NGGP: NextGen Garbage Pile (aka: No one reads the topics or stays on topic) *spawn*

Status
Not open for further replies.
What I find interesting is the Orbis leak mentions Jaguar explicitly while the Durango doc doesn't, just says x86. Perhaps the Durango CPU is more customized. Someone suggested you could upgrade the CPU's from one ADD and one MUL unit to two FMA's instead, this way you could effectively double the flops. I figure the L2 and memory couldn't support the bandwidth, but by using the embedded RAM, maybe that alleviates the bandwidth issue.

:yes:

That's quite possible, like VMX units on Xenon.
 
:yes:

That's quite possible, like VMX units on Xenon.

Aegies seems to agree with you on differences between the Durango CPU and Orbis one. Apparently, the Durango is kind of more special?

I keep reading Durango´s hardware is far more custom than Orbis. I hope this leads to a better MS system. The rumors around it are not pleasing.
 
Which is the case for Durango, no? The question is now if the performance hit to these extra 4 CU's is made up by the efficiency improving silicon in Durango.

edit: aegis is commenting on the recent rumors:

https://twitter.com/aegies/status/296320251983953920
https://twitter.com/aegies/status/296320930555236352


He also says the details are incomplete and we're still missing extra hardware.

Yeah though there maybe more to modifications on the CPU side for Xbox 3 as pointed out by McHuj, because like Proelite mentioned everything detailed seems to only be about efficiency for Xbox 3, not some missing CUs that aren't being talked about. And my take based on everything so far, these 4 CUs may not perform at 100% if used for rendering. So could their performance be 90%? 50%? And then if used in that manner you're looking at what seems to be a "plain" 8-core Jaguar for a CPU vs a possibly modified CPU in Xbox 3.

I'm still struggling to see how Durango would come close to the performance of Orbis. Hypothetically, I could see a 1.2 TF GPU closing the gap with 1.4 TF through an improved ISA and embedded RAM thanks to efficiency gains. That's only a 15% or so difference, but the other compute capability. I'm not sure can be close easily.

This is why I brought up GPU compute usage not being standard in game development. Looking at PS4's design it would have a big advantage in this area compared to Xbox 3. But what good is this advantage if it's not used? Maybe down the road GPU compute will be more of a standard in development and Sony is banking on that, but as of now its potential seems like it could be a wasted primarily in 3rd party efforts. And because of that, PS4's advantages are negated somewhat making the two more comparable.
 
Last edited by a moderator:
What could they have done to the 4CUs that would cripple them for rendering tasks? Surely anything the would have done would have been for the benefit of the system, yet it seems Orbis has gone down in some peoples eyes after this leak.
 
Basically for Orbis you want the dev to code any and everything physics or effects simulation related for the 4 CU compute part, kind of like how there was a lot of coding for SPE's to do those things this gen. But I gather the most complicated stuff we've seen this gen would easily fit in that 4 CU so creating any game in the same vein as current gen would see that 4 CU go underutilized.. or neglected. So devs will need to use some imagination to run more things off of it to utilize the system more efficiently. But at least this time it'll be a lot easier to implement compared to CELL, so I think devs will eventually get around to that being the norm. That frees up the CPU for the game loop, AI stuff and whatever else usually done there.

So essentially the weak CPU got freed up a bit, and so did the 14 CU GPU (if you wanted to do compute on it since 18 CU doing rendering and compute at once has high latency), but at the cost of chopping off 4 CU which can't do rendering that well anymore? And like bg said is it 50% or 90% or whatever % capable now. This might be due to having to use it in a hybrid crossfire mode.

Then Durango apparently now has a beefed up Jag? So that'll handle the physics and simulation routines better, and then you have the low latency and efficient GPU using DME to try to make up for this 2 CU deficit and bandwidth disparity.
 
Are we sure they cant do rendering related tasks very well though? Surely being physically the same or very similar as the other CUs in the system means they should still be relatively good at working on the same kind of task? What could they have removed from these CUs that would make this not the case?

Could it just mean that the management of tasks across the two pools of CUs is less efficient so you loose a bit of rendering power over just firing tasks at all 18 CUs in a non split config, but Compute functions see an improvement due to the split pool because of no contention for rescources? In this case it would make sense to me the 4 CUs are still very potent for rendering tasks but they may take more fine grained management and effort to achieve best efficiency.

To bring it back to the vs argument a little, do we think much has changed after this leak? I would have thought until we hear more about Durango things are pretty much the same or Orbis is looking maybe even a little bit better due to the 4 CU customisations, they wouldnt have done it if it was to the detriment of the system...
 
Last edited by a moderator:
There's not enough info to go off of to say for sure, or even if we're interpreting "minor boost" correctly. There a few scenarios of why and you outlined one, and I thought it might have only been referring to the extra ALU units before, but who knows.
 
What could they have done to the 4CUs that would cripple them for rendering tasks? Surely anything the would have done would have been for the benefit of the system, yet it seems Orbis has gone down in some peoples eyes after this leak.

No gone down for me, but trying to rectify the "comparable" comments.

The perspective of this view is based on the "hardware balanced at 14 CUs". This to me would indicate the hardware is designed around 14 CUs for rendering. I'm sure someone could put a more technical explanation around this, but to me this would suggest the 4 CUs are tapping into resources not designed with them in mind for rendering. This would keep them from performing at 100% of their ability. There are one or two other ways I could go with this, but I'm focusing on this one for now.
 
I guess we are just lacking the detail, which isnt helped by ambiguous wording of the leak. I would be very surprised though if the + 4 CUs were not great at rendering tasks even if not at 100% efficiency. However the increase in Compute ability if required may be well worth the tradeoff. Id be very surprised if these CUs could only be used well for compute and are otherwise idling. Even SPUs in Cell were used for graphical tasks and i expect these CUs to be significantly better at it.
 
No gone down for me, but trying to rectify the "comparable" comments.

The perspective of this view is based on the "hardware balanced at 14 CUs". This to me would indicate the hardware is designed around 14 CUs for rendering. I'm sure someone could put a more technical explanation around this, but to me this would suggest the 4 CUs are tapping into resources not designed with them in mind for rendering. This would keep them from performing at 100% of their ability. There are one or two other ways I could go with this, but I'm focusing on this one for now.

For me go down.In my mind perception it has gone from seing Orbis like a AMD 7860 to a 7830.Sony had cell!.The best compute processor in the world.20 watts in 28 nm or so!.And they retrain 4 precious CUs that on the PC world make a game like Battlefield 3 hit almost 60 fps at 1080p to fight what?.Another MS trick like VMX in Xenon?...
 
:yes:

That's quite possible, like VMX units on Xenon.

I had wrote the following post in another thread, but because it seems to be related to what you are implying, I want to share it here:

ROG27 said:
I think this generation is a flip-flop of last gen processor-wise.

I think MS went for a CPU-centric machine and Sony a GPU-centric machine.

This is in line with what each platform will be focusing on. Next Xbox will presumably focus on CPU-intensive processes (i.e., Kinect, multitasking, AI, etc.). PS4 will presumably focus on GPU-instensive processes (i.e., graphics, animation, physics simulation, etc.).

That being said, there will be logic in each that affords flexibility when needed.

I believe the 4 CU block on the GPU (specialized for GPGPU functionality) primarily represents this flexibility for Sony.

As for Microsoft, I believe beefed up (or additional) SIMD FPUs on the Jaguar CPU side primarily represents this flexibility.

To add to this hypothesis, each system has a memory architecture that is either biased towards making efficient use of the CPU (Microsoft with a large pool of slower DDR3 being balanced with a small pool of fast, extremely-low-latency, coherent ESRAM and DMEs) or the GPU (Sony with a high-bandwidth GDDR5 UMA pool).

I think the design intent for both system is pretty clear here. Knowing this, it is pretty safe to say that the Durango's "special sauce" is beefed-up (or additional) SIMD FPUs on the CPU side.

What do you guys think?
 
Wow! This is fun. I love watching how platform affinity affects peoples perception and our technical nature creates a disconnect from "big picture" kind of thinking. Upcoming console launches are always the most fun yet disturbing times around here.

Couple items just as food for thought, as you ring every extra theoretical flop from your system of choice.... If you were to graph perceived visual fidelity vs computational power, small increments of visual improvement require exponential increases in power. And so while DF and the like may be able to identify subtle performance and quality differences with their discerning powers of perception, trust me, these are lost on 95% of the gaming public. And this potential 50% or 33% (depending on how you do the math) discrepancy in theoretical flops, probably won't affect game quality for the masses enough to drive purchasing decisions as much as features will. Put another way, the perceived visual improvement between xbox1/ps2 to xbox360/ps3 will be greater than xbox360/ps3 to xbox720/ps4 even though from a computing power perspective things look to be orders of magnitude more. And if orders of magnitude jumps can't move the needle much, what's 50%?

So in the end, which is better or worse, may not be decided by chip designers and hardware engineers, but rather whomever comes up with the next great must have entertainment application and/or interface. I can see Durango supporters lamenting the loss of 3GB and 2 cores, but when you come home and your system recognizes you had a bad day from your demeanor, and tells you which of your friends is also having a bad day and recommends a movie to watch together remotely to cheer you up, and maybe a game afterwards, along with other news that can seamlessly bounce to other devices; perhaps you'll miss it less. Or maybe Orbis's touch controllers will automatically detect player heart rate and scale game intensity accordingly. So many possibilities.

Anyway. Breath taken. Please continue.
 
I don't think it's a downgrade to have 4CUs dedicated to computing and 14CUs dedicated to rendering instead of having 18CUs for both. DX11 level of graphics requires a lot of GPU computing: You need it for Depth of Field, Motion Blur, Particles, Fluids, Clothing, etc. So you would detach GPU resources to these tasks anyway. Devs can be assured that there will always be 14CUs for rendering tasks, no matter how much GPGPU tasks they're throwing at the game. Doesn't sound too bad for me. But I still agree that Sony most likely wants the 4 computing CUs to be as flexible as possible.

The cool thing about the HSA is, that you can use these effects not only for visuals (like in Nvidia PhysX games), but also for gameplay mechanics, since the HSA aims to have an ultra-fast communication between CPU and GPU elements without an annoying copy overhead.
 
No gone down for me, but trying to rectify the "comparable" comments.

The perspective of this view is based on the "hardware balanced at 14 CUs". This to me would indicate the hardware is designed around 14 CUs for rendering. I'm sure someone could put a more technical explanation around this, but to me this would suggest the 4 CUs are tapping into resources not designed with them in mind for rendering. This would keep them from performing at 100% of their ability. There are one or two other ways I could go with this, but I'm focusing on this one for now.

I´ve read a theory that makes sense the most. The balanced hardware means you can get the highest performance when dedicating those 4 CUs to assist the CPU instead of using them for rendering tasks.
 
was checking the neogaf thread - aegies recently posted on the orbis thread that the xbox 3's audio dsp was been emulated by a xeon 4core/8 thread cpu !! the devkit has 2 xeons and one of them is for jaguar emulation . He also noted out that there is a reason why the 12 cu gpu is called as 768 shader cores .

so can the gpu in durango be something from the successor of gcn/gcn2 ?
 
was checking the neogaf thread - aegies recently posted on the orbis thread that the xbox 3's audio dsp was been emulated by a xeon 4core/8 thread cpu !! the devkit has 2 xeons and one of them is for jaguar emulation . He also noted out that there is a reason why the 12 cu gpu is called as 768 shader cores .

so can the gpu in durango be something from the successor of gcn/gcn2 ?

Yes, and even VLIW.
 
Isn't it funny that someone always comes up with new fairy dust for Durango that will close the gap to Orbis?

When Orbis was rumored to have a 1.8 TFLOPS GPU, the "leakers" said that Durango will have a secret sauce for the GPU (some kind of super FLOPS) since it had a 7970 in the first devkit. When Orbis was rumored to have 4GiB of super fast GDDR5 RAM, the "leakers" said that Durango will have some wizzard jizz that will close the bandwidth gap. Now Orbis is rumored to have 512 GFLOPS on the computing side (8 Jaguars + 4 GCN CUs) and "only" 1.4 TFLOPS on the GPU side and all of a sudden Durango will have some sort of Super-Jaguar with ultra-beefy FPUs that close the gap again. What happened to the secret sauce for the Durango GPU? No longer required since Orbis was downgraded to 1.4 TFLOPS on the GPU?

All these Durango rumors sound contradictory as hell. I'm not believing any of it until someone comes out with concrete information.
 
Everybody is looking on the CUs, but Durango will be 50% less ROPs and 33% less TMU according to rumors and some math. Isn't a lot?
 
Isn't it funny that someone always comes up with new fairy dust for Durango that will close the gap to Orbis?

When Orbis was rumored to have a 1.8 TFLOPS GPU, the "leakers" said that Durango will have a secret sauce for the GPU (some kind of super FLOPS) since it had a 7970 in the first devkit. When Orbis was rumored to have 4GiB of super fast GDDR5 RAM, the "leakers" said that Durango will have some wizzard jizz that will close the bandwidth gap. Now Orbis is rumored to have 512 GFLOPS on the computing side (8 Jaguars + 4 GCN CUs) and "only" 1.4 TFLOPS on the GPU side and all of a sudden Durango will have some sort of Super-Jaguar with ultra-beefy FPUs that close the gap again. What happened to the secret sauce for the Durango GPU? No longer required since Orbis was downgraded to 1.4 TFLOPS on the GPU?

All these Durango rumors sound contradictory as hell. I'm not believing any of it until someone comes out with concrete information.

I'm a 360 fan, but there is no denying from these specs, the ps4 will be more powerful, the only thing, i can hope for is ps4 being 20% more powerful and not 50%.
 
Status
Not open for further replies.
Back
Top