AMD: R7xx Speculation
- Thread starter Unknown Soldier
- Start date
- Status
Jawed
Legend
Seems like it must have been.
Looking at the D3D Rightmark fill rate test here:
http://www.techreport.com/articles.x/14934/7
RV670 delivers, at best, about 12GT/s, which is what's expected theoretically. So HD4870 can be expected to hit about 30GT/s, since HD4850 hits ~20GT/s (67% of theoretical) with 56% of the bandwidth.
30GT/s is only a smidgen behind GTX280, quite a difference from the gap between HD3870 and 9800GTX
Jawed
NO WAY THAT IS REAL-TIME!
But if so then im just spechless especially at the real-time reflections of everything including the chick, ray tracing based?
NO WAY THAT IS REAL-TIME!
But if so then im just spechless especially at the real-time reflections of everything including the chick, ray tracing based?
Isn't the background static and only Ruby, the taxi and the robot being rendered by the card? Thought I saw something along those lines when the video first appeared.
Jawed
Legend
You're serious?The taxi brakes to a halt as Ruby runs out, the robot then sends the taxi flying across the street (knocking debris off the building it hits) as well as a dragging a fire escape out of the side street. At least two people in the background go scarpering off and other traffic crosses on the junction behind the action. The robot creates impact craters in the tarmac and the car on the left shakes with the impact.
Jawed
When NDAs in regard to new ATI cards are supposed to end?
I willing to learn more.
I n regard to chip size, could ATI have trade some registers for LS, density may be higher.
And company like clearspeed claim that LS can be as fast as registers.
I willing to learn more.
I n regard to chip size, could ATI have trade some registers for LS, density may be higher.
And company like clearspeed claim that LS can be as fast as registers.
Last edited by a moderator:
The taxi brakes to a halt as Ruby runs out, the robot then sends the taxi flying across the street as well as a dragging a fire escape out of the alleyway. At least two people in the background go scarpering off and other traffic crosses on the junction behind the action.
Jawed
Yeah no idea. Thought I had seen something along those lines. The car and wall to the right also get damaged so it could be all rendered real time. Pretty impressive either way.
Yeah it seems that flops is the performance basis for those numbers. I'm hoping that the 4870 is a decent jump over G92. Everything we know so far is very appealing....except for performance. Just one more week....
Mintmaster
Veteran
The 20GT/s number is based on those 3DMark screenshots from several pages back, right?
Those tests have never been BW limited, as they use fairly magnified textures. While it's true that aside from G80 most recent NVidia cards have been well below peak, several of us have guessed that the reason is attribute interpolation throughput and that shouldn't affect RV770.
A good piece of evidence against BW being a factor in that multitexturing test is the speed of the 9600GT and 8800GT:
http://www.techreport.com/articles.x/14168/4
(Note that the Palit 9600GT is the one tested)
The 8800GT is 45% faster, which matches its 50% theoretical advantage despite having less total bandwidth. In other words, efficiency was about the same with 40% less BW per TMU. Moreover, the 8800GT has much less BW than a 4850 and gets well over 20GT/s.
I'm not too sure why the 4850 is lagging a bit here, but remember that we're looking at numbers from some guy on a forum with beta drivers.
Jawed
Legend
Do the 4870's coming next week have 1GB of RAM or only 512 due to the GDDR5 shortage?
http://www.fudzilla.com/index.php?option=com_content&task=view&id=7947&Itemid=1
http://www.fudzilla.com/index.php?option=com_content&task=view&id=7947&Itemid=1
Please stop the 480 VS 800 shader bs, CJ said 800 and GPU-Z says 800. Beside that it wil have 40 TMU and core communiction on R700 goes to GDDR5 bandwith (this seems confirmed by an ATI R700 slide that showed the two GPU's on the same memory ringbus)
Only thing left is in what kind of way 40 TMU's are going to be limited bij 4850 by 57 Gb/s mem bandwith. Some say 4850's will easely hit 1100 Mhz so there is some OC headroom but will it be enough?
Only thing left is in what kind of way 40 TMU's are going to be limited bij 4850 by 57 Gb/s mem bandwith. Some say 4850's will easely hit 1100 Mhz so there is some OC headroom but will it be enough?
DegustatoR
Legend
480 SP + high shader domain clocks = "800 SP"?..My first estimate based on that die shot, if it is indeed a RV770 die shot, is that AMD is the first to a 2GHz shader clock with the HD4870...
No, 6*10*5 SPs = 300 SPs at 2.67x the core clock = 800 SPs
EDIT: Just to make myself clear, it might also be 600 SPs @ 1GHz and much denser or whatever. Point remains though that I would be extremely surprised if there wasn't some shader clock trickery going on here. It's the only sane explanation for the astonishingly low die size, anyway...
EDIT: Just to make myself clear, it might also be 600 SPs @ 1GHz and much denser or whatever. Point remains though that I would be extremely surprised if there wasn't some shader clock trickery going on here. It's the only sane explanation for the astonishingly low die size, anyway...
GPU-Z says 32.
So it will probadly come with 5 SIMDs, each 160SPs, to keep die relatively small.
The open question are the Texture Units:
- 4 fatter ones with 8 TF each
or
- 8 thinner ones, without the useless sampler units
Geez GPU-Z doesn't mean crap. It's just a database...it doesn't read anything off the chip except clock speeds.
I wonder if this is related to the AMD video - http://www.theinquirer.net/gb/inquirer/news/2008/06/17/amd-piles-tera-ist-hype
I wonder if this is related to the AMD video - http://www.theinquirer.net/gb/inquirer/news/2008/06/17/amd-piles-tera-ist-hype
DegustatoR
Legend
Well, considering how this supposed "800 SPs / 40 TMUs" chip is doing somewhere around G92 territory in games, i'd say that there are some other possible explanations...
- Status
Similar threads
