but no evidence to believe othewise...
AMD: R7xx Speculation
- Thread starter Unknown Soldier
- Start date
- Status
DegustatoR
Legend
If it's AFR then it'll be affected by AFR issues. Maybe by a lesser degree thanks to an improved CF link but it will be anyway.
And up until now i haven't seen anything that suggests that it will be something other then ("improved") AFR solution.
Regardless of whether it is or not, I think most customers will be swayed by the average frame rate numbers.
I had an OC BFG version and it was clock at 500 and I could take it to 550 without modding the BIOS voltage settings. Memory didn't allow for alot of OC room as the HSF unit only covered the GPU.
I agree. The TDP of 4850 vs 4870 is a strong indication that the 4870 gpu is using a higher voltage. Also the functional units seem to be similar enough to rv670 to be designed for a similar clock rate (and after all rv670 reached a bit higher clock).
This is all true, but until there's confirmation it's AFR you could as well hope for something else
.Mintmaster
Veteran
Apples to apples, sure, but who is not going to be tempted by a 512MB card with similar performance at the $299 price point? The 320MB GTS sold quite well for this reason, despite the shortsightedness of such a purchase.
In any case, the point is that AMD has a huge cost advantage there. I'm sure a $300 4870 512MB makes a lot more money than a $400 GTX 260 896MB. Psychologically, I think 1GB ram looks a lot more than 896MB, so even at $375 it'll be much more attractive, IMO.
I was wrong in my performance projection, though, because I thought the 4870 would be 850 MHz instead of 750. Now we're looking at only 20% boost in clock. The bandwidth will help a lot, but not always, so I'm not as confident about it beating the 260.
Megadrive1988
Veteran
Yet, AMD might be able to claim R700 / 4870X2 CrossFire (4 GPUs) having ultimate performance over GTX280 Triple SLI.
The 4870 probably won't outright beat the GTX260 but seeing as how the GTX260 seems to get hurt w/ AA and AF increased compared to the 4850 (where in some games, it outright matches or beats it), I wouldn't be surprised if the 4870 was very close to the GTX260 in a lot of situations
When you mean 4850 you should probably say that instead of rv770, or do you expect the 4870 to be only on par with a 9800+?
and I think jimmyz @xs managed his 4850 at 700mhz, but he did re-seat the hsf. I don't think we'll see how good an overclocker it is until there's a tool that supports it beyond that.
In hardware.fr tests the 4850 was already beating the 260 in 5 of ~20 tests and I don't think it was ever very far back. I'd be surprised if 4870 wasn't at the very least competitive with it in the worst situations, while beating it quite often.
Oh yeah, my bad. Meant the 4850.
Well the 9800GTX+ might be quickly irrelevant if the G92 memory issues don't get fixed. The 4850 is slaughtering the 9800GTX once AA and resolutions go up as we've all known the 9800GTX really starts suffering in #'s once 4xAA and higher are on (along with other settings). If the G92b is just higher clocked, it might slow down its descent w/ AA and so on, but to what extent?
This is how they stack up at the maximum resolution / AA tested by techpowerup. Maximum AA level is 4x.
COD4: +4%
COH: +5%
Crysis: +0%
Quake Wars: +16%
Far Cry: -16%
FEAR: +11%
Prey: +20%
Quake4: +291%
SC3: +33%
STALKER: +15%
Supreme Commander: +5%
UT3: +3%
3dmark03: +6%
3dmark05: +5%
3dmark06: -8%
Its advantage at 8xAA is a lot more impressive than at 4xAA.
If I had to guess, it would seem AA is not being applied in Q4 just by looking at the numbers. Going from 1280 to 2048 res, the hit on the X2 and the 4850s would make it look that way.
I find their "diminishing returns" argument to be full of holes, BS marketing speak. There's a reason why CPUs hit a diminishing returns bottleneck, and it's because their workloads are primarily single threaded. Increasing ILP and clock for single threaded workloads does show diminishing returns.
On the other hand, 3D cards have workloads that are embarrassingly parallel. For any given semiconductor node, simply dropping more functional units on the same unit area will yield increased returns vs separate chip packages. Multichip configurations only have the advantage of increased aggregate memory bandwidth. From an economic standpoint, you might get better economy in terms of yields from 2 smaller chips, vs 1 uber chip as well (if the one uber chip is toast, you lose more)
However, in terms of performance, there is no "diminishing returns" IMHO from simply putting more and more on the same unit area as process shrinks occur rather than multichips. What ATI could have said was that there is a sweet spot in terms of cost per performance from going with smaller chips rather than one big difficult to manufacture chip.
The problem for ATI is that because they don't have the uber-big-chips, their marketing message will only be successful in the short-term as the 4870x2 goes up against the GT200. But as soon as NV shrinks it and tweaks it, and roles out the inevitable smaller cut-down versions, NV will have a story at the high end and middle of the market.
I just don't think the "we're only going to build medium sized chips" strategy is going to work in the long term, not when your competitor is willing to build and market medium, large, and uber-large chips, giving them more options.
On the other hand, 3D cards have workloads that are embarrassingly parallel. For any given semiconductor node, simply dropping more functional units on the same unit area will yield increased returns vs separate chip packages. Multichip configurations only have the advantage of increased aggregate memory bandwidth. From an economic standpoint, you might get better economy in terms of yields from 2 smaller chips, vs 1 uber chip as well (if the one uber chip is toast, you lose more)
However, in terms of performance, there is no "diminishing returns" IMHO from simply putting more and more on the same unit area as process shrinks occur rather than multichips. What ATI could have said was that there is a sweet spot in terms of cost per performance from going with smaller chips rather than one big difficult to manufacture chip.
The problem for ATI is that because they don't have the uber-big-chips, their marketing message will only be successful in the short-term as the 4870x2 goes up against the GT200. But as soon as NV shrinks it and tweaks it, and roles out the inevitable smaller cut-down versions, NV will have a story at the high end and middle of the market.
I just don't think the "we're only going to build medium sized chips" strategy is going to work in the long term, not when your competitor is willing to build and market medium, large, and uber-large chips, giving them more options.
Only if all resources are expanded proportionally with each increase in unit count. Otherwise you do indeed see non-linear scaling, and thus a "point of diminishing returns".
You've actually just described said "point of diminishing returns".
Mind you the GTX scores are also from a highly OC'd Liquid Nitrogen cooled setup! The Radeons at stock should beat the GTX's in tri-SLI. It is on the other hand much easier to cool 3 GPUs than 4...
There's also the issue of driver optimization, these things are much trickier for multi-GPU setups. I'll wager that ATI can probably squeeze a little more out of their drivers because R700 was supposedly designed from the ground up for scalability.
Last edited by a moderator:
One thing regarding that computerbase.de 8xAA numbers. Why AMD is not promoting 8xAA performance? 4850 is a mainstream card and IMO most buyers will not have uber res monitors so gaming at 1280x1024 or 1600x1200 will be very common. And according to those 8xAA numbers this card should shine in that kind of tests. They should really advertise that AA performance more.
- Status
Similar threads
