Averagejoe
Regular
With this specs no a change,now durando has change from been on par to be more powerful.
NGGP: NextGen Garbage Pile (aka: No one reads the topics or stays on topic) *spawn*
- Thread starter Barso
- Start date
- Status
With this specs no a change,now durando has change from been on par to be more powerful.
"It's not enough to succeed, others must fail"
I think it's because the "special sauce" everyone was holding on to is almost gone, even the die hard fans are dropping this. The info about the DME says they won't help the graphics, (they are helping the I/O a lot though). So the only remaining way to make the graphics equal would be to cripple the PS4. It's the recent new "thing". Now it's because sony are incompetent, they have 18CU and GDDR5, but they really crippled the whole system for absolutely no intelligible reason. The specs are less, they were just hopeful targets that sony failed to achieve. They reserved the crippled CUs exclusively for the Move even in games that won't use it. So there, it will be close to 12CU performance with 18CU.
For me, the reason I think they'll be close enough, is because someone told me they would be. It's that simple. I'm looking for a technical explanation for that, and I haven't found it yet.
Bagel seed
Veteran
To the people who keep bringing up superDae, ask yourself, if Vgleaks and everyone else was saying Durango will be a 2.2TF monster beating Orbis handily, and Dae implied the opposite, would you believe him and keep bringing it up...?
Silent_Buddha
Legend
According to the VGLeaks information which everyone seems to be treating as gospel, the system has 18 total CUs capable of 1.84 TF.
According to the same source the hardware is designed/balanced such that 14 CUs are dedicated to graphics rendering while 4 CUs are reserved for "other" things.
According to the same source, when used for graphics rendering they only provide a minor benefit.
Now, you either believe all it, none of it. Or you are just picking and choosing what you choose to believe.
What we don't know is how that division is taking place. Is there some sort of physical separation (this would go to easily explain why their contribution to graphics rendering would be "minor")? Are 4 CUs somehow so different as to make their rendering contribution less significant? Is it just a software OS reservation? Does the OS require a certain amount no matter what? Does that then mean only X% of those 4 CU's can actually be released for use by something other than the OS for graphics rendering?
Here's a hint. We don't know. We have absolutely zero idea if it's a contiguous 18 CUs or if there's some physical separation causing a 14+4 configuration or if there's just software/OS restrictions or if it's something else entirely.
Claiming all 18 can be used for graphics rendering is just as true/false as saying 4 CUs can't efficiently be used for graphics rendering as we just do not know enough information.
Can all 18 be used equally for graphics rendering? Sure that's a possibility.
Can only 14 be used equally for graphics rendering with 4 being sub-optimal? Sure, that possibility is just as likely. But if you believe anything coming out of VGleaks about the specs, then surely this one has a slightly higher probability. Or you choose not to believe it, and then at that point all of the information from VGleaks about Orbis comes into question as to whether it is valid or not.
I have no real opinions about it either way. I do like to speculate on what it could possibly be. If all 18 CUs can be used equally, it's no skin off my back. If there's some reason that 4 CUs can't be used as well for graphics rendering, again no skin off my back.
Somehow claiming that one situation is more "correct" than another situation however, serves no purpose as everyone is working off of incomplete information that we don't even know for sure is even correct.
Regards,
SB
anexanhume
Veteran
Yes, because he'd be claiming orbis was 2.8TF. He's an attention fiend.
And who would you believe?
Obviously this thread is full of fanboyism, from both sides, not because one of the next gen consoles can be more powerfull, it is because all people are "adding" things to their system and "substracting" things to the other system, without the full info about specs.
Betanumerical
Veteran
Please quote the part where they shut down the 4 CU as been useless like you claim,this is starting to sound like mayor nelson comment about cell on 2005 when he call Cell SPE a bunch of useless DSP's.
Now we know how useless spe were.
Im saying that the article destroys the myth of them being useless because of what i posted. I interpret the last bit as them saying all the CU's contain at least this. Meaning the other 4 are just as capable if not more then the other 14.
Averagejoe
Regular
Very good post.
In fact they would probably perform the same because developers don't persue power they want playable versión in fact durango could have games that perform better con it,bit it will not because they perform the same.
Doesn't it seem odd to claim the 4cu's are a minor boost for graphics if they are actually better at it than the other 14. Terrible wording makes it really open as to what it could be.
You added the word "designed", you made up an information that 14 are "dedicated", and claim 4 are "reserved" for imaginary things. All of this is in your head. It's not in the source. You're making things up.
Well, some people are assuming things too, like all 18 CUs are equals, this is not in the source.
Betanumerical
Veteran
True, but the source at least mentions the minimum of which all 18 CU's provide, what those other 4 CU's provide on top of that we do not know.
I think the major issue is this line
Minor is very arbitrary.
For some people <10% is minor. For some <50% is minor. All it says is that is the benefit is >0%. We don't know the upper bound
When I read the wording the way I take it as that when the system is using 14CUs, it is maxing out some other aspect of the rendering pipeline, hitting its peak limit. You are free to use the extra 4CUs for rendering however there is some aspect (bandwidth, memory, fill rate, anything) that is constrained once you hit 14 that makes it useless to throw any more CUs at rendering.
Sure, you may get slightly more efficiency but its diminishing returns (minor boost). Sony probably had some transistors leftover in its budget and, being burned by overly custom designs before, decided to just up the number of CUs to be used for some extra math as the developer saw fit. SO yes in terms of all the CUs doing computational tasks, the GPU can operate at theoretical peak of 1.8TF but from a strictly rendering pipeline perspective, anything over 14CUs will be wasted due to some yet unnamed constraint. Therefore the GPU operates as a 14CU GPU and the overall design is 'balanced' for a 14CU GPU. There.
Sure, you may get slightly more efficiency but its diminishing returns (minor boost). Sony probably had some transistors leftover in its budget and, being burned by overly custom designs before, decided to just up the number of CUs to be used for some extra math as the developer saw fit. SO yes in terms of all the CUs doing computational tasks, the GPU can operate at theoretical peak of 1.8TF but from a strictly rendering pipeline perspective, anything over 14CUs will be wasted due to some yet unnamed constraint. Therefore the GPU operates as a 14CU GPU and the overall design is 'balanced' for a 14CU GPU. There.
It's not in the source that Orbis is using Dried Unicorn Blood as a PCB dye, either. (remember you heard it here first!)
There's a difference between reasonable assumptions and making things up. The source doesn't say the CUs are different. So we assume they're not., because until we have more info it's a reasonable assumption. This is supported by the CU redundancy requirements, and by the numbers in the source, 410Gflops for 4 is matching the other CUs: 1.84T/18=102G. 4x102G=410G. It's the same processing power, so what would they modify? These modifications need to make it a better system than leaving the 18 untouched. There's a good angle that they are wired differently, or are accessed more low-level, or have an independent dispatcher... but there has to be a reason to do that, a gain versus leaving them as is, and they don't disappear.
The reason the "minor benefit" claim is ambiguous is because we don't know the point of reference. Is it a minor benefit compared to not using them for graphics? Is it a minor benefit compared to them being "stock" CUs? Is it a minor benefit because they only represent ~20% of the total FLOPs when used? Or is it a minor benefit compared to using them to assist the CPU where it would be a 400% gain? And we still don't know what the author considers "minor".
This seems like a sensible reading, but it's hard to think what kind of constraint might be so specific in a non-partitioned UMA box.
My take on all this is that the CPU in Durango will make up for the deficiency of compute.
I believe that 4 of the CU in Orbis will be reserved for Pseye(s) and/or physics work.. BUT can be used for rendering; though due to the overall balance of the system would only provide a minor boost. Could also be that not the entirety of the 4 CU's are reserved for PSeyes. Could be that only two and a half put of the four are, and so it would only be a minor boost if using that block of 4 for rendering.
The way I see it shaking down though is, lets forget about the 4 CU for a moment:
In terms of rendering, we've got 14 vs 12 CU... now if the new Xbox has a beefier CPU, say with .4TF of compute vs .1TF that difference will have to come from Orbis's 14 CU... so now we've got:
New Xbox: 1.6TF coming from the CPU and GPU
PlayStation: 1.5TF coming from the CPU and GPU
It's a wash. (as many developers have been saying)
This would also tie into DaE and friends comments.
So there are really two things yet undetermined in this whole shebang (assuming the leaks are still accurate)... the +4 CU and their role, and the CPU in Durango.. but all signs are pointing to more flops being uncovered yet to make this a more level playing field.
Btw, what kind of improvements *are* likely in Durango's CPU? like MOST likely? I'm curious of as to what kind of improvements can be made to double or quadruple the flops, and what one is likely to expect.
Oh and I doubt we'll be getting anything other than vague specs on the 20th.
I believe that 4 of the CU in Orbis will be reserved for Pseye(s) and/or physics work.. BUT can be used for rendering; though due to the overall balance of the system would only provide a minor boost. Could also be that not the entirety of the 4 CU's are reserved for PSeyes. Could be that only two and a half put of the four are, and so it would only be a minor boost if using that block of 4 for rendering.
The way I see it shaking down though is, lets forget about the 4 CU for a moment:
In terms of rendering, we've got 14 vs 12 CU... now if the new Xbox has a beefier CPU, say with .4TF of compute vs .1TF that difference will have to come from Orbis's 14 CU... so now we've got:
New Xbox: 1.6TF coming from the CPU and GPU
PlayStation: 1.5TF coming from the CPU and GPU
It's a wash. (as many developers have been saying)
This would also tie into DaE and friends comments.
So there are really two things yet undetermined in this whole shebang (assuming the leaks are still accurate)... the +4 CU and their role, and the CPU in Durango.. but all signs are pointing to more flops being uncovered yet to make this a more level playing field.
Btw, what kind of improvements *are* likely in Durango's CPU? like MOST likely? I'm curious of as to what kind of improvements can be made to double or quadruple the flops, and what one is likely to expect.
Oh and I doubt we'll be getting anything other than vague specs on the 20th.
I do think the author clearly intends for the reader to marry "minor benefit" and "balanced for 14cus" though. To me this can only mean 3 things.
1. The 4 are physically less suited for rendering tasks
2. They are physically separated or connected in such a way that makes them less suited for rendering tasks
3. As i stated above, the rendering pipeline is optimized (or saturated) by 14cus. Any additional cus thrown at rendering are constrained by this saturation.
I honestly don't see a way to interpret it as written where this chip can effectively use 1.8tf or >14cus for rendering in a meaningful way.
- Status
Similar threads
