• Hey, guest user. Hope you're enjoying NeoGAF! Have you considered registering for an account? Come join us and add your take to the daily discourse.

VGleaks: Orbis Unveiled! [Updated]

i-Lo

Member
29% is not minor.

To me it is not, and perhaps the same to you. However, I did not write the article for VGleaks and thus I am only guessing what the author may have considered "minor" while writing the article, given the objective figures in question.

0405_qqq6q.gif

You know at this point, I feel like people who ask questions for which answers have been given (and no this is not a hyperbole) at least over a few dozen times should just be added to ignore list because they could not be arsed to do a little research.

Lol, I'm not a "master race" guy, system ram is usually unrelated to gameplay, but when did I inquire about specifically game playback?

Oh no, I just thought you may know something. I am genuinely curious and you seem to be a knowledgeable guy. This question is open to anyone who would know for certain.
 

Razgreez

Member
btw what are ALU's?

ALU are arithmetic logic units i.e. they perform math. Don't worry about the rest just consider them to be the ones that do the actual work. What people seem to be missing here is that each AMD CU usually contains 64 ALUs but, more importantly, one scalar ALU. This scalar ALU is specifically designed to execute "one off" or SIMD inefficient code. In order to give developers better control of a CU this would either need to be disabled or made programmable

If made programmable they could possibly "default" back to normal scalar ALU functions making the CU as powerful as normal CUs for rendering. If completely disabled they become somewhat less rendering efficient
 
Oh no, I just thought you may know something. I am genuinely curious and you seem to be a knowledgeable guy. This question is open to anyone who would know for certain.

To be honest? It's hard to tell what game does what with the system ram, some games debug tools will tell you how much system memory is being used for the game. I think Crysis is one of those. I'm guessing a lot of those is caching? No idea to be honest.
 

i-Lo

Member
ALU are arithmetic logic units i.e. they perform math. Don't worry about the rest just consider them to be the ones that do the actual work. What people seem to be missing here is that each AMD CU usually contains 64 ALUs but, more importantly, one scalar ALU. This scalar ALU is specifically designed to execute "one off" or SIMD inefficient code. In order to give developers better control of a CU this would either need to be disabled or made programmable

If made programmable they could possibly "default" back to normal scalar ALU functions making the CU as powerful as normal CUs for rendering. If completely disabled they become somewhat less rendering efficient

Do you reckon this is what may have been done to those four CUs (given they still yield 410GFlops overall) leading VGleaks to make the "minor" comment?

To be honest? It's hard to tell what game does what with the system ram, some games debug tools will tell you how much system memory is being used for the game. I think Crysis is one of those. I'm guessing a lot of those is caching? No idea to be honest.

Thanks nonetheless.
 
considering the overall perfomance of the ps3 is 30% more than the x360 it would be minor.

That is different. PS3 may have a flop advantage(with RSX too), but 360's GPU is more advanced and is clearly more powerful than RSX at the majority of tasks. Cell is definitely more powerful than Xenon, but its not simple using those advantages, where it is simple to use Xenos's advantages.

Anotherwards its a completely different situation, and this Orbis/Durango situation is much more apple to apples, and if there is a literal 30% advantage here your much more likely to see it.
 

Subtle

Member
So can someone tell what this means? Thread is moving too fast. No Dragonball power levels. Digimon power levels would be appreciated :)
 
This is the way I read the original vgleaks post earlier. I TRIED to ask about it twice cause I noticed it was never brought up, but no one noticed my post

IMO this is what the post says. Let me point out how its basically just how I THINK you should interpret the original post correctly and is not just speculation.

Each CU contains dedicated:

- ALU (32 64-bit operations per cycle)
To me, this means every CU has at least one ALU.

- Texture Unit

- L1 data cache

- Local data share (LDS)

About 14 + 4 balance:

- 4 additional CUs (410 Gflops) “extra” ALU as resource for compute Now two this is like an addendum to the above bolded information, and it means 4 of these CU has a total of TWO ALU's. Hence theres you secret sauce. Or more literally 4 CU's of the 18 are "special" and packing an little extra step in there swagger.

I have yet to read speculation about this on b3d, but will do now as it seems they may share the same train of thought and have a lot more technical know how then me. This IMO this was just basic English comprehension and paying attention to details in the way this updated info was posted(it is written below the original description of the CU's and is indented).

btw what are ALU's? I dont even know what significance this would hold. I'm assuming it just makes those 4 CU better suited for compute processes as they have 2 instead of 1. But what are they and what do they do?

Hmm... that's interesting... could mean something significant if true.
 

thuway

Member
so for those of us who havent read the whole thread, are the specs worse or better than the rumors?

We have found out that a portion of Orbis's GPU will be customized for compute (animation, lighting, physics, etc.) - typically things done on the CPU. This is essentially the promise of the Cell Processor but delivered in spades and multiple times over. (GPGPU + HSA)

Its better.

There are also idiots on the forum who are taking the above fact and spinning it into something negative. Little do they realize, that if the GPU had to do those certain tasks, it would cripple its performance.
 
We have found out that a portion of Orbis's GPU will be customized for compute (animation, lighting, physics, etc.) - typically things done on the CPU. This is essentially the promise of the Cell Processor but delivered in spades and multiple times over. (GPGPU + HSA)

Its better.

There are also idiots on the forum who are taking the above fact and spinning it into something negative. Little do they realize, that if the GPU had to do those certain tasks, it would cripple its performance.

what are your thoughts on what I posted above, and what would it mean?

Memory bandwidth is important too. That measures how much data can be read/written to ram in a second. The PS4 (Orbis) is supposedly using very fast GDDR5 memory. The Wii-U is using really slow DDR3. So while it seems like 4GB vs 2GB isn't that big of an advantage, when you factor in how much faster the data can be moved around, PS4 has a decisive memory advantage over Wii-U.

Yea it may only have 2x the RAM, well actually for games it has 3.5x the amount, but it has almost 14x the bandwidth(this doesn't factor in Wii U edram tho).
 
RAM total is only 2x WiiU ?

Yet people still say the difference between Wiiu and ps4 will be similar to Wii and ps3 when ps3 has like 6x the ram of Wii.

Memory bandwidth is important too. That measures how much data can be read/written to ram in a second. The PS4 (Orbis) is supposedly using very fast GDDR5 memory. The Wii-U is using really slow DDR3. So while it seems like 4GB vs 2GB isn't that big of an advantage, when you factor in how much faster the data can be moved around, PS4 has a decisive memory advantage over Wii-U.
 
We have found out that a portion of Orbis's GPU will be customized for compute (animation, lighting, physics, etc.) - typically things done on the CPU. This is essentially the promise of the Cell Processor but delivered in spades and multiple times over. (GPGPU + HSA)

Its better.

There are also idiots on the forum who are taking the above fact and spinning it into something negative. Little do they realize, that if the GPU had to do those certain tasks, it would cripple its performance.

so basically we can expect to see phys-x effects in orbis games?
 
so basically we can expect to see phys-x effects in orbis games?

What it means is the devs have choices on how they want to use it. It seems to be designed in a way to suggest to devs a certain way to use it(compute, but even then it doesnt have to be for phys-x type effects), but they dont have to and still could use it for rendering. Its like an RPG. You can spec your warrior to be a tank and take damage, or you try an repec him to be more damage dealing if you want lol.

Its gonna come down to the dev and the game.

Imo, the missing parts to the picture in the case of Durango must be some physics oriented hardware as well. I suppose they have included something to that effect, alleviating the burden of the GPU, which would make Orbis and Durango at parity (similar GPU output, GPGPU capabilities of the same order, different but ambitious memory configuration).
No way does the 1.2 rumored TFlops of the GPU account for that...

Maybe. One would hope for Durango's sake. Everyone is just hoping at this point since the vgleaks article was less detailed. Orbis would still hold a slight GPU advantage, and a large bandwidth advantage, while Durango would hold a total memory advantage. Like most with technical know how have pointed out(T Lottes) bandwidth>more memory, but i could see some certain game if exclusive to 360, putting that extra memory to use(a few years down the line as the generation evolves.).

Now to me that all means they still have there advantages and disadvantages, with Orbis having a slight overall advantage. Almost parity as you put it. BUT I think the real differential will come down to how much OS overhead Durango really has. Memory(2 or 3GBs?), CPU cores(1 or 2 cores? will be at least 1), CU's(i know this never has been rumored but it could happen depending on what this system is gonna be doing, toaster, frying eggs, time machine ect ect) all could be affected.

To me it all harkens back to that outdated 2010 MS doc. System overall power was said to be 8x 360, but games would be 6x 360 with 2x being taken by other non releated gaming applications. Now obviously things have changed, and I think Durango specs have gone up since then, but the overall objective for Durango remains the same(as in now it may just be 10x 360, with 8x to games, u get the idea).
 

Pistolero

Member
Imo, the missing parts to the picture in the case of Durango must be some physics oriented hardware as well. I suppose they have included something to that effect, alleviating the burden of the GPU, which would make Orbis and Durango at parity (similar GPU output, GPGPU capabilities of the same order, different but ambitious memory configuration).
No way does the 1.2 rumored TFlops of the GPU account for that...
 

Perkel

Banned
so basically we can expect to see phys-x effects in orbis games?

if rumors are true. Yes. and with better effect than Nvidia psyx. Still games will look better than Durango. Also someone mentioned that those 4 CU are a bit different so they may be even more advanced than other normal CU.
 

mrklaw

MrArseFace
unlikely. The chips are rumored to have unbelievably low yields.

Really? But with no edram on the chip, what's so complicated? 20CUs (with 2 disabled leaving 18 to use) isn't huge and jaguar doesn't add much to the size.

What are the yields on something like a 7850/7870?
 

deanos

Banned
Imo, the missing parts to the picture in the case of Durango must be some physics oriented hardware as well. I suppose they have included something to that effect, alleviating the burden of the GPU, which would make Orbis and Durango at parity (similar GPU output, GPGPU capabilities of the same order, different but ambitious memory configuration).
No way does the 1.2 rumored TFlops of the GPU account for that...
prepare yourself for disappointment. there is no parity.
 
We have found out that a portion of Orbis's GPU will be customized for compute (animation, lighting, physics, etc.) - typically things done on the CPU. This is essentially the promise of the Cell Processor but delivered in spades and multiple times over. (GPGPU + HSA)

Its better.

There are also idiots on the forum who are taking the above fact and spinning it into something negative. Little do they realize, that if the GPU had to do those certain tasks, it would cripple its performance.

Isn't the Cell a better SIMD processor? It's basically what it's designed for. You might as well call it a SIMD processor. Hell, it supports up to 128-bit SIMD operations while those ALU's are limited to 64-bit.

SPE - SIMD Processing Unit (yes, the 'S' stands for Synergistic, but that's just PR talk).
 
Imo, the missing parts to the picture in the case of Durango must be some physics oriented hardware as well. I suppose they have included something to that effect, alleviating the burden of the GPU, which would make Orbis and Durango at parity (similar GPU output, GPGPU capabilities of the same order, different but ambitious memory configuration).
No way does the 1.2 rumored TFlops of the GPU account for that...

Your mistake is that you assume there is parity, and are searching for evidence to support that conclusion. That's completely backwards. Based on the evidence we have there is no parity, nor do we have any reason to conclude there is any undisclosed physic hardware that has never even been rumored. Perhaps you should make peace with the possibility that Orbis is simply more capable.
 

gofreak

GAF's Bob Woodward
I think the B3D speculation is along the lines of what JohnnySasaki86 is saying.

Not that there are 18+4 CUs.

Still 14+4 CUs.

But the theory suggested is that the 4 aren't split out under a different threading regime.

Simply that they have extra resources, and THOSE resources offer a minor boost for rendering performance compared to a normal CU. But those tweaks really target compute jobs (and presumably the hardware threading would prioritise compute jobs on those CUs).

'Hardware balanced', it is suggested, might be poor English for 'normal hardware' in the 14 other CUs, balanced for regular rendering.

It's a totally reasonable read of what vgleaks reported. It would put us a bit more back to square one on the GPU side - code would be able to transparently leverage the 18 CUs for rendering - but with knowledge that a few of the CUs are tweaked for better compute performance.

The only thing that doesn't jive with all this, and that might jive more closely with the idea of the 4 CUs working under a different (manual?) threading system, is Eurogamer's report which said the 'compute module', 'is not a part of the main graphics pipeline'.

I guess we might get clarification down the line once Sony starts talking about it at developer conferences and things. Not sure if vgleaks is in a position to clarify this (?) They might have all the docs, but maybe they haven't parsed out what exactly this means. Kind of frustrating to think they or Kotaku might have all this info amongst their PDFs but either don't fully understand it or don't want to share all the details... :)
 

mrklaw

MrArseFace
Pretty interesting convo. So the 4 CU's have an extra SIMD? That's the theory proposed anyway. What do GAF'ers techies have to say about this?

I don't think that is doable - a CU is either carrying out graphics or compute tasks. It'd be pretty complicated to change the internal layout of a CU just to add an extra SIMD unit, plus e architecture is designed to schedule graohics *or* compute tasks to each CU, not both at the same time. I don't think the SIMD units are even directly addressable.

I think B3D are over thinking it. The description was simple - 14+4 = 18, hardware balanced for 14 (so ROPs etc are based on 14 being used for graphics), and CUs come in groups of 4 with shared cache etc so 4 for a 'compute coprocessor' sounds good.

The 'minor' increase in graphics is just a word they are fixating on. You could use it for graphics but because it wouldn't be tied into the main pipeline you would need to control it more manually and so maybe get less efficient results from it than if it was purely 18CUs for graphics. Most likely if it isn't being used for physics etc it'll be use for the same kind of tasks that the CELL used to take off the RSX. That worked quite well at letting the RSX punch above its weight.
 

Pistolero

Member
Your mistake is that you assume there is parity, and are searching for evidence to support that conclusion. That's completely backwards. Based on the evidence we have there is no parity, nor do we have any reason to conclude there is any undisclosed physic hardware that has never even been rumored. Perhaps you should make peace with the possibility that Orbis is simply more capable.

I have no stake in the outcome, believe me! But people in the know suggested as such, and heavily may I add, to the point that dismissing the more probable end result would be a form of...neglect. I might be wrong, of course, but I doubt it...
 

TheOddOne

Member
Your mistake is that you assume there is parity, and are searching for evidence to support that conclusion. That's completely backwards. Based on the evidence we have there is no parity, nor do we have any reason to conclude there is any undisclosed physic hardware that has never even been rumored. Perhaps you should make peace with the possibility that Orbis is simply more capable.
I don't know how you could be so overly defensive over his post, because he asks a valid question. What if roles were reversed? Imagine Durango info was released first and everybody started to champion it for being more powerful, while we know very little about Orbis? Would you not call that premature? Seems unfair to throw that guy under the bus and basically dismiss his points.
 
My guess would be that the 4 CUs are simply on a different part of the datastream highway (or think of it as a factory line).

Consider that for a PC, the CPU and the GPU are basically miles apart, and connected through a very limited bus, that doesn't allow for very efficient two-way communication (lots of lag, limited bandwidth), to the point where most games barely burdened the CPU at all until recently, and many just ran on GPU almost completely.

So far, consoles usually have had a better setup where CPU and GPU components can work together more efficiently. For instance in PS3, the Cell processor's SPEs are closer to compute units and could do all sorts of things like physics, culling, animations etc., and as the GPU could post its framebuffer back to the main memory, also post-processing jobs (like MLAA as probably the famous example)

For modern GPUs, I understand that the GPUs work well because they do things massively paralel, that requires a certain high latency caching setup.

If you move 4 of those CUs to the CPU side, then you can do a lot of physics and pre-processing in the more 'low-latency' section of the pipeline, which would be far more efficient.

I fully expect this to be similar in Durango, by the way.
 
I don't know how you could be so overly defensive over his post, because he asks a valid question. What if roles were reversed? Imagine Durango info was released first and everybody started to champion it for being more powerful, while we know very little about Orbis? Would you not call that premature? Seems unfair to throw that guy under the bus and basically dismiss his points.

It's basic skeptical inquiry...evidence is surely scant on both sides, but now we have a clearer picture on Orbis, and a less clear picture on Durango. Until the picture is clearer on Durango, comparisons seem a bit crap at the moment.
 
The only thing that doesn't jive with all this, and that might jive more closely with the idea of the 4 CUs working under a different (manual?) threading system, is Eurogamer's report which said the 'compute module, 'is not a part of the main graphics pipeline'.

I guess we might get clarification down the line once Sony starts talking about it at developer conferences and things. Not sure if vgleaks is in a position to clarify this (?) They might have all the docs, but maybe they haven't parsed out what exactly this means. Kind of frustrating to think they or Kotaku might have all this info amongst their PDFs but either don't fully understand it or don't want to share all the details... :)

My theory on that, is this is already pretty confusing and unclear, and that could explain why they interpreted it the way they did(if they were already told some complicated unclear info from there source. It like that game telephone lol). If you read their words on this "compute model" they seemed pretty unsure about what they were talking about. Thats why I didn't pay much attention to that whole part once this info was revealed.

To me, It look like vgleaks has just copied info directly off of some doc into there article and there wasn't much interpretation involved.

Who knows, because the 14+4 thing implies some form of separation as well. I was thinking they were separate in the sense they were modified and therefor upgraded compared to the other CU's. I dont know if its not part of the rendering pipeline or not(hell I dont even know what constitutes that) but we know they can be used for rendering. Maybe there "separate" in the way the programmer can access them and how they go about using them, but hardware wise all 18 of them are together like normal. "not part of the rendering pipeline" may be more of a software reference than a hardware reference?

Regardless, just like you said in your last comment, we'll probably just have to wait to get further clarification down the line once Sony start revealing the information to developers.
 

TheOddOne

Member
It's basic skeptical inquiry...evidence is surely scant on both sides, but now we have a clearer picture on Orbis, and a less clear picture on Durango. Until the picture is clearer on Durango, comparisons seem a bit crap at the moment.
Like I said, it was a very valid question. I personally believe Orbis will be stronger, but I think Durango can hold its own. Now we just need all the info on the table to see how powerful each one will be.
 

imjust1n

Banned
I think of the wiiU as a net book with a decent graphics card. The orbis I think of high end pc with out the baggage (meaning) core cpu power and high memory and ext ext. I think of the new xbox a step below the orbis. And pc gets higher and higher. Steam box a better innovated pc gaming experience
 

deanos

Banned
comment from B3D:
According to these rumors, Orbis is basically an APU + GPU design, but integrated into a single SoC instead of a MCM or SiP, which allows for minimum latency and maximum bandwith at the same time.

On the APU side we have the eight Jaguars combined with 256 GCN shaders, together burning an incredible amount of 512 GFLOPS! Just for the record: An Intel Core i7 4770k desktop CPU (only) delivers 448 GFLOPS. Imagine what this could mean for AI, animations, physics, etc. On the GPU side we have a processor that is basically a small HD7850 Pitcairn. In my eyes this GPU is strong enough to deliver enjoyable graphics for a next gen gaming console, especially when keeping the perverse TDP of modern high end cards in mind. We're most likely talking about 3rd gen HSA for Orbis, which means a unified adress space for CPU and GPU(s), pageable system memory for the GPU(s) with CPU pointers, and a fully coherent memory between CPU and GPU(s). Simply put, no copy work between the CPU and the two iGPs. This will be a hell of a speedup compared to a modularly designed PC (take a look at the superb 28nm Temash SoC rendering Dirt: Showdown at 1920x1080 with 5W!!!)

The reasons for going with APU + GPU are better programmability and lower latencies. You can look it up in this slide from the 2012 Fusion Developer Summit. You don't want your GPU to be saturated because of both GPGPU algorithms and graphics tasks. A single (AMD) GPU can't handle either task at the same time, which would end in a lot of headache for the programmer. AMD names two solutions for this problem: You can wait for the 2014 20nm feature Graphics Pre-Emption, or you just use an APU dedicated to computing together with a second GPU dedicated to graphics rendering. Sony is doing the latter, obviously.

It seems as if the Orbis rumors are getting more and more specific. The difference between this leak (8 x Jaguar/ 256 GCN SPs APU + 896 GCN SPs GPU) and the last leak (8x Jaguar CPU + 1152 GCN SPs GPU) is a much better balance between the computing power and the graphics power. Eight Jaguars for a Pitcairn seemed a bit underpowered, anyway. But one thing is missing in this leak: There is no dedicated DRM hardware at all, neither ARM nor SPE. It would really surprise me if they are launching without a proper in-hardware DRM.
http://beyond3d.com/showpost.php?p=1699631&postcount=187

this 14 + 4 approach sounds better by the day
 

Santar

Member
I was really really hoping it'll have bc, if it doesn't Sony should really hope the Duraongo doesn't either or they'll look pretty bad in comparison.
 

UrbanRats

Member
We have found out that a portion of Orbis's GPU will be customized for compute (animation, lighting, physics, etc.) - typically things done on the CPU.

So ehm.. i've read you make this distinction several times in this thread, and then some people (spongebob) differentiated between those and "graphics".
Now, being completely ignorant on the subject i'm wondering: aren't those elements (lighting, animations etc) a big part of what makes "good graphics"? Why the distinction? What else there is to calculate, that doesn't go under that computing specification? Only textures, triangles and shaders?
 
Top Bottom