AMD: R8xx Speculation
- Thread starter Shtal
- Start date
3:0Make it 2:0
I'm voting MCM and board TDP of 190W (1GB RAM)
Not sure why they would need such a large board unless it is a MCM.
Jawed
Legend
The cooler on the first R600 was much bigger than the product eventually shipped with.
http://vozforums.com/showthread.php?t=18979
I know that's got the OEM bracket on it too, but still, the early cooler was pretty bloody pessimistic!
Jawed
http://vozforums.com/showthread.php?t=18979
I know that's got the OEM bracket on it too, but still, the early cooler was pretty bloody pessimistic!
Jawed
I was just assuming an X2 card made up of the MCMed cores.
Well, they were still trying to hit their clockspeed targets even with all the leakage.
Lucid?!?
More like master-slave relation. That is, the master GPU hosts the system interfacing (PCIe) with some kind of proprietary inter-chip hi-speed link with the other GPU.
In theory, the same could be done now with the CF-X port (RV770/790).
So 4 cores in total? I think 2x 6-pin wouldn't be sufficient for that.
But 2x 180mm2 128bit shouldn't require so big PCB. Maybe the 1200SPs part in "dual-core" configuration?
Yeah, I misread your post.
I thought you meant for the enthuiast cards, +$500.
I see now you are still referring to the RV870 performance card.
For some reason it still sounds like they have only 2 physical GPUs to manufacture.
RV810 and RV830/840(with rumors of RV830 being scrapped for a smaller GPU)
The upper cards are just made with multiples of the RV840, at least that's how I interpreted those recent rumors.
Jawed
Legend
Haha, been messing with Charlie recently?
frameavenger
Newcomer
^ 
Can't blame a reporter for trying to break some news. Nice try Chollie.
I personally think that the pictured card is 'rv870xt', which is an mcm package of 2x830/840...Whatever the displayed die of ~180mm2 is. If the shoe fits...
Considering the mobile information which puts the 'rv840' pro at 60W load, I think we'll see something like this:
rv840pro: < 75W (no peg)...60W
rv840xt: >75W (1x6-pin)...85W?
rv870pro: <150W (1x6-pin)...110W?
rv870xt: >150W (2x6-pin)...160W?
R800: < 300W (1x8pin, 1x6-pin)....Two mcm packages in typical 'x2' config.
Something like this makes sense because the XT models get that extra power (plug) in every instance, comparable to the next steps pro model, which could be used for higher clock speeds. It separates each market. IE, the 'pro' model would not have the juice to obtain even close to what the XT could obtain in clock speed. For instance, a pro model @ 750-800mhz and the XT @ 950-1ghz.
I figure 40nm was always meant to usher in the 900-1ghz core speeds. If rv740 would've worked out ideally, it would've been the rv670 to the r600. Not a huge performance bump, but essentially a re-do with a smaller power envelope and slightly less bandwidth. The ROPs especially benefiting from the higher clock, but even the cut down design yielding a superior product because of clocks. Obviously the problems with rv740/tsmc 40nm lead to that not happening and I believe us getting the 4750-spec as a 4770. I would not be in the least surprised if the initial mark for the high-end product was nearer the 1ghz marker with 5ghz GDDR5...such as was displayed with the launch slides. Obviously the massive power leakage kept that from happening (and probably from nvidia creating gt214) but still. Two things I find interesting:
1. Load power ideally at stock is 50W, as shown by xbitlabs. There must've been something severely fooked at TSMC for them to need to require a 6-pin as it has.
2. Every 4770 I've seen can obtain 1ghz using the stock cooler and voltage tweak.
I figure they had a plan, but when yields ended up in the can and/or with huge variable leakage, they just decided to release the thing with the power connector to cover their bases and let whomever could actually get one go nuts.
I think they have it all ironed out with 'evergreen', which I still think is a gpu (not a range of gpus), and have the tiers planned accordingly. That could certainly cover the $149-morethanIcanafford market.
Can't blame a reporter for trying to break some news. Nice try Chollie.
I personally think that the pictured card is 'rv870xt', which is an mcm package of 2x830/840...Whatever the displayed die of ~180mm2 is. If the shoe fits...
Considering the mobile information which puts the 'rv840' pro at 60W load, I think we'll see something like this:
rv840pro: < 75W (no peg)...60W
rv840xt: >75W (1x6-pin)...85W?
rv870pro: <150W (1x6-pin)...110W?
rv870xt: >150W (2x6-pin)...160W?
R800: < 300W (1x8pin, 1x6-pin)....Two mcm packages in typical 'x2' config.
Something like this makes sense because the XT models get that extra power (plug) in every instance, comparable to the next steps pro model, which could be used for higher clock speeds. It separates each market. IE, the 'pro' model would not have the juice to obtain even close to what the XT could obtain in clock speed. For instance, a pro model @ 750-800mhz and the XT @ 950-1ghz.
I figure 40nm was always meant to usher in the 900-1ghz core speeds. If rv740 would've worked out ideally, it would've been the rv670 to the r600. Not a huge performance bump, but essentially a re-do with a smaller power envelope and slightly less bandwidth. The ROPs especially benefiting from the higher clock, but even the cut down design yielding a superior product because of clocks. Obviously the problems with rv740/tsmc 40nm lead to that not happening and I believe us getting the 4750-spec as a 4770. I would not be in the least surprised if the initial mark for the high-end product was nearer the 1ghz marker with 5ghz GDDR5...such as was displayed with the launch slides. Obviously the massive power leakage kept that from happening (and probably from nvidia creating gt214) but still. Two things I find interesting:
1. Load power ideally at stock is 50W, as shown by xbitlabs. There must've been something severely fooked at TSMC for them to need to require a 6-pin as it has.
2. Every 4770 I've seen can obtain 1ghz using the stock cooler and voltage tweak.
I figure they had a plan, but when yields ended up in the can and/or with huge variable leakage, they just decided to release the thing with the power connector to cover their bases and let whomever could actually get one go nuts.
I think they have it all ironed out with 'evergreen', which I still think is a gpu (not a range of gpus), and have the tiers planned accordingly. That could certainly cover the $149-morethanIcanafford market.
Similar threads
