doctor in a pseudo science health thread.

Besides that, it has been announced that downloadable games will be playable while they are being downloaded. I bet the tech will also be used for loading from BD.

 

[teh_pwn]

Yeah I'm not convinced the GDDR5 will be a monumental benefit. Bigger bandwidth but higher latency over DDR3.

Lets say higher bandwidth can deliver 10 boxes in 10 seconds, whereas narrower bandwidth lower latency can deliver 3 boxes in 4 seconds. Yeah in 10 seconds on DDR3 I can only get 7.5 boxes compared to 10 . But what if I only practically need 3? I can get them faster on DDR3. And this is not even factoring the ESRAM which can deliver me 1/2 box in 1 second. What if half a box is as much as I need for a damn good period of time.

All of a sudden the Durango outperforms PS4. NOTE: boxes and ratios completely fictional to illustrate a point.

I think the real clincher will be who has the fatter GPU muscle. And we know how that's allegedly looking ATM.

GDDR exists because it's designed to operate optimally for graphics cards. If DDR3 was competitive, GDDR5 would not be used.

I imagine lower latency might matter for PC applications like Xilinx Vivado, which is entirely CPU compute. But lots of apps are moving to GPU acceleration, so I'm just not seeing the benefit for DDR3 for consoles other than cost.

Besides that, it has been announced that downloadable games will be playable while they are being downloaded. I bet the tech will also be used for loading from BD.

Do you mean from HDD?

If not and the console OSes are too dumb to dynamically allocate different standby caches in memory, that's a horrible waste of RAM for a feature many people don't care about.

Optical discs are a necessary evil for reliability, cost, and capacity. As disc readers for software, they have fallen far behind flash drives, SSD, and even HDD. I cannot wait for mass Gb internet everywhere so things can be over the network.

 

[ypo]

Xbox only fans don't seem to have meltdowns like the Sony only crowd do...

From some uber mad dude.

Dude no... haven't you read up on this week's news? Here is the performance breakdown path of all games (including those that have yet to be conceived):

PS4 >>>> PC >>>>>>>>>>>>>>>>> Durango >> PS3/360 >>>>>>>>>>>> Wii U

That GDDR5 RAM makes all the difference in the world.

In many cases, the Durango version of all early games will likely struggle to perform better than the PS3/360 versions due to how slow its DDR3 RAM is and how much sluggish it will appear to be running against its PS4 and PC counterparts while trying to run all of those high-end graphical processes.

Note: I don't actually believe this. Just parroting GAF's ridiculous group-think consensus that has manifested over the past 60+ hours since the PS4 was crowned the next-gen victor on Wednesday night.

 

[Klocker]

From some uber mad dude.

it's not anger as much as frustration with the stupidity being spouted about this GDDR5 hyperbole.

 

[Perkel]

Someone correct me if i am wrong.
Small example of bandwidth vs size of ram. We assume there are no bottlenecks (like inferior GPU CPU etc). We love to play 1FPS games to further simplify our test.

We have two memory pools:

M1 - 20MB which has 20Megabytes/s bandwidth
M2 - 40MB with 5MB/s bandwidth.

We want to make a forest in our game.

Now we load for example 1MB tree (mesh,textures, everything) into our M1 and M2.

M1- 1 out of 20MB is reserved
M2- 1 out of 40MB is reserved

Because developers reuse assets we will now make a forest from this 1MB tree.

M1 thanks to 20MB/s bandwidth can output 20 tree models a second.
M2 thanks to 5MB/s can output 5 tree models a second.

Now because our tree forest can't be just one tree "copypasta" we will add 4 other tree variations (each 1MB).

M1 - 5 of 20MB reserved for 5 tree variants
M2 - 5 of 40MB reserved for 5 tree variants

M1 output is 20 trees forest which is consisted of 5 variations of tree model.
M2 output is 5 trees forest which is consisted of 5 variations of tree model.

Now we add 5 more variations to our forest

M1 - 10MB/20MB reserved and can output 20 trees with 10 variations
M2 - 10MB/20MB reserved and can output 5 trees with 5 variations.

As you can see we can add more variations and still slower bandwidth simply won't be able to show you forest. What size gives you is variety of assets
What bandwidth gives you is how much of those assets can be output into final image.

And there is another thing that you must understand. Developers reuse assets ALL TIME.

This is why no one uses cheap DDR3 over expensive GDDR5 in GFX cards. DDR3 is good for doing things like browsing internet, checking mail, writing, watching youtube, playing old games which were not created with high bandwidth in mind.

Someone said that games which take only 2GB of ram don't need more than 2GB/s bandwidth. That is completely untrue. Since developers reuse assets you can have game in which only 100MB is used but because they reuse a lot of assets you can run out of bandwidth.

This is also why EDRAM in X360 and PS2 was so important it gave it bandwidth to better use small amout memory which had better effects on overall end picture than just more main slow memory.

This is also why 4GB of GDDR5 ram was better than 8GB of DDR3 ram and still would be better than 16GB or 32GB of DDR3 ram.

And this is also why it is so important for next gen games because those games will be pushing a lot of things on screen.

 

[JohnnySasaki86]

Unlike the 360 you won't have to fit everything in the embedded ram this time around.

How? It's still only 32mb of SRAM.

 

[Drek]

I believe that the bottleneck will be the speed of the bluray drive on both systems. From what i can remember Sony is going for a 6x drive, i guess MS will use similar drive speeds. If the information on wikipedia is correct a 6x drive can read 27mb per second, and that doesnt even take seek times into account. If we consider that the OS and other functions will take away RAM from both systems...i just calculated with 6 GB for games, then it will take almost 4 minutes to fill the RAM. I can imagine that games need to be installed if we want acceptable loading times!

I kinda want to be a dick here and just write "DATA COMPRESSION, HOW DOES IT WORK?" with an ICP pic, but that doesn't really do much for the discussion I supposed.

Look, 27MB/s from a blu-ray drive gets unpacked by decompression algorithms that both systems are going to have on hardwired chipsets. Both will likely have damn near the exact same chipset for this in fact (if not the exact same).

The system basically "unpacks" the 27MB/s on it's way to memory, making it something far bigger along the way.

Now obviously this is still a bit of a bottleneck on current generation hardware and this is why we've seen such significant use of harddrive installs (so that you can then fill memory from the faster HDD and the optical drive at the same time). But the solution to that bottleneck isn't just ignoring the additional memory, it's load times.

Games that stream entirely from an optical disc (like many open world games of today) rely heavily on compression/decompression. That happens to be a rapidly evolving field that has seen some pretty massive gains since the PS3 and Xbox 360 were finalized, FYI, so being concerned about the media being able to feed all this ram is not particularly valid. It'll be sorted out just fine and developers will still use every scrap of memory they have at their disposal.

Is this confirmed at all? I've heard this thrown around here and there but I didn't see anything solid.

It supposedly has an ARM CPU included on the mobo to handle smaller OS operations and to maintain Sony's announced "game suspend" feature (and I'm sure other low power features. If this is true the purpose is obviously rather clear, ARM CPUs are very energy efficient and would be ideal for things like auto-updates without a full system boot and the like.

it's likely on the extreme low end relative to the rest of the hardware and will therefore probably have zero legitimate effect on games.

 

[AzaK]

Somebody correct me if I'm wrong. People are pointing the BD drive as a bottleneck for the GDDR5 ram speeds, but isn't the benefit of 176 GBps shown with how fast the GPU and CPU can access memory? I thought this is how it works:

1. BD/HD drive feeds data to RAM at its own speed.

2. That data sits in RAM, ready to be used whenever the CPU or GPU needs it.

3. When the GPU needs to pull a ton of assets that have already been loaded into RAM, this is where that 176 GBps bandwidth comes into play. The GPU and CPU have more immediate access to content already loaded into RAM and can process it faster.

Sure, but that assumes the data is in RAM ready to go. With 8GB you might find it could be but it needs to come from disk at some point. Either through an upfront level load cost, or the engine streams extra stuff in that you might eventually get to. Essentially using a more liberal algorithm that people use now (Due to limited memory).

It's all about your algorithms being as optimal as possible. One that makes sure textures are in memory when the system needs them.

Someone correct me if i am wrong.
Small example of bandwidth vs size of ram. We assume there are no bottlenecks (like inferior GPU CPU etc). We love to play 1FPS games to further simplify our test.

We have two memory pools:

M1 - 20MB which has 20Megabytes/s bandwidth
M2 - 40MB with 5MB/s bandwidth.

We want to make a forest in our game.

Now we load for example 1MB tree (mesh,textures, everything) into our M1 and M2.

M1- 1 out of 20MB is reserved
M2- 1 out of 40MB is reserved

Because developers reuse assets we will now make a forest from this 1MB tree.

M1 thanks to 20MB/s bandwidth can output 20 tree models a second.
M2 thanks to 5MB/s can output 5 tree models a second.

Now because our tree forest can't be just one tree "copypasta" we will add 4 other tree variations (each 1MB).

M1 - 5 of 20MB reserved for 5 tree variants
M2 - 5 of 40MB reserved for 5 tree variants

M1 output is 20 trees forest which is consisted of 5 variations of tree model.
M2 output is 5 trees forest which is consisted of 5 variations of tree model.

Now we add 5 more variations to our forest

M1 - 10MB/20MB reserved and can output 20 trees with 10 variations
M2 - 10MB/20MB reserved and can output 5 trees with 5 variations.

As you can see we can add more variations and still slower bandwidth simply won't be able to show you forest. What size gives you is variety of assets
What bandwidth gives you is how much of those assets can be output into final image.

And there is another thing that you must understand. Developers reuse assets ALL TIME.

This is why no one uses cheap DDR3 over expensive GDDR5 in GFX cards. DDR3 is good for doing things like browsing internet, checking mail, writing, watching youtube, playing old games which were not created with high bandwidth in mind.

Someone said that games which take only 2GB of ram don't need more than 2GB/s bandwidth. That is completely untrue. Since developers reuse assets you can have game in which only 100MB is used but because they reuse a lot of assets you can run out of bandwidth.

This is also why EDRAM in X360 and PS2 was so important it gave it bandwidth to better use small amout memory which had better effects on overall end picture than just more main slow memory.

This is also why 4GB of GDDR5 ram was better than 8GB of DDR3 ram and still would be better than 16GB or 32GB of DDR3 ram.

And this is also why it is so important for next gen games because those games will be pushing a lot of things on screen.

You're missing one minor point. What if you can't actually process that much data per frame? As in, your GPU can't handle it. Doesn't matter how fast the memory is if the GPU CPU/GPU contention etc means you can actually only get 1/4 of that out of RAM in a frame. I'm not saying that the PS4 will have this problem, but I am a little hesitant to just say "Faster" always means better.

 

[rudieboy77]

Another thread discussion about actual specifications versus hypothetical specifications.

 

[Perkel]

You're missing one minor point. What if you can't actually process that much data per frame? As in, your GPU can't handle it. Doesn't matter how fast the memory is if the GPU CPU/GPU contention etc means you can actually only get 1/4 of that out of RAM in a frame. I'm not saying that the PS4 will have this problem, but I am a little hesitant to just say "Faster" always means better.

Read 3rd sentence "We assume there are no bottlenecks (like inferior GPU CPU etc)"

I just did it for comparison of what bandwidth and size do.

 

[Ol No Bones]

From some uber mad dude.

Who's mad. I'm just saying that if the ps4 is weaker at all you will see serious meltdowns. I have yet to encounter an Xbox boy that is freaking out about the ps4 news. I just too many people are assuming too much about the Xbox.

 

[Raydeen]

8gig of DDR3 don't matter.

8gig of GDDR5 won't matter.

All matters is do you a machine that plays the next HALO or UNC?HARTED.

 

[zoukka]

We are going to be talking about RAM for the next 7 years aren't we...

 

[pottuvoi]

How? It's still only 32mb of SRAM.

This time ROPs and TEX units can read and write to both SRAM and main ram.

This should allow some interesting things.
IE. For post effects having source buffer/texture in SRAM and writing to frame buffer in main ram should have very nice performance. (Especially if there is need for 'random' reads.)

 

[LabouredSubterfuge]

8gig of DDR3 don't matter.

8gig of GDDR5 won't matter.

All matters is do you a machine that plays the next HALO or UNC?HARTED.

I think you're in the wrong thread.

 

[Shtof]

Is that because MS will force their version to look equal to the competition? Meaning we'll only see a difference on PC? Wasn't this a rumour/fact going around?

Lol, I don't think that's a possibility, if Sony couldn't do this in the 6th or 7th generations.

 

[Izick]

Is that because MS will force their version to look equal to the competition? Meaning we'll only see a difference on PC? Wasn't this a rumour/fact going around?

This is some conspiracy-theory level of shit. There's now way that's even feasible.

 

[Arkham]

That picture is hilarious because the lower Ram is like "bitch I don't even care, I don't even wanna fuck that female over there."

So, your preference is to play with a male box not a female box. Gotcha.

 

[Scrooged]

The difference is the PS4 will bankrupt developers faster.

 

[AgentP]

The difference is the PS4 will bankrupt developers faster.

With easier architecture and quicker dev times?

 

[BrainZEROX]

Unlike the 360 you won't have to fit everything in the embedded ram this time around.

Microsoft's approach (the rumored one more exactly) is interesting. Orbis needs large bandwidth because everyone (CPU, GPU, background task processing chip, video compression/decompression chip, audio chip, DMAs...) will access the GDDR5 memory pool. A large bandwidth is a necessity to serve everyone without anyone starving. I think this is an artifact of GCN's architecture and Microsoft took a different path than Sony here. They also deal with a unified memory architecture but Microsoft seems to have isolated the most hungry one, the GPU, so it will primarily address the ESRAM not the DDR3 pool. DDR3 will then be accessed by comparatively less "hungry for bandwidth" units and this is where the DMEs make sense as they seem to have their own path. Continuously filling the ESRAM will not contribute to saturate the bus that gives access to DDR3 to the CPU and other non GPU chips and they share 68 GB/s. In that regard, I think the gap between Orbis and Durango is much closer than we might think. Sure Orbis has 176GB/s bandwidth to GDDR5 but it has to be shared so, for the GPU only, it probably means quite a bit less than 176GB/s. Durango's GPU could deal with constant 102GB/s without having to share if the DMEs are able to fill ESRAM continuously but you have to balance your algorithms so the GPU works in ESRAM mainly. In my understanding (I may be wrong of course), that would mean it's not a matter of "176>102 gameover kid", it's a bit more complex and Microsoft's tradeoff (complexity versus cost) is quite interesting I think.

 

[onQ123]

The difference is the PS4 will bankrupt developers faster.

How the fuck did you come to this conclusion?

 

[Y2Kev]

We are going to be talking about RAM for the next 7 years aren't we...

We kinda were this gen too. Ask all the Skyrim players!

 

[i-Lo]

The difference is the PS4 will bankrupt developers faster.

My head hurts.

 

[Crimson_Gold]

The difference is the PS4 will bankrupt developers faster.

Huh? Please explain this. I fail to understand how this could be the case

 

[SgtCobra]

With easier architecture and quicker dev times?

lol pretty hard to counter this reply.

 

[BrainZEROX]

The difference is the PS4 will bankrupt developers faster.

That doesn't make a lot of sense. Both PS4 and Durango seem pretty much developer friendly machines.

 

[teh_pwn]

The difference is the PS4 will bankrupt developers faster.

There has been a myth for a long time that increased hardware capability increases development costs.

This is misplaced. Better reasons:
1. Platform specific complexity, lack of code reuse means paying hundreds of software engineers and computer scientists to create code that will be thrown away later. Those are large salaries that could be put to making unique games.
2. Bad management with lavish marketing campaigns, CG, and modeling things that don't matter rather than using hardware capabilities for free stuff like high res textures and IQ.

 

[Dead Prince]

we just been Scrooged.

 

[polyh3dron]

Durango will have the lowest common denominator, so multiplatform games will still probably look the same on both consoles unless Bethesda feels like shipping a broken game on one of the consoles which has never happened before.

and lol @ "PS4 will bankrupt developers faster".

 

[bgassassin]

And maybe Durango is a Tile Based Rendering:

http://www.pjblewis.com/articles/tile-based-forward-rendering/

From what I understand it's not designed for that.

 

[JABEE]

We kinda were this gen too. Ask all the Skyrim players!

Yeah I wish we lived in the mythical world where all multi-platform games ran the same this generation. It didn't happen for most of the titles on PS3 for the first two years of the systems existence, and there were frame rate issues for the rest of its life up to today.

From the looks of it, the GPU will be substantially more powerful than the rumored Durango specs, and the Ram is a more advanced version of the same capacity. I think there will be noticeable differences in multi-plat games just as there were on the 360 which had a GPU and Ram advantage.

 

[Scrooged]

Publishers gonna demand bigger, flashier games that use the most out of the system. I don't care how easy the tools are getting, you still need to create the art and build the actual game + mechanics. AAA games are costing more than ever. It's going into the hundreds of millions of dollars. How many times have you heard that a certain game needs to sell X million copies to break even? Why do you think games cost $60 and yet publishers are still trying to find every which way to charge us more in the form of DLC and microtransactions? Why do you think over 100 dev studios have closed down over the past 5 years?

 

[Hoo-doo]

Publishers gonna demand bigger, flashier games that use the most out of the system. I don't care how easy the tools are getting, you still need to create the art and build the actual game + mechanics. AAA games are costing more than ever. It's going into the hundreds of millions of dollars. How many times have you heard that a certain game needs to sell X million copies to break even? Why do you think games cost $60 and yet publishers are still trying to find every which way to charge us more in the form of DLC and microtransactions? Why do you think over 100 dev studios have closed down over the past 5 years?

Art and models made for current-gen games are at a much, much higher fidelity than the PS3 or the 360 can even display. Costs might rise, but not nearly as much as you're implying.

If anything devs can now make games without being ridiculously constrained by ancient hardware, which also takes a ton of time and money for optimization.

 

[FlutterPuffs]

Publishers gonna demand bigger, flashier games that use the most out of the system. I don't care how easy the tools are getting, you still need to create the art and build the actual game + mechanics. AAA games are costing more than ever. It's going into the hundreds of millions of dollars. How many times have you heard that a certain game needs to sell X million copies to break even? Why do you think games cost $60 and yet publishers are still trying to find every which way to charge us more in the form of DLC and microtransactions? Why do you think over 100 dev studios have closed down over the past 5 years?

It has been said over and over again by devs that it's much easier this time than during the jump from SD to HD. For instance Square Enix's Luminous Engine:

Square Enix chief technical officer Yoshihisa Hashimoto claimed that despite its startlingly realistic visual fidelity and new features, Luminous Studio can reduce game development costs by up to 30 per cent, and expects it to enable faster development cycles.

Same progress is being made all over so relax.

 

[Sky Chief]

Yeah I wish we lived in the mythical world where all multi-platform games ran the same this generation. It didn't happen for most of the titles on PS3 for the first two years of the systems existence, and there were frame rate issues for the rest of its life up to today.

From the looks of it, the GPU will be substantially more powerful than the rumored Durango specs, and the Ram is a more advanced version of the same capacity. I think there will be noticeable differences in multi-plat games just as there were on the 360 which had a GPU and Ram advantage.

Exactly. This gen devs were frequently way too ambitious as well resulting in many sub 30 fps framerates and sub HD resolutions. Hopefully the alleged GPU and RAM advantage of the PS4 will alleviate these issues in the PS4 version.

 

[AgentP]

Microsoft's approach (the rumored one more exactly) is interesting. Orbis needs large bandwidth because everyone (CPU, GPU, background task processing chip, video compression/decompression chip, audio chip, DMAs...) will access the GDDR5 memory pool. A large bandwidth is a necessity to serve everyone without anyone starving. I think this is an artifact of GCN's architecture and Microsoft took a different path than Sony here. They also deal with a unified memory architecture but Microsoft seems to have isolated the most hungry one, the GPU, so it will primarily address the ESRAM not the DDR3 pool. DDR3 will then be accessed by comparatively less "hungry for bandwidth" units and this is where the DMEs make sense as they seem to have their own path. Continuously filling the ESRAM will not contribute to saturate the bus that gives access to DDR3 to the CPU and other non GPU chips and they share 68 GB/s. In that regard, I think the gap between Orbis and Durango is much closer than we might think. Sure Orbis has 176GB/s bandwidth to GDDR5 but it has to be shared so, for the GPU only, it probably means quite a bit less than 176GB/s. Durango's GPU could deal with constant 102GB/s without having to share if the DMEs are able to fill ESRAM continuously but you have to balance your algorithms so the GPU works in ESRAM mainly. In my understanding (I may be wrong of course), that would mean it's not a matter of "176>102 gameover kid", it's a bit more complex and Microsoft's tradeoff (complexity versus cost) is quite interesting I think.

Sounds like you are trying to convince yourself. Did you see the leaked docs and flow chart? You can assuming they are wrong, but the bottom line is the 32MB eSRAM is jsut a cache. 32MB is 0.39% the size of the 8GB pool. It is not some faster but equal size chunk of memory. What you sound like you are describing is the PS3's split memory, where the full bandwidth of the XDR is available to the CPU and the full bandwidth of the GDDR3 is available to the GPU. This is the PC model. The Durango model is an evolution of the 360 one, a small cache to help alleviate the bandwidth issues of the unified memory.

Each move engine can read and write 256 bits of data per GPU clock cycle, which equates to a peak throughput of 25.6 GB/s both ways. Raw copy operations, as well as most forms of tiling and untiling, can occur at the peak rate. The four move engines share a single memory path, yielding a total maximum throughput for all the move engines that is the same as for a single move engine. The move engines share their bandwidth with other components of the GPU, for instance, video encode and decode, the command processor, and the display output. These other clients are generally only capable of consuming a small fraction of the shared bandwidth.

The careful reader may deduce that raw performance of the move engines is less than could be achieved by a shader reading and writing the same data. Theoretical peak rates are displayed in the following table.

Copy Operation Peak throughput using move engine(s) Peak throughput using shader
RAM ->RAM 25.6 GB/s 34 GB/s
RAM ->ESRAM 25.6 GB/s 68 GB/s
ESRAM -> RAM 25.6 GB/s 68 GB/s
ESRAM -> ESRAM 25.6 GB/s 51.2 GB/s

I fail to see how the GPU can sustain 102GB/s. I can't find that flow chart with the bandwidth between sub-systems, I'm not sure if it is even based on fact, though it was pretty complete.

 

[BruiserBear]

It's strange seeing so much talk about the RAM differences in these consoles, when the real differentiator is gonna be the GPU and ROPs advantage of the PS4.

 

[TimeEffect]

This is some conspiracy-theory level of shit. There's now way that's even feasible.

Sorry, was just going off of something I read a while back about MS's policies/standards, hence why I asked

 

[ProfessorMoran]

8gig of DDR3 don't matter.

8gig of GDDR5 won't matter.

All matters is do you a machine that plays the next HALO or UNC?HARTED.

We're in the era of HALO without Bungie, what's the point? It's over man, it's all over.

 

[TheExecutive]

I guess my question is the motherboard. I mean you can't just put 4 more gigs on without leaving some room on just in case. Same goes for MS

Right?

 

[BrainZEROX]

Sounds like you are trying to convince yourself. Did you see the leaked docs and flow chart? You can assuming they are wrong, but the bottom line is the 32MB eSRAM is jsut a cache. 32MB is 0.39% the size of the 8GB pool. It is not some faster but equal size chunk of memory. What you sound like you are describing is the PS3's split memory, where the full bandwidth of the XDR is available to the CPU and the full bandwidth of the GDDR3 is available to the GPU. This is the PC model. The Durango model is an evolution of the 360 one, a small cache to help alleviate the bandwidth issues of the unified memory.

No, I'm not trying to convince myself, I'm just trying to understand. What I'm describing is a system with a GPU that uses 32MB of ESRAM as a cache... So, we do agree but my wording may have been confusing as english is not my native language, sorry for that.

I fail to see how the GPU can sustain 102GB/s. I can't find that flow chart with the bandwidth between sub-systems, I'm not sure if it is even based on fact, though it was pretty complete.

I got my information from Vgleaks too:

- 8 gigabyte (GB) of RAM DDR3 (68 GB/s)
- 32 MB of fast embedded SRAM (ESRAM) (102 GB/s)

But you're right though, when the DMEs are pumping data, they "eat" one significant part of that bandwidth (up to 25.6GB/s from your informations it seems) so that's not 102 GB/s for the GPU only.

 

[Y2Kev]

Sorry, was just going off of something I read a while back about MS's policies/standards, hence why I asked

We know it isn't true though. There are superior PS3 games.

They USED to have a rule that said you can't have free content on other systems that you charge for on XBL. But...they changed that. I guess because they are part Succubus.

 

[Angelus Errare]

Some of the posts in here...you guys are taking the GAF tagline a bit too seriously.