No, it's because each ALU is capable of a MADD (2 flops) per clock.
AMD: R7xx Speculation
- Thread starter Unknown Soldier
- Start date
- Status
No, it's because each ALU is capable of a MADD (2 flops) per clock.
http://www.tomshardware.com/cn/382,news-382.html
HD4870 & HD4850 at first only w/ 512MB (1GB cards later), RV770 PRO ~9800GTX, RV770 XT ~30% faster. Engine clk: TBD (like i said before).
HD4870 & HD4850 at first only w/ 512MB (1GB cards later), RV770 PRO ~9800GTX, RV770 XT ~30% faster. Engine clk: TBD (like i said before).
SEP in Q3?
Only because of the baseline architecture (R600) proved to be inferior to G80.
Prior to R600, ATI never implemented this dual GPU rubbish (minus the Rage Fury MAXX--but that's almost irrelevent now).
I guess we have to wait for ATI to move out of the shadow of R600 before they start competing on the ultra high end, monolithic GPU market.
You mean what R680 should've been, instead of the HD3870 X2?
- 55nm
- 256 bit bus
- 512MB-1024MB GDDR4
- 480 stream processors
- 32 TMU's
- 16 ROP's
- Universal clock domains akin to R600/RV670
Hm...Looks sort of like RV770, don't ya think?
I dont think AMD lost any market share after the RV670 launch.
http://www.digitimes.com/mobos/a20071228PD207.html
Also the numbers from JPR take longer to reflect the impact. It not something like the steam survey.
Agreed and agreed.
I don't think AMD will return to the high-end, monolithic mega-chip market, at least not anytime soon. For now, they've chosen the approach of offering an affordable mainstream solution with a reasonable price/performance ratio and a CF-on-a-card solution for the higher end. Now there are two possible outcomes:
1) this strategy will be successful, earning AMD money. No need to change it, then.
2) it won't be successful and AMD will continue losing money. And nobody will go designing a huge GPU if they're broke.
1) this strategy will be successful, earning AMD money. No need to change it, then.
2) it won't be successful and AMD will continue losing money. And nobody will go designing a huge GPU if they're broke.
Nvidia has a good point though. Making multi-GPU work seamlessly is dependent on high-speed off-die communication and a shared memory pool. The worst case performance for multi-GPU solutions today is just too low. Then again we've all learned not to worry about what Nvidia says in public.....
Well, without wanting to facetious or anything... if multiple CPUs was such a great solution why do we have 2GHz CPUs rather than 10 x 200MHz CPUs, and why were multi-core CPUs that last resort once clock scaling went wrong? If getting good performance out of multi-core CPUs was easy it would have been done already, and for CPUs multi-core is pretty much the same as multi-die-single-core when it comes to the software side (cheaper to build of course but that's integration for you).
GPUs are different for reasons which we're all familiar with, however they're now reaching an analogous hard-limit on their single-die scaling. But I don't see how multi-die is somehow going to magically become less undesirable or magically less of a problem now than it's been in the past. It runs counter to pretty much the whole ethos of large-scale integration which has driven the industry for forty years, and it makes the software more difficult too (which is the killer in my humble opnion). If SLI/X2/GX2 call it what you will was the right thing to do to give the consumer price/performance it would have happened already and we'd all be running 32-way SLI <think of a GPU from four years ago>.
I think what I'm getting at here is that being forced into a situation where you have to build multiple independent non-integrated units which aren't able to communicate with each other as efficiently and effectively as if they would if they were integrated really shouldn't be portrayed as win for anybody. Overall it's a worse solution for everyone. The consumers will end up buying "cores" that most of their software can't make effective use of, just like they do now when they buy dual-core and quad-core PCs because "it makes the Internet smoother". The games developers spend more time getting the code to work, time which the end-user has to pay for.
The first problem is how to implement the NUMA configuration between multiple discrete chipsets.
I'll believe the 30% faster than 9800GTX when I see it in real world actual played benched games and not synthetics, fly-bys, walkthrus or cutscenes.
- Status
