• 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 don't understand the need for reserved CUs. GPGPU is perfectly doable on a classic GPU, without the need to have dedicated units for that.
In the end even the orbis GPU might be used for computational stuff, and so is the Durango one.

I think what will really matters is the number of CUs. Having a part of them arbitrary assigned to a task may come off as a constraint more than anything else.

They are probably not reserved like that. Usually all the components of a gpu scales with the compute units. This could very well mean that scaling everything to have a 18CU gpu would be too expansive, and so they decided to scale all the other parts of the gpu to 14CUs and provided extra float point performance that wouldn't make much difference for graphics, but is useful for other tasks.
 

Spongebob

Banned
Wow... people are explaining it and you're not getting it. If you want to PLAY games - you need more than just "graphics". 1.2TF for Durango is used for graphics and "else". With Orbis - 1.4TF can be used for JUST graphics (if chosen) and the CUs to be used for "else".

A portion of that 1.2TF will have to be used for "else". SO let's say that 400GF are used for "else" in a single video game. That means that Durango will use 800GF for "graphics" and Orbis will use 1.4TF for "graphics". What, about this simple math, do you not get?

Is any of this sinking in? I hope because you are now being ignored, junior.

I know this.

I'm also taking into account what Proelite is saying about Durango's FLOPS being greater than Orbis's.

With all due respect, given your post history, it's a fair assumption, yes.

Edit: sorry.
 
They're not being reserved.

They just have to be used more explicitly in your code. If I'm writing my rendering code, I make library calls, and the library and hardware take care of the rest. The GPU splits the work up per vertex or per group of fragments, or whatever, and schedules the threads and executes them.

These 4 CUs won't operate under that regime. They won't mix in 'automagically' to the work you do through the graphics libraries. You'll have to set up the work explicitly, separately, for those 4 CUs.

The work could be whatever you want that a CU can do - graphics or GPGPU. So they're not being 'reserved' per se, they're not being prescribed for any one particular use. However the motivation for splitting the CUs out of a hardware managed regime is so that if you want to do GPGPU, you don't have to leave the scheduling and context switching up to a hardware balancer that might not do things optimally when mixing graphics and compute tasks.

"I think"

Interesting. So devs would get more flexibility. Glad it's still 18 CU's
 

FourMyle

Member
4 additional CUs

Vn3aIlk.gif
 

Lord Error

Insane For Sony
That is exactly how I feel.
No matter what CPU they chose, it would not fare well in compute tasks. Without those, you'd be getting games that behave literally like what you're playing now, only in higher resolution and even better framerate than what you'll been getting here. To me that's far less interesting than games that will finally be able to use larger scale fluid simulations during gameplay and other stuff like that.
 

Codeblew

Member
Are people clinging to the $400 price tag, just so they can post negativly when this thing launches at a higher price?


I am expecting this thing to be $499. Sony won't do $599 again, that is for sure. $399 seems low but I would be happy with that. I am buying it day 1 no matter what the price is.
 
Am I missing something here.

VGleaks mentions that only a minor gain in rendering performance will be gained despite a 29% increase in the # of CUs being dedicated to rendering. From that it's reasonable to assume that these other 4CU aren't as capable in their ability to render, right?

VGleaks may know the specs, but I think you are putting a bit too much faith in simple wording. You are basing your whole argument on the word choice they use, while others are actually discussing the issue using potential scenarios....

Also, it seems the FLOPs for both systems are fairly well known, unless things have changed. Both use AMD, so they are directly comparable ie. how can they be different in Durango unless they were using Nvidia?
 

Razgreez

Member
Am I missing something here.

VGleaks mentions that only a minor gain in rendering performance will be gained despite a 29% increase in the # of CUs being dedicated to rendering. From that it's reasonable to assume that these other 4CU aren't as capable in their ability to render, right?

As i stated earlier i assume the Scalar ALUs are likely disabled making them quite a bit weaker (less efficient) at general graphics processing
 

Spongebob

Banned
VGleaks may know the specs, but I think you are putting a bit too much faith in simple wording. You are basing your whole argument on the word choice they use, while others are actually discussing the issue using potential scenarios....

I know this.

I'm also taking into account what Proelite is saying about Durango's FLOPS being greater than Orbis's.

.
 
Mock the Jaguar all you want, but it had Tempest 2K and a good Aliens game. Will the PS3/360 be able to say that? And you could call France on the controller.

Cool story bro.

You're the guy that ran through the Durango leak threads claiming it was nothing more than a DVR. Your post history doesn't lie.
 
Too much RAM??

You can never have too much RAM. If anything the RAM on orbis is frankly a complete let down.

Within 2 years the ps4 will be bandwidth starved and devs would be back to making compromises again.

Anything less than 8GB is frankly a dissapointment


In this case the type of ram is more important than the amount, yes it's 4 gb of RAM, but it's also really fast, that's more important than the amount
 

Perkel

Banned
Regarding the 14 + 4 CUs

(not 100% sure)

VLIW is rather inefficient when it comes to anything else than pure GPU tasks. So a GPU with a VLIW architecture wouldn't perform that good on tasks like physics, etc. So what if the 14 CUs are indeed based on VLIW architecture (Southern Island is not - so maybe the Liverpool GPU is older and does not feature GCN).

To make that up somehow AMD designed or put together 4 extra CUs for those tasks in a SIMD architecture. I just wonder why not just take a pure Southern Island GPU which is all SIMD and leave it at that.

So basicly I guess the GPU could be "old" the extra CUs are a newer architecture to have better GPGPU possibilities.

As they are not as efficient for physic as SIMD they will give a big boost for calculations.

Before leak i wondered if they could include hardware for physic like Ageia before Nvidia bought them did. 4CU for CPU units is essentialy that hardware.

Games like GTA4, Red Faction Guerilla and other physic based would fly on something like this.

I fully expect to see things like

THIS

used as standard for orbis.
 

Plinko

Wildcard berths that can't beat teams without a winning record should have homefield advantage
I am expecting this thing to be $499. Sony won't do $599 again, that is for sure. $399 seems low but I would be happy with that. I am buying it day 1 no matter what the price is.

In what alternate universe is $399 "low"?
 

sangreal

Member
Am I missing something here.

VGleaks mentions that only a minor gain in rendering performance will be gained despite a 29% increase in the # of CUs being dedicated to rendering. From that it's reasonable to assume that these other 4CU aren't as capable in their ability to render, right?

they're capable but apparently segregated from the load balancing hardware. they're not there to render, but to perform gpgpu tasks without bogging down the graphics pipeline. presumably, these tasks will also still need to be executed on durango somewhere
 

TronLight

Everybody is Mikkelsexual
As they are not as efficient for physic as SIMD they will gave a big boost for calculations.

Before leak i wondered if they could include hardware for physic like Ageia before Nvidia bought them did. 4CU for CPU units is essentialy that hardware.

Games like GTA4, Red Faction Guerilla and other physic based would fly on something like this.

I fully expect to see things like

THIS

used as standard for orbis.
Please do this.

At least in first parties, I want cloth physics everywhere.
 
I don't really understand why people are dissing the GPU here if it is a downclocked 7870... .

It isn't...

A 7970m is a downclocked 7870.
If orbis gpu had all 20 CUs it would be a downclocked 7970m (so a downclocked downclocked 7870).
But it isn't that either, it has only 14 CU's for rendering out of those 20, and 4 for GPGPU stuff like physics).

So no, it isn't a downclocked 7870, much closer to a 7770 more likely.
 

Durante

Member
I've wondered if perhaps maybe you can give those guys fast path/interface to the CPU's L2 cache so that the communication between CPU and CU could be super low latency and very high bandwidth.
Yes, that would be one idea. As I said earlier, if you do anything in hardware to make those 4 CUs more suited for general purpose computation, I can totally see the reason for the fixed split.
 
We've said this over and over and over again. Orbis and Durango will be good at different thing. Don't be foolish to think one is better than the other in all frames of work. However, this news about CU is something that makes Orbis EVEN better. People are morons. You lot are trying to spin something positive into a negative.

This is akin to Cell processor, that actually is worth hyping. Fuck is wrong with you GAf.

okay, I'm confused. so the gamecube / xbox comparison no longer stands?

surprised a developer would get it so wrong, unless they hadn't used the extra horsepower.
 
Maybe not. According to what we know - physics, AI, etc will have to be gouged from the 1.2TF performance of Durango's GPU, limiting what can be done with graffox - whereas on the Orbis - it has it's own dedicated unit so the entire 1.4TF can be dedicated to graphocks.

There will be a difference. Of course, what can change in the future is the MS "special sauce" which, I contend, is nothing more than a continuation of the "jump in" PR campaign.

I'll say it again: PR > Specs

I don't think AI would be something you would like to put on the gpu.

Wow... people are explaining it and you're not getting it. If you want to PLAY games - you need more than just "graphics". 1.2TF for Durango is used for graphics and "else". With Orbis - 1.4TF can be used for JUST graphics (if chosen) and the CUs to be used for "else".

A portion of that 1.2TF will have to be used for "else". SO let's say that 400GF are used for "else" in a single video game. That means that Durango will use 800GF for "graphics" and Orbis will use 1.4TF for "graphics". What, about this simple math, do you not get?

Is any of this sinking in? I hope because you are now being ignored, junior.

That's not exactly how it works either. Physics and other non graphical tasks don't just occur in a vacuum completely unrelated to graphics. Having split resources like this means that any task the 14CUs could perform that are tied to the physics is going to wait for these 4CUs to finish their jobs, and their throughput is going to be much lower than if those 14 CUs could take the job...

If the 12 CUs from durango setup are more advanced for example and have no trouble switching the context from a compute job to a graphics one you could potentially speed up the compute job (12 CUs vs 4) and perhaps even end up having more time to do the graphics tasks...

I have no idea if that's the case, but switch compute with vertex shaders and graphics with pixel shaders and you pretty much describing the Xenos/RSX situation.
 

Spongebob

Banned
they're capable but apparently segregated from the load balancing hardware. they're not there to render, but to perform gpgpu tasks without bogging down the graphics pipeline. presumably, these tasks will also still need to be executed on durango somewhere

Why does everyone think that I forgot about this?

Everything that I was saying took into account Proelite's comments.
 

sono

Member
Sounds awesome.

Credit card limit raised - check.

Fabulous Graphics potential - check.

Will it run a web browser or is that just a software issue ?
 

lantus

Member
So relative to today's newest hardware on PC, how old would we say this PS4 hardware is? I think that is a good barometer to where they would price this at. I'm fully expecting a maximum of $499.
 

yurinka

Member
Going by Atari math, that's going to be a 512bit console! Orbis and Durango ushering in the 512bit generation!
8 Jaguar 64 bit cores. According to Microsoft's Polygon you get a 9th core for free purchasing the Halo 4 pizza at Pizza Hut, Doritos and a couple of Mountain few.
 
In what alternate universe is $399 "low"?

I think that was in regard to specs, not consumer's economic standards.

Personally, those looking for a price point under $399 are going to get a wake-up call and shows how long of a generation this was. There's development, marketing, manufacturing, and distribution costs to take into account with the initial price point. Then profits need to come in for awhile for fueling bonuses and ensure the company actually benefits from the product. Once all of those are in order, then we'll see prices go down.
 
Top Bottom