I'm anticipating atleast 2 clock domains for GPU and CPU and maybe more for a memory controller etc. I wonder if they'll be able to clock the GPU higher since they're using a SOI process now. Also, does anyone know if they're integrating the southbridge into this die or will it be off chip for now?
Fusion die-shot - 2009 Analyst Day
- Thread starter AnarchX
- Start date
brain_stew
Regular
This should bring back a bit of life into PC Gaming
Enough horespower to play any console port on APU alone
A 480SP DX11 ATI GPU being the low end is going to prove a massive boon for PC developers, what a stepup from Intel integrated graphics! :smile:
With even the low end parts shipping with such a beefy graphics solution, surely GPGPU development has got to look a lot more viable?
I wonder if AMD could modify one of these chip's for Nintendo's next console? Surely the prospect of a single chip solution with low R & D costs has got to be tempting for Nintendo (heck, a similar sort of offer looks to have won Nvidia, Nintendo's mobile contract)? Say, hack off two of those CPU cores, attach 1GB of GDDR5 and you've got a pretty compelling 720p gaming machine, no? Should be a nice enough enough upgrade over a PS3/360 and an insanely huge upgrade over the Wii, yet with a 28nm process available fpr a 2012 launch it should prove pretty damn cheap and again, most of the R & D work is done already. Having a standard modern DX11 architecture has got to be a big win for Nintendo as well, it'll make ports much simpler and developers will be very happy to be working with something familiar. UE3/CE3/Unigine/id Tech 5 could all be ported over in no time at all meaning great results from the get go. A very comprehensive BC solution could be provided through software emulation, given the gulf in processing power.
Heh, a bit of a tangent there, I know, its just with a few slight modifications, these Fusion processors could make a great little console on a chip.
Last edited by a moderator:
The big reddish die in the first post is not showing the complete chip -- the DDR interface in the right edge is at weird "length" (I counted 64+32 bit data pads) as if the chip has being sawed off of the bottom line. That would really imply double the visible SIMD width of the IGP multiprocessors for a total of 480 ALUs (6*80).
brain_stew
Regular
I'd definitely like to see how Crysis fairs on one of these. If ATI can deliver on the promise of "Crysis without a dedicated GPU" with decent quality (say high settings/720p/~30fps) then I think they're onto a winner, that's a pretty nice USP to have at the low end over Intel. You get to play all PC gaming has to offer with AMD's solution with fairly nice "better than console" quality without any bulky dedicated GPU which your Dell/HP/Asus desktop/laptop may not ship with or instead may make it difficult to install.
CPU performance should be more than enough for the average home user, even if Intel offer faster performing CPUs at the low end, the average home user just isn't going to see any tangible benefit from that, but the ability to play any PC game on the market in some fashion or another (and it would seem surprisingly decent quality) is something that the average consumer can immediately identify with and benefit from straight away and its easy point to market.
CPU performance should be more than enough for the average home user, even if Intel offer faster performing CPUs at the low end, the average home user just isn't going to see any tangible benefit from that, but the ability to play any PC game on the market in some fashion or another (and it would seem surprisingly decent quality) is something that the average consumer can immediately identify with and benefit from straight away and its easy point to market.
I have a hard time seeing the point. People who care about gaming will want a much better GPU anyway (Especially in 18 months when this thing is out!). People who dont care about gaming, well, dont care.
I only see a point if GPGPU takes off in such a big way, which it very well could I guess. Then, having a GPU attached to the CPU for GPGPU stuff, would be excellent. In a way just a different multicore CPU strategy. You could think of the SP's almost like Cell SPU's, really, then. Of course theyre far away but you get the idea.
I only see a point if GPGPU takes off in such a big way, which it very well could I guess. Then, having a GPU attached to the CPU for GPGPU stuff, would be excellent. In a way just a different multicore CPU strategy. You could think of the SP's almost like Cell SPU's, really, then. Of course theyre far away but you get the idea.
Yep - it then matches the picture in post #18
The bandwidth dilemma will still haunt the IGP solutions. Unless some major shift in the platform happens, like wider memory interface (expensive SKUs and boards), or adopting faster DDR/XDR tech, the main target market will remain focused on casual gaming. Such solutions would benefit moslty closed custom platforms, like game consoles, where proprietary hardware additions (eDRAM frame-buffer) could complement the IGP, to unlock it's full potential. The discrete GFX solutions are here to stay for some more time.
brain_stew
Regular
There's probably somewhere around a billion PC gamers out there,sure that includes people that play solitaire once a month and is spread over an incredibly diverse range of games and delivery methods, but the point still stands, the majority of people that have a computer do some gaming gaming in some shape or form at least periodically. A massive chunk of those aren't ever going to be savvy enough or want to lay down extra cash on discrete hardware dedicated to gaming but if they have it already, what's to say a lot of them won't use it?
The way the market works atm most people with a PC are locked out from any game released in the last 3/4 years if it isn't a PopCap title. This changes all that, as AMD's low end consumer should have enough GPU grunt to play any PC game in some fashion at least, quite possibly more often than not at more than satisfying settings for most, especially since AMD plans to refresh the GPU packaged each and every year. That opens up PC games (and GPGPU applications) to a much bigger market which can surely only be a good thing?
Countless have tried to run WOW on Intel integrated graphics for example but its such a terrible experience I doubt many persist, if such customers instead buy an AMD chip in the future they get to play WOW but not lose any other tangible real world functionality at all. I spend a fair amount of time offering PC hardware advice to a very (primarily console focused) popular gaming forum, and yes, a lot of people on low end notebooks/desktops do ask if they're capable of playing modern games, it might just be for the latest Valve release, WOW a Train Sim or C & C but the only constant is that anyone with a current integrated solution is in for a shitty experience if they don't upgrade their graphics hardware, and since many use laptops or awkward prebuilt PCs this is often not an option for them, so they're totally SOL.
Not to be stubborn, but if there's such a demand for this, why wouldn't companies just make better IGP's already?
Where's the advantage to Fusion, over simply better IGP's? There may be a few here and there, maybe save a bit of cost here or there, but I dont see anything major. And what you lose is flexibility, because it's fixed. A motherboard design could upgrade to a better IGP in a more modular fashion.
Myself, if I could have a 8 core CPU, or rather a Fusion with 4 cores and a low end GPU integrated for the same cost, I'd certainly prefer the former, I can worry about the GPU later.
Just some thoughts.
Why were there quad core CPUs with 4GB of memory when most consumers only email and browser and few apps used 2 threads, let alone 4-8?
Marketing.
Intel has a much stronger position with OEMs and a powerful CPU (even if way overkill) is easier to justify that a GPU that is "for gaming." Consumers have a hard time paying $50 extra for a "gaming" feature yet how PCs are tiered your typical system, as it gets nicer (more memory, larger HDD, bigger display, etc) also scales with the CPU as a basic metric. A nicer computer needs a nicer CPU (a lot of people call computers CPUs afterall).
As for users expanding their systems, this isn't something most consumers can do.
But would that 128MB be enough if the chip really has that 480, or even just 240 ALUs?
Comapnies who? nv, out of market, amd, going for fusion, intel igp's, give me a break, they'll land in cpu's anyway.
It could save some power in mobile market, especially versus the midrange-discrete-mobile gpu's. Another advantage is almost zero latency in the cpu-gpu communication.
brain_stew
Regular
Its a perfectly valid point but the main counter I would bring to you is where was the motivation to improve IGPs before? We 're talking about super low margin parts, where volume and low cost is king. It was in ATI and Nvidia's best interests to gimp integrated graphics to a certain degree to encourage sales of their low end discrete cards which bring larger margins. Sure it wasn't in the interests of consumers but with low performing integrated graphics solutions in the best interests of all the serious players why would things change?
AMD have identified graphics as a core strength and a genuine USP that they potentially have against Intel in the CPU market. If they can increase the proliferation of GPGPU then their CPUs are going to be able to handsomely outperform Intel in a wide range of applications. If the hook early on is that for the same price as Intel's low end solution you get genuine gaming ability for the same price then that's a pretty marketable idea. Integrated graphics themselves may be low margin, but CPUs generally command a higher margin (yeah, I know this hasn't necasarily been the case with certain products lately) and if a nice integrated product helps sell that higher margin product then it suddenly makes sense to improve it, no?
The whole single chip dynamic changes a lot, and honestly it sounds like something that marketing would be able to work well with. A 3ghz quad core fusion CPU with a 480SP DX11 GPU, is definitely something with marketing potential.
Surely the concept would just be like "turbocache" and "hypermemory" cards of old? The game or application would be able to dip into the main system memory pool (which its directly connected to already, right) once its filled up its memory cache. A 64/128MB GDDR5 cache, would be cheap and probably provide a nice speed boost. It could be an optional extra for system integrators to add but is an extra dedicated 64 bit GDDR5 memory channel really going to increase costs a great deal? I don't know how feasible this is in reality, I'm still very much an absolute amateur when it comes to this stuff, but I'm learning and its no less interesting/exciting.
I think the mobile market is where a chip like this has huge potential, so every watt saved counts. Currently notebooks with anything better than Intel integrated graphics see a significant increase in price. Yet notebooks with a low end discrete GPU as their main differentiator routinely sell for $100-200+ more than their integrated counterparts, and in numbers. AMD will absolutely be the only solution for low end or thin and light notebooks with any sort of workable graphics solution because Intel need to rely on a separate discrete card for parity, that means higher costs, bulkier machines and worse battery life. Honestly, notebooks is where this solution just makes a lot of sense, a $500/$600 laptop with a quad core and a 1/2 teraflop DX11 graphics solutions in a standard 14" laptop casing sounds pretty appealing to me.
Last edited by a moderator:
Sideport was always useless when used exclusively (apart from saving power), so 128MB shouldn't be much of a problem. Besides, by that time surely 2gb gddr5 chips will be out, so 256MB with one chip.
Those 20GB/s+ AnarchX claims though would require over 6Ghz gddr5 ram which seems like a stretch on that timeframe for a IGP (and it is still less than what system memory will provide - assuming dual-channel ddr3-1333 will be used).
Yes, unless you use it in exclusive mode (most bios have option for this). So far though sideport just barely made a difference, since fastest you could get was around 16bit ddr3-1600 (3.2GB/s).