It's in the VGLeaks diagram here. The text description is:
Nothing there about additional scalar processors.
- Status
One of the pics shows 4 VSPs per SIMD, with each VSP being expanded into 4 vector pipes.
I think that's pretty much GCN
SC = CU
SIMD=SIMD
VSP = Vector ALU
4 vector pipes = 4 vector pipes
12 CU (SC) * 4 SIMD * 4 ALU (VSP) * 4 vector pipes = 768
I'm guessing marcberry won't be heard from again.
Either contribute constructively or go away. There is very little in the VSP idea that's got actual evidence. You can argue an ambiguity of the meaning of 'VSP' in the VGLeaks diagram, but the text is clear. You claim there's a patent explaining this. I've politely asked for but you've not provided it yet. Google throws up nothing, so your assertion, "that is their is a PATENT from AMD for VSPs it's big" rings hollow - VSPs are so big, no-one's talking about it?
At this point, you need to present technical support for you POV, best done by serving up this patent and proving us all wrong. Without such technical documentation you are completely contradicted by the information so should just agree to disagree, rather than suggesting everyone's too slow you comprehend your incredible understanding.
So I guess VSP refers to Vector Scalar Pipes instead of Vector Scalar Processor?
If people have something against no DVD, that really seems closed minded. This new system is probably not targeted at gamers, more like TV/movie/app users who might play a game or two. This generation is going on 7+ years. Anybody buying a 360 now is either replacing an older one or just using it to watch content/use Kinect.
There other things in the rumors that are more odd & need more picking apart. Yes, Kinect not being mandatory seemed odd. But that's still not worthy of attention either. There are ways they could do that & still be mandatory(dev assumes every user has one & if they don't have one, it won't work).
The part I think everybody glossed over was that the new revised Xbox 360 won't even be called Xbox 360. It will be called Xbox <new brand>. Same goes for Durango. It will be called Xbox <New Brand>. So the new 360 will be a subset of Durango. Every Durango will have the new 360 SOC making it fully backward compatible. Sounds more like they're making Xbox 360 Version 2013 instead of a whole new generation. They're going to be milking the 360 design for years to come, but calling it something else. Sounds like a great way to transition existing users to the next generation, but I was kind of hoping that rumor about Durango supporting 360 games via a separate add-on was the direction they were going. The biggest issue I have is I can't see the 360 SOC being used for the low power state. Also, does the new revised 360 support both Kinect models or just one? If one, which one?
As for the controller supporting touch. I'm kinda blah on that idea. It would be nice to see something different. I was hoping for the smart watch/VMU idea.
The minor upgrade to the UI seems plausible. They have almost 8 years of UI development leading up to this. If it works, why change just for the sake of change? I like the multi-tasking bit, but again it's something we've already heard about.
$150 for Xbox 360 w/ no DVD drive. $99 would be better, but they might go free with Xbox Live sub.
$400 for Durango & no Kinect? $250 with Xbox Live sub?
$500 for Durango & Kinect? $350 with Xbox Live sub?
In the end some of it seems plausible, some not. Just like a lot of the rumors we've gotten the past 6-12 months. LOL
Tommy McClain
There other things in the rumors that are more odd & need more picking apart. Yes, Kinect not being mandatory seemed odd. But that's still not worthy of attention either. There are ways they could do that & still be mandatory(dev assumes every user has one & if they don't have one, it won't work).
The part I think everybody glossed over was that the new revised Xbox 360 won't even be called Xbox 360. It will be called Xbox <new brand>. Same goes for Durango. It will be called Xbox <New Brand>. So the new 360 will be a subset of Durango. Every Durango will have the new 360 SOC making it fully backward compatible. Sounds more like they're making Xbox 360 Version 2013 instead of a whole new generation. They're going to be milking the 360 design for years to come, but calling it something else. Sounds like a great way to transition existing users to the next generation, but I was kind of hoping that rumor about Durango supporting 360 games via a separate add-on was the direction they were going. The biggest issue I have is I can't see the 360 SOC being used for the low power state. Also, does the new revised 360 support both Kinect models or just one? If one, which one?
As for the controller supporting touch. I'm kinda blah on that idea. It would be nice to see something different. I was hoping for the smart watch/VMU idea.
The minor upgrade to the UI seems plausible. They have almost 8 years of UI development leading up to this. If it works, why change just for the sake of change? I like the multi-tasking bit, but again it's something we've already heard about.$150 for Xbox 360 w/ no DVD drive. $99 would be better, but they might go free with Xbox Live sub.
$400 for Durango & no Kinect? $250 with Xbox Live sub?
$500 for Durango & Kinect? $350 with Xbox Live sub?
In the end some of it seems plausible, some not. Just like a lot of the rumors we've gotten the past 6-12 months. LOL
Tommy McClain
Question for some of the more knowledgeable folks around here:
If Global illumination is a time consuming affair that is/has typically been done during development and is then "pre-baked" into light-maps, would it not be possible for MS to provide cloud based light calculating servers which offloads that workload from the devs and the local machine while simultaneously providing a high quality lighting solution which is interactive?
Granted, it doesn't/wouldn't provide instant calculations (lightmaps), but for most instances, it wouldn't need to.
I was just thinking of ways which would allow a cloud based assisted compute model could work in the interim while we all wait for zero lag gigabit internet.
It seems to me that Environment Lighting, AI, and voice commands are all low hanging fruit in that regard.
What do you all think?
If Global illumination is a time consuming affair that is/has typically been done during development and is then "pre-baked" into light-maps, would it not be possible for MS to provide cloud based light calculating servers which offloads that workload from the devs and the local machine while simultaneously providing a high quality lighting solution which is interactive?
Granted, it doesn't/wouldn't provide instant calculations (lightmaps), but for most instances, it wouldn't need to.
I was just thinking of ways which would allow a cloud based assisted compute model could work in the interim while we all wait for zero lag gigabit internet.
It seems to me that Environment Lighting, AI, and voice commands are all low hanging fruit in that regard.
What do you all think?
VSP appears to be something they've made up to illustrate the fact, as discussed in the text, that the SIMDs work differently from the previous AMD SIMDs by servicing vectors made from four scalars instead of a vector operand (if I've got that right). I see nothing anywhere on the internet to suggest VSP is an official AMD concept, let alone an implementation that pairs every SIMD unit with a scalar processor, and even the VGLeaks description suggests as much.
Possible, but then you're adding server costs to each Durango, needing to provide the local hardware and xxx TFlops in some server somewhere. There's a thread on distributed computing where you could ask this for a better, technical discussion, as long as you ignore the bickering already in that thread.
Thanks
I see.
I thought this might be a more appropriate thread for the question as the server farm then becomes an extension on the console for compute resources.
Sort of like a dual APU ... with a REALLY slow connection between them but with substantially more compute resources.
Regarding the costs associated, I'm not sure that's a fair cost association as the inevitable end-game for consoles seems to be "the cloud" anyways, so this would simply be a slow build up to that eventuality. Not to mention the ability to save costs per unit while still providing a top-tier gaming experience which isn't bound to a baseline spec created at a specific point in time.
But if you feel that this is the wrong thread for that discussion I apologize (again
).I'm wondering if the diagram was drawn from the point of view of a shader working on pixel quads. A given pixel's work should work through its components sequentially in one lane, and the other three pixels in the quad would be given adjacent SIMD lanes.
There is one mention of "vector shader processor (VSP)" in the southern islands ISA document, but it's such a one-off I'm wondering where it's coming from.
Granted, the fact that it is in the SI ISA document kind of gives a hint as to what architecture Durango's GPU is based on.
Here is an AMD patent. Look at the first drawing.
http://www.google.com/patents/US20120019542
Hopes it helps.
It's a concept not limited to any specific console, and not rumoured for Durango, so it's worth getting it's own investigation as a thought exercise.I see.
But if you feel that this is the wrong thread for that discussion I apologize (again
The VSP there is a SIMD unit working concurrently on four values. I'm very happy that's all we're talking about with the VSP - a SIMD working on four scalars instead of a vec4.
That seems closest to a VLIW4/VLIW5 GPU, and is less consistent with the Durango leaks than GCN. The arrangment of the L1, LDS, scheduler, registers, and texture units don't match the Durango leak.
I'm curious whether VSP means the same thing in all places.
+1..
The AMD patent tells you what a VSP is, they do not use VSPs in any other AMD documents in showing how GCN 1.0 works.
Later i will go thru some real math and post the patent if you have not found it. We know now that this is not a GCN 1.0 GPU. The VGleaks and or SuperDae is keeping something from us, this is why MS did not try to take it down. We are now 96hours away and we are just now looking into VSPs and it has been their all the time. Also look at the VGT their is one for each of the 12SC cores compared to having just two per GPU. Soon the math.
Later i will go thru some real math and post the patent if you have not found it. We know now that this is not a GCN 1.0 GPU. The VGleaks and or SuperDae is keeping something from us, this is why MS did not try to take it down. We are now 96hours away and we are just now looking into VSPs and it has been their all the time. Also look at the VGT their is one for each of the 12SC cores compared to having just two per GPU. Soon the math.
Actually, they do bring up a VSP once in the Southern Islands (GCN) ISA document.
are you referring to this patent?
Integrated Vector-Scalar Processor
http://www.google.com/patents/US20100332792
Or you could just post and get it over with, because right now it seems like no one is convinced it's worth the effort to search for i.
I would argue that it's generally good form to substantiate one's own claims instead of the "make my point for me" method of debate.
- Status
Similar threads
- Sticky