From the rumor you trusted earlier, the PS4 has 32 rops.
NGGP: NextGen Garbage Pile (aka: No one reads the topics or stays on topic) *spawn*
- Thread starter Barso
- Start date
- Status
no link file under hearsay/rumor/pretend inside sources
i am sure others have mentioned it on this board in some thread or another recently. maybe even this one. i know i saw it being discussed recently.
the difference is said to be slim, but why it would be true at all given the specs and cu counts we have is an interesting question.
could it be they're both just running 8 rops? do rops even determine that?
edit: lulz at thread tags as always. but it's missing the flops capacitor...
ROPs (and bandwidth) would determine resolution and frame rate, not triangle/verts.
But yes I'm expecting a higher ROP count and the bandwidth to support them on Orbis (from what specs I've seen).
I expect they will both be "similar" but the frame-rate and resolution edge for multiplats will go to Orbis if the specs hold up.
I'm expecting 20-30fps on Durango and 40-60fps for the same game/content on Orbis.
Albofighter
Newcomer
When the consoles are announced, will you guys be able to tell us every tech detail you know? I would love to see how sony and microsoft had their vision and what kind of priorities they had.
Charlie said devs are disappointed because final hardware is inferior to initial dev kits. Hope that's not true but we'll have to wait a little longer I guess.
I wonder if vgleaks will post something next monday. It seems every week we're getting new details.
Come on man, you really don't wanna miss out on things like hair, mustache, fluid dynamic particles and advanced lighting physics in nextgen games. Trust me even if 3rd parties don't abuse the 4CUs, Sony 1st parties would vindicate their existence so much that 3rd parties would get seriously inspired and encouraged to abuse them right after.
So far Charlie has been mildly amusing. I'm not sure if he can be considered among the "good" sources. Following occam's razor, I won't pay 50$ to find out.
Huh? ROPs and bandwidth are not the sole determiner of resolution and frame rate. Triangles/verts factor in as do other things.
Track record for Charlie (his articles that I remember):
Fermi being the end of Nvidia.
Oban being Durango's SOC name
Durango is PPC
1% change for Durango to have x86
PS4 having 2.5D or 3D technology
Durango having less than 8GB of ram
Devs are disappointed with beta kits.
Disappointing Surface RT sales somehow signaling the end of MS.
I just hope your right about durango, from these specs it looks like ps4 is 50% stronger then durango, i'm just hoping durango could at least get close to the bandwidth and gpu specs of obis cause multi platform games will suffer, especially if they are targeting 30fps on orbis.
He loves Microsoft
and Intel, and Nvidia.
arijoytunir
Regular
dont forget devs are actually more excited for durango more than orbis asper the rumors.
From business perspective, the fun factor is more important than the tech factor. As long as MS focuses on delivering new and exciting experience, the consumers will welcome them. If the developers sees the early potential, they will be happy too. This is not to say Durango is less or more powerful than Orbis. Specs may not be that important in the new climate.
I'd be interested to see Sony's new approach. If it's the same old, same old Playstation Move, PS Eye, 3D combo, and 4K concept, it would be a non-event for consumers at large.
EDIT: If Durango can be packaged and sold as an entertainment center or home PC instead of a game console, Microsoft may be able to market it openly and aggressively in China. That market is huge.
Sony may be thinking of something similar, which is why their CTO was talking about experimenting with some sort of PS4 Linux. IMHO, it's the product concepts that's more interesting in the end for both devices.
I'd be interested to see Sony's new approach. If it's the same old, same old Playstation Move, PS Eye, 3D combo, and 4K concept, it would be a non-event for consumers at large.
EDIT: If Durango can be packaged and sold as an entertainment center or home PC instead of a game console, Microsoft may be able to market it openly and aggressively in China. That market is huge.
Sony may be thinking of something similar, which is why their CTO was talking about experimenting with some sort of PS4 Linux. IMHO, it's the product concepts that's more interesting in the end for both devices.
I think I got this...
So, what you are essentially saying is that the modified jaguar CPU in Durango has double the floating point units (I.e., not shared amongst 2 cores like in stock jaguar) and thus double the flops (I.e., 102 x 2 = 204 Gflops).
You are also alluding to the audio processor doubling as a FPGA. FPGAs are incredibly flexible but are not known for having a high flop output for the transistor density. How many additional Gflops could it potentially produce...another 100 to 200 high-quality Gflops max, if that?
In the best case scenario for Durango, you have a 200 Gflop CPU and a 200 Gflop FPGA in addition to a 1200 Gflop high-efficiency, rendering-centric GPU (which may not be able to double as a GPGPU compute unit at all so as to maximize rendering efficiency). 1800 GFLOPs in total for Durango in this instance. Definitely a CPU-centric machine if this is true.
In the best case scenario for Orbis, you have 1800 GFLOPs worth of GPGPU CUs (400 GFLOPS of which are primarily compute worthy and secondarily for rendering if need be - 1400 GFLOPS of which are primarily rendering worthy and secondarily for compute if need be). In addition, you have a stock 100 GFLOP jaguar CPU. 1900 GFLOPs in total for Orbis in this instance. Definitely a GPU-centric machine if this is true.
The memory architectures of each will play to the strengths/focal points of each.
Does that sound about right?
So, what you are essentially saying is that the modified jaguar CPU in Durango has double the floating point units (I.e., not shared amongst 2 cores like in stock jaguar) and thus double the flops (I.e., 102 x 2 = 204 Gflops).
You are also alluding to the audio processor doubling as a FPGA. FPGAs are incredibly flexible but are not known for having a high flop output for the transistor density. How many additional Gflops could it potentially produce...another 100 to 200 high-quality Gflops max, if that?
In the best case scenario for Durango, you have a 200 Gflop CPU and a 200 Gflop FPGA in addition to a 1200 Gflop high-efficiency, rendering-centric GPU (which may not be able to double as a GPGPU compute unit at all so as to maximize rendering efficiency). 1800 GFLOPs in total for Durango in this instance. Definitely a CPU-centric machine if this is true.
In the best case scenario for Orbis, you have 1800 GFLOPs worth of GPGPU CUs (400 GFLOPS of which are primarily compute worthy and secondarily for rendering if need be - 1400 GFLOPS of which are primarily rendering worthy and secondarily for compute if need be). In addition, you have a stock 100 GFLOP jaguar CPU. 1900 GFLOPs in total for Orbis in this instance. Definitely a GPU-centric machine if this is true.
The memory architectures of each will play to the strengths/focal points of each.
Does that sound about right?
Hecatoncheires
Newcomer
so everybody uses their physics hardware to add more non-interactive gfx just a different kind...love it
That's the point: Nvidia PhysX is non-interactive. you can't use GPGPU algorithms for interactive gameplay computations due to the copy overhead between CPU/DDR3 and GPU/GDDR5 on every modular PC. The communication between CPU and GPU takes way too long and kills the speedup that you would gain from utilizing the GPU for these kind of tasks. That's why Nvidia's CUDA based GPGPU approach "PhysX" is used for visual effects only, which can be output directly to the renderer and don't have to be sent back to the CPU.
AMD's HSA claims to allow for interactive gameplay GPGPU algorithms: With the 28nm 3rd gen HSA tech that is rumored for the new consoles, you will have features like the unified adress space for CPU and GPU, which allows you to have CPU and GPU working on tasks together without that nasty copy overhead. That means that there is a whole bunch of new GPGPU algorithms that can only be utilized with a heterogeneous processor. Instead of having just some graphics effects like occlusion, lens flare, DoF, motion blur, and smoke and mirror effects like particles, fluids, textiles, you will see some GPGPU effects that can interact with gameplay (e.g. the character shielding his or her eyes because of eyes, mouth beacause of smoke), you will see GPGPU algorithms for A.I., for driving physics (I bet Yamauchi Kazunori was really looking forward to working with this), for much more complex animations, for complex pathfinding, and so on.
DirectCompute is an essential for DX11 graphics. Durango will need to utilize GPGPU algorithms (or at least use some other processing elements that compensate for it) in order to deliver this level of graphics anyway. And it needs a heterogeneous processor approach to use these effects for gameplay mechanics.
What he's saying is if you take the best case scenario for Durango and compare it to the worst case scenario for Orbis (and then ignore a number of rumored hardware features), maybe the Durango won't look so bad.
Love_In_Rio
Veteran
Proelite said that in neogaf.
Rangers, the most plausible theory about the 4 special CUs we can make from the leaks is that are normal CUs with a scalar ALU added to help branching code and less stalls in the vector ALUs.
And the "minor boost" ( not "help" ) is thank to these CUs having this extra ALU that gives an extra graphics push.
When i read "minor boost" i can´t avoid thinking in the Fantastic Car when Michael Knight pushed the button to chase the bad guys, and think Kutaragi is Michael Knight and the bad guys MS
Well, your theory is as well plausible and would involve Orbis being more like an APU with 4 CUs plus a 14 CU GPU, but then i don´t see the way the 4 CUs would give "minor boost" to graphics if needed. Besides, this doesn´t match with the HSA provided evolution.
And about Jaguar fpu, in the standard model each core has its fpu, so AndyH surely heard something but not the right thing.
Love_In_Rio
Veteran
AstoundingHolmes
Newcomer
Are we sure that Durango is Jaguar based? People have mentioned the rumors don't explicitly state Jaguar. Perhaps it's richland/vishera downclocked with some special sauce? For instance, that AndyH guy said each core was modified to have its own FPU, but it seems Jaguar already has an FPU per core (correct?), they're just not so stellar, which led to some saying they're upgraded instead. However, what if it's based on vishera, like an 8350? Then the comment makes more sense. Say an 8350 modified for an FPU per core, downclocked to 1.6GHz? At 1.6 even vishera shouldn't be the heat/watt monster we know it as, and given the smaller GPU CU count, they might have room.
Granted, to me, that belies what we've heard about the 32MB of eSRAM not having ROPs. Going off of ERP's comments of how the eSRAM might be used to lower latency of reads/writes/misses and thus speed up compute on the CUs, wouldn't that suggest that MS WANTS to do heavy duty stuff on the CUs? Seems redundant to customize both CPU & GPU for heavy duty physics, no? Maybe, MS has seen that the speed up is great enough that they can have the GPU essentially run double duty, pushing all the physics calculations you'd want through the 12 CUs so quickly that it doesn't have a sizable impact on the time spent rendering the scene. Additionally, this utilizes the 32MB of eSRAM in a way faintly similar to the way IBM's power 7 processors are set up, which might explain the "like a super computer" rumors that popped up for the Durango.
As to the 14+4 issue surrounding orbis, what if it's a result of legacy design? That is to say, what if orbis was originally an APU + GPU. It would help explain the CPU context switching patent by Sony. Additionally, Arthur Geis (the ageis poster from GAF) mentioned that those old Durango rumors about MS using a 6670 were actually for orbis, and that people had confused them. Also, IIRC, around that time there was talk of a 2 GPU solution for Durango, but perhaps this too was for orbis. If true, these seemingly suggest an APU+GPU solution. However, somehow someway this changed (maybe they caught wind of what MS had planned?), and Sony went with a SoC instead. Now, maybe, prior to this event, Sony planned to use the CUs (let's conveniently count 4) in the APU to provide processing support for the CPU in a more integrated, HSA like setup, and so some modifications were done to say the ACEs or to add an extra SIMD unit or increase texture caches or whatever it is the latest rumors are suggesting, specializing them in some way for GPGPU work with expectation they would be under utilized for rendering with the separation. The aforementioned event occurs, and this design gets updated into a SoC, but because developers have already been working with this setup, Sony isn't able rework the designed integration of the 4CUs too radically. As a result, we have a puzzling 14+4 setup with some vestige literature suggesting these 4CUs will provide minimal rendering support, despite that it may no longer holds as they're now just more specialized CUs in an 18CU SoC GPU. This might explain why eurogamer may perceive them as separate - their documents are out of date.
Sorry for the overuse of "maybe", but I just want to highlight this as guess work. I don't want to come off poorly on my first post.
Granted, to me, that belies what we've heard about the 32MB of eSRAM not having ROPs. Going off of ERP's comments of how the eSRAM might be used to lower latency of reads/writes/misses and thus speed up compute on the CUs, wouldn't that suggest that MS WANTS to do heavy duty stuff on the CUs? Seems redundant to customize both CPU & GPU for heavy duty physics, no? Maybe, MS has seen that the speed up is great enough that they can have the GPU essentially run double duty, pushing all the physics calculations you'd want through the 12 CUs so quickly that it doesn't have a sizable impact on the time spent rendering the scene. Additionally, this utilizes the 32MB of eSRAM in a way faintly similar to the way IBM's power 7 processors are set up, which might explain the "like a super computer" rumors that popped up for the Durango.
As to the 14+4 issue surrounding orbis, what if it's a result of legacy design? That is to say, what if orbis was originally an APU + GPU. It would help explain the CPU context switching patent by Sony. Additionally, Arthur Geis (the ageis poster from GAF) mentioned that those old Durango rumors about MS using a 6670 were actually for orbis, and that people had confused them. Also, IIRC, around that time there was talk of a 2 GPU solution for Durango, but perhaps this too was for orbis. If true, these seemingly suggest an APU+GPU solution. However, somehow someway this changed (maybe they caught wind of what MS had planned?), and Sony went with a SoC instead. Now, maybe, prior to this event, Sony planned to use the CUs (let's conveniently count 4) in the APU to provide processing support for the CPU in a more integrated, HSA like setup, and so some modifications were done to say the ACEs or to add an extra SIMD unit or increase texture caches or whatever it is the latest rumors are suggesting, specializing them in some way for GPGPU work with expectation they would be under utilized for rendering with the separation. The aforementioned event occurs, and this design gets updated into a SoC, but because developers have already been working with this setup, Sony isn't able rework the designed integration of the 4CUs too radically. As a result, we have a puzzling 14+4 setup with some vestige literature suggesting these 4CUs will provide minimal rendering support, despite that it may no longer holds as they're now just more specialized CUs in an 18CU SoC GPU. This might explain why eurogamer may perceive them as separate - their documents are out of date.
Sorry for the overuse of "maybe", but I just want to highlight this as guess work. I don't want to come off poorly on my first post.
Love_In_Rio
Veteran
It will be funny if in the end MS - that still is not a "known" member of the HSA foundation - embraces homogeneous computing instead of heterogeneous after being a long time partner of AMD.
- Status
Similar threads
