![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
XBOX360 Revealed: Implications for R520?
- Thread starter ^eMpTy^
- Start date
Lorien
Supreme [H]ardness
- Joined
- Aug 19, 2004
- Messages
- 5,197
There are no implications since R500(xbox 360 console gpu) has very little in common with R520(upcoming PC consumer market gpu)
The tech thats on the xbox360 we'll be finally able to have in a consumer gpu next generation in the form of R600.
The tech thats on the xbox360 we'll be finally able to have in a consumer gpu next generation in the form of R600.
Sly said:
That's pretty much the complete article.
I'm sure the cores are quite a bit different...I just figured that they would be based on the same technology roughly, so like the same performance per clock thereabouts as R520...so when they say the fillrate at 500Mhz is lower than a 6800Ultra...I find that very wierd...
Maybe we're getting to the point where fillrate isn't the limiting factor anymore?
S_Z said:
Hmmm..ok here's my question then...if they are done with the R600 design to the point that it's going to be in the xbox360...why not just release the technology on the PC at the same time? Maybe because the performance isn't up to par? But if performance is lacking, why would microsoft use it vs R520? Any ideas?
I don't think having a slow speed is all that limiting. I mean the x800 chips are clocked faster yet the nV40 have a higher shader through put. Dave B over at B3D has also been stressing that we need to NOT focus on FillRate, Clock and # of Pipes to be the SOLE gage of perfromance on the new hardware. Also keep in mind that given the small form factor that they may not have and endless supply of cooling options like a PC case does...
^eMpTy^ said:
Fillrate depends mostly on the speed of the videomemory, not on the speed of the GPU.
Since the XBox will run at relatively low resolutions (1080i at most?), it doesn't need all that much raw fillrate. Seems like they want to concentrate on more advanced shading instead, which would mean that they can use cheaper memory in the unit.
I suppose the R520 will be paired with faster memory and more fillrate, because PCs can run at higher resolutions, and the overall price can be higher.
I think you misunderstood him. I dont think he (or anyone else) is saying that the R600 design is finished in any way. Its just that R600 will be based on the R500 (Xbox 360) architecture, with unified shader model and so on. The R520 is said to still be based on the R300 architecture (heavy modified of course) and maybe have some things in common with the R500 but it wont be a completely new architecture.^eMpTy^ said:
I of course have no idea if this info is correct, its just what I have read in the B3D forums.
Lorien
Supreme [H]ardness
- Joined
- Aug 19, 2004
- Messages
- 5,197
R600 is a ways off and the reason its not being released yet can be one or all of the following things.
- No need since they already know what NV will bring to the table with G70 performance wise and only a chip like R520 w/MVP(SLI response) support is enough to maintain a performance/features lead/parity.
- It could also be that the tech in xbox360 gpu was too advanced and not neede/not ready for prime time in the PC consumer side. Releasing it in a console makes perfect sense for troubleshooting with regards to driver development.The console being a closed system gives them the best place to work out the kinks of a new architecture.
- Xbox360 gpu is also on a new process 90nm which again gives em the opportunity to test the waters. ATI are known to not take risks when moving to a new manuf. process. Remember the mid/low end Radeon 9600Pro(RV350) came out first at 130nm before the high end X800 series(R420).
- From what ive heard R600, the PC consumer version of xbox360 gpu, is supposedly going to be released in a 65nm process which has just been unveiled by TSMC so process availability may be another factor.
http://www.beyond3d.com/forum/viewtopic.php?t=22537
Edit: Another reason is why release R600 when the OS to support it is not here yet, i.e. longhorn and WGF2.0 which this chip supports.
- No need since they already know what NV will bring to the table with G70 performance wise and only a chip like R520 w/MVP(SLI response) support is enough to maintain a performance/features lead/parity.
- It could also be that the tech in xbox360 gpu was too advanced and not neede/not ready for prime time in the PC consumer side. Releasing it in a console makes perfect sense for troubleshooting with regards to driver development.The console being a closed system gives them the best place to work out the kinks of a new architecture.
- Xbox360 gpu is also on a new process 90nm which again gives em the opportunity to test the waters. ATI are known to not take risks when moving to a new manuf. process. Remember the mid/low end Radeon 9600Pro(RV350) came out first at 130nm before the high end X800 series(R420).
- From what ive heard R600, the PC consumer version of xbox360 gpu, is supposedly going to be released in a 65nm process which has just been unveiled by TSMC so process availability may be another factor.
http://www.beyond3d.com/forum/viewtopic.php?t=22537
Edit: Another reason is why release R600 when the OS to support it is not here yet, i.e. longhorn and WGF2.0 which this chip supports.
either that or microsoft paid them a shitload to make them a gpu that was 2 generations ahead of whats out now so that the xbox360 will be not only king turd when it comes out but will contiune to be untill the r600 is released. After the r600 comes out, it will still be top of the line and the need for an new xbox will be greatly reduced.
Then they can sell more games, which means more money to combat the losses they'll make selling the console in the first place. The longer the xbox 360 is their current console, the more games they'll sell and they'll make alot more money then if they released a newer console sooner.
And why wouldn't ati just release a card now based on whatever the xbox 360 is getting? Money. Why release a card thats 2 generations ahead of whats out now? Release something half way now, the r520, let it sell then release the r600 later when nvidia comes up with something that'll threaten their sales.
Thats my grasp of the situation, tho i might have all my facts wrong, so i dunno
Then they can sell more games, which means more money to combat the losses they'll make selling the console in the first place. The longer the xbox 360 is their current console, the more games they'll sell and they'll make alot more money then if they released a newer console sooner.
And why wouldn't ati just release a card now based on whatever the xbox 360 is getting? Money. Why release a card thats 2 generations ahead of whats out now? Release something half way now, the r520, let it sell then release the r600 later when nvidia comes up with something that'll threaten their sales.
Thats my grasp of the situation, tho i might have all my facts wrong, so i dunno
geekcomputing
[H]ard|Gawd
- Joined
- Aug 30, 2004
- Messages
- 1,100
is it just me or does the name xbox 360 sound dumb.
xbox 2 would be better.
no one is going to call it xbox 360 , that is just too long , at least for the southerners.
xbox 2 would be better.
no one is going to call it xbox 360 , that is just too long , at least for the southerners.
Brent_Justice
Moderator
- Joined
- Apr 17, 2000
- Messages
- 17,755
I'm very intrigued with the graphics processor it has, yummmmy.
S_Z said:
please, for god sake someone find any official post saying this. I cannot BELIEVE that this is so actively believed. Its been said right from the announce of he R520 series its a completely new core. So someone find where its yet another extension please.
edit: R420/R480 180M transistors 130nm
R520=300-350M transistors
R520=GDDR3 and GDDR4 Support plus support of up to 1gig in memory
R520=90nm poduction, sorry i dont believe they dropped to a much more advanced fab process for a refresh.
R520 to have support for UP TO 32 pipes thoguh 24 expected on launch each capable of doing 1.3times the work of todays pipes.
Longhorn WGF1.0 compliant
What have i been spelling out? Its a core made to last, much more advanced then the R420 and quite a but more advanced then the Nv40. The G7xxx series is quite assuredly going to be shorter then the R520 in transistors, though the G70 will be based off a newer core according to you? Dont think so. Its simply not an extension. Last thing is that in terms of speed the most ATI could go is 600mhz to start, under the same WPC as the R420 if its yet another refresh the core would actually prove to be quite a crappy one in terms of improvment. I have no idea where this rumor took a foothold but its total bs. If anything i see the R600 design being based off this core simply because the R520 is such a radical change from all previous ATI core designs.
FanATIc said:
Yeah as far as I have heard R520 is a completely new core. It has PS3.0 and FP32...so it has to be waaaaay different than R420...
I don't think anyone can make any statements about G70 right now though...considering that R520 was delayed from last year to this year, one could argue that the G70 which was moved up from next year to this year, will be a more advanced core...
Brent_Justice
Moderator
- Joined
- Apr 17, 2000
- Messages
- 17,755
I'm interested in the "*48-way parallel floating-point dynamically-scheduled shader pipelines- Unified shader architecture" part of the specs.
Sounds a lot like virtual pipelines to me, maybe 24 pipes in the traditional definition, but can dynamically schedule it so it's like it has 48 if need be. Something like that maybe, just theorizing here, can't wait to hear more about it.
Also the Unified shader architecture, sounds like SM 4.0 to me (WGF 2.0). Again, just guessing.
Sounds a lot like virtual pipelines to me, maybe 24 pipes in the traditional definition, but can dynamically schedule it so it's like it has 48 if need be. Something like that maybe, just theorizing here, can't wait to hear more about it.
Also the Unified shader architecture, sounds like SM 4.0 to me (WGF 2.0). Again, just guessing.
Brent_Justice said:
What they mean is that they've moved away from separate vertex and pixel units, and now have shader units that can perform both vertex and pixel operations.
Which means that the chip can dynamically reconfigure itself... if you have a lot of geometry to process, it could reconfigure itself to eg 47 vertex units and 1 pixel unit... on the other hand, if you need to draw lots of pixels, it can reconfigure itself to eg 1 vertex unit and 47 pixel units.
ATi has mentioned this idea many times before, and it makes a lot of sense... NVIDIA hasn't shown any interest in it so far, claiming that having separate units, each optimized for its own task, is a less complex and a more efficient solution.
So I'd love to see what ATi's solution can do in practice.
I think these units will just be VS3.0/PS3.0-capable by the way. And I don't expect any geometry shader functionality yet... But I hope I'm wrong
^eMpTy^ said:
Im not sure Nvidia did bump back a core to be honost. Just because a project called the NV50 (problably a basic yearly refresh) was scrapped and they went to a different code name system: G60= Geforce 6xxx Series, G70= Geforce 7xxx series, G80= Geforce 8xxx series, doesnt necissarily to me mean they are bringing something new completely to the table. Dont forget they will be introducing it right along side ATI so they couldnt of started work right when that NV50 thing was happening. That was right around christmas. I think it would be a loss personally for them to use the NV40 base for only a year.
Guess we'll see.
Scali said:What they mean is that they've moved away from separate vertex and pixel units, and now have shader units that can perform both vertex and pixel operations.
Which means that the chip can dynamically reconfigure itself... if you have a lot of geometry to process, it could reconfigure itself to eg 47 vertex units and 1 pixel unit... on the other hand, if you need to draw lots of pixels, it can reconfigure itself to eg 1 vertex unit and 47 pixel units.
ATi has mentioned this idea many times before, and it makes a lot of sense... NVIDIA hasn't shown any interest in it so far, claiming that having separate units, each optimized for its own task, is a less complex and a more efficient solution.
So I'd love to see what ATi's solution can do in practice.
I think these units will just be VS3.0/PS3.0-capable by the way. And I don't expect any geometry shader functionality yet... But I hope I'm wrong
Sounds very cool but thats still one hell of a leap. X800/6800 are 16, 6 so 22 total pipes.
Up and coming should be 24, 8 so 30total pipes with the R520 having support for 40total.
The R500 spec still has some pipes not yet accounted for or even mentioned in any video card horizon. Wont believe anything until someone officially comes up with an answer. Both ATI and Nvidia have actually commented that less pipes but making each one longer would be a more realistic future, such as how CPUs work today. I find it odd they'd make such a massive leap against what they have been saying.
DarkMonkey
Limp Gawd
- Joined
- Dec 30, 2004
- Messages
- 454
But without speculation and rumors, how are they supposed to pass the time until the product is released?
mohammedtaha
2[H]4U
- Joined
- Oct 14, 2004
- Messages
- 3,648
speaking of ati ... does ati.com load for you guys ?
TehQuick said:here we go again, speculation and rumors. Would u guys wait until the actual products are out before comparing them?
How is that fun?
Anyways, my understanding was that the r500 architecture had been binned entirely ages ago by ATI and replaced with what is now the r520, and that the gpu in the Xbox360 is a custom version of the r520 core designed to fit in with the funky console type architecture, not some futuristic r600 type part.
Also, don't the nvidia 6 series already have a fancy 32x0 virtual pipeline thing going on, maybe the 48 reference on the xbox specs is the same sort of deal.
mastercheeze
2[H]4U
- Joined
- Jul 23, 2004
- Messages
- 2,799
t3mp said:
Nope.
http://electronics.howstuffworks.com/hdtv1.htm
http://hardforum.com/showthread.php?t=900035
They're both 1920x1080, the difference is in how they are shown. Progressive (1080p) displays the whole image at once, 60 times a second. Interlaced (1080i) displays alternating halves of the image 60 times a second, i.e, you still need the whole resolution image.
edit- DANG mjannusch beat me to it!
mohammedtaha
2[H]4U
- Joined
- Oct 14, 2004
- Messages
- 3,648
MartinX said:
even though this can be very true .. the same old core but faster GDDR3 that does 700mhz 1.43ns it would mean we would be getting the same heat emitted as a X800XT .. but if this is the new core it can be a lot cooler .. correct ?
Rune75
[H]ard|Gawd
- Joined
- Aug 16, 2004
- Messages
- 1,347
^^^^ Using 90nm tech on the xbx360 would let the gpu run a lot cooler than your standard x800 video card
Lorien said:
FanATIc said:
I don't know if you can use transistor counts to tell if it is a new core. From what I have read, they speculate that adding FP32 will add aprox 30% more transistors. Add 8 more piplelines and SM3.0 tech and the 300 million transistors reported for the r520 is reached.
I am not saying that it is not a new core, I am just pointing out it is possible it is still based on the R420 core.
peltman78 said:
well I doubt they just ditched r300 technology altogether...I'm sure that this is in some way an evolution of that technology...but certainly r520 will be a big departure simply because adding FP32 and SM3.0 makes it significantly different...those changes alone make R520 a "new" core in my eyes...
^eMpTy^ said:
Very true. I guess it all depends on how you define new core. If that is your definition than I am ok with that.
I am more curious, if (speculating totally here) the R500 is a new core that is the base for the R600, does it have the Geometry shader hardware that is supposed to be a requirement for WGF 2.0? I am very interested to see what you can do with that if it does.
peltman78 said:
I didnt just name transistor count, i named quite a few things that will all allow the core to stretch for along time to come. 30% more=230M, Nvidia did that with 222M and SM3.0 tech. So unless you yourself came to the conclusion that ATI dropped in another 70-110M more transistors for pipes while Nvidia, again, will be doing it with under 300M, then i dont understand your way of thinking.
Why is it so hard to grasp it being a completely new core?
Empty, every core from every company is an evolution. They dont just pop up out of no where.
R300=107M transistors .15Fab
9800XT Core=115N transistors .15fab
R420=180M transistors .13Fab
R520=300-350M transistors 90nm fab
see a slight discrepency?
Most advanced NV3x core was the 5950U
130M transistors .13fab
NV40 222M transistors .11fab
FanATIc said:
No Shit? That's exactly what I was saying.
NV40 is 130nm...
You jump to a new process when the new process makes sense...not just because you have a new core...though the two events often coincide because cores are tailored to the process they are manufactured on to a degree...
Lorien
Supreme [H]ardness
- Joined
- Aug 19, 2004
- Messages
- 5,197
Yes because electric engineering phd's and people who work in the 3d graphics industry are !!!!!!!s right? Your attitude is typical of the population of this forum and one of the main reasons why it remains stagnant and devoid of quality posters. Don't worry tho i dont plan on staying here longsteviep said:
Its a new process, one which they dont have experience with. It is much better to have someone like MS carry the burden of paying for the manufacturing(ATI simply designed the chip and MS is in charge of the rest) and get any data/input from then on how its doing(yields, achievable clock speeds, etc) before releasing it on thier side as a pc consumer product.
Think of it as a pool of ALUs that can work on pixels and vertexes. Reading the last patent from ATI and knowing now what we know of R500 we can understand it better.
"The shader core has 48 Arithmetic Logic Units (ALUs) that can execute 64 simultaneous threads on groups of 64 vertices or pixels. ALUs are automatically and dynamically assigned to either pixel or vertex processing depending on load."
The patent refers to an arbiter device that distributes to load of this pool of 48ALUs between pixel tasks and vertex tasks dynamically. This was one of the main doubts people had about this chip. How can all those 48 ALUs remain busy all of the time or at least most of the time to reduce pipeline stalls or when branch prediction fails?
Here's a diagram
http://www.beyond3d.com/news/images...imulti.gif&comment=ATI Multi-Threading Patent
There's some more speculation going on at b3d.
http://www.beyond3d.com/forum/viewtopic.php?t=22693&postdays=0&postorder=asc&start=140
The good stuff starts in page 9 of that thread
Here's some possible R600 morsel as well
http://www.beyond3d.com/forum/viewtopic.php?t=22851
Lorien said:Yes because electric engineering phd's and people who work in the 3d graphics industry are !!!!!!!s right? Your attitude is typical of the population of this forum and one of the main reasons why it remains stagnant and devoid of quality posters. Don't worry tho i dont plan on staying here long
He has a point actually, since a lot of these engineers are ATi employees... combine that with the fact that there are no NVIDIA employees (is that NVIDIA's company policy, or is that B3D's policy?), and you can get pretty biased conversations indeed.
Lorien
Supreme [H]ardness
- Joined
- Aug 19, 2004
- Messages
- 5,197
Scali said:
Err the 2 companies have different policies with regards to letting their employees speak about released/unreleased products. Take a guess which one is more friendly/open to the community?
Edit: And it is not "a lot" but if you consider 3 maybe 4 "a lot" be my guest