but it wouldn't have the ability to provide the PCIe lanes required. It would need a new chip. Something like an FX steamroller, which was supposed to come out Q1 2014 but AMD have kind of... well they've gone silent about it. Like it probably won't happen and we might all forget they ever published a road map with it on.
http://www.asus.com/Motherboards/SAB...990FXGEN3_R20/
asus did pcie3 on AM3+ allready
Whilst PCIe 3.0 might be important in tri/quad SLI setups (and probably more so in crossfire since they've binned the bridges) the number of people using those types of systems must be in the low six digits across the entire world. And the number of them that would bother with quad SLI and then not want the absolute fastest CPU performance (i.e. not AMD) I'd be surprised if it even got into 10s of thousands.
For 'enthusiast' grade systems single card, or at most dual card @1440p is still the benchmark, GPUs need to get much faster before pcie3 is neccessary. Even in triple head setups the difference is pretty slight.
The games which tend to need high framerates are FPS games when you come to think of it,and you will realise where Mantle support is first coming out for,
Mantle supports is already confirmed by some of the biggest devs,like DICE,Activision and Crytek. Now look at the engines that two of them use,ie,CryENGINE3 and Frostbite 3 which will be used on multiple next generation titles,and idTech5 from id software runs well on AMD FX CPUs(FX8150 is faster than a Core i5 2500K). The last two major engines will be Source3 and Unreal Engine 4,and the preceding engines already threaded reasonably well,so they will probably will run reasonably well on AMD CPUs. Moreover,even games like Skyrim run perfectly fine on older CPUs,and Skyrim Online is going to use a multi-threaded engine.
PS2 has gone from a lightly threaded game to a one which can use multiple threads just in the last week,and the peformance boost is amazng.
The reason?? A PS4 version is being released.
The thing is a lot of the older engines will be gone in the immediate future as many games are multiplatform,and the new consoles are DX11 compatible. A lot of these very engines are only DX9 too. The ones remaining will be used in games which give good framerates with older hardware,ie,targeting many casual gamers too,not hardcore ones with decent hardware.
Even PC exclusive games like LOL,DOTA and Minecraft which are lightly threaded run perfectly fine on even older CPUs. Its the reason they are popular,ie,they are scalable. You get decent framerates at realworld settings with a range of hardware.
Last edited by CAT-THE-FIFTH; 20-11-2013 at 12:38 PM.
minecraft is on xbox
interesting:
AMD has more PCIE lanes available than Intel - AMD has 32 whereas (including PCH) intel has 20 , and they all cant be full speed , as its limited by the DMI.
heres a test.
can someone with a haswell board run a heavy 3d benchmark , AND do a big data transfer over a GB nic? , having a thought something has to throttle.
My apologies - I didn't pick that up. I don't know if the APU will use some of the PCIe lanes internally if the GPU isn't disabled, but even if it does KAveri would still be able to offer a full 16 pcie2 lanes to a single card. The fact that you can crossfire with the APU demonstrates that its possible to have both that and a PCIe GPU enabled.
Problem with trying to shift anything from FM2+ to AM3+ is that AM3+ requires Hypertransport links, whjich simply aren't there on the Kaveri die*. So it's impossible to use Kaveri silicon with AM3+ without a lot of messing around. It's more interesting that the BD/PD FX dice apparently have PCIe lanes on them, as this *might* make it possible to use PD FX silicon on FM2+ (which I'd guess would be AMDs preferred solution, converging on a single consumer platform). But since it seems to be pretty much a foregone conclusion that FM2+ will be superceded by a new DDR4-supporting platform in a year or so, it wouldn't make a lot of sense to move FX from an existing platform nearing EOL, to another existing platform also nearing EOL.
AMD have pretty much always used the same silicon for desktop and server parts. If server DDR4 adoption is starting to approach critical mass, AMD are surely going to start looking at creating server parts with DDR4 memory controllers. The last time this happened (DDR2/DDR3) they went for a hybrid controller, but AFAIK that's just not practical with DDR3/DDR4, so they're going to have to drop DDR3 support at the outset. That only really leaves 3 options for FX: create a new enthusiast chipset with DDR4 support, design a separate Steamroller die with DDR3 memory controller, or keep serving AM3+ with old parts for a year until the new consumer DDR4 chipset is ready. Which of those do we think will see the best ROI?
*EDIT: interestingly though, to get an 8 core part out of Kaveri you'd have to do a 2-chip MCM, and that would have 32 PCIe 3 lanes available: 16 from each kaveri chip. Interesting, n'est pas?
Would this actually happen though? ie a real world usage example would be? BF4 gaming with backup to NAS in the background? Would that be heavy enough data transfer? If you're gaming you're gaming, and on my machine that means everything else shut down, and all net ports blocked except those the game needs to run.
PLX is essentially a switch - it doesn't magically give you more PCIe bandwidth - if you have two cards saturating their lanes, each will be left with what they would've had in the first place, and probably with a bit of extra overhead loss from the PLX.
In reality though, you won't get a noticeable drop in gaming performance until you go down to something like 4x PCIe 2.0 (even less in most games). PCIe 3.0 might be good for bragging rights, but there's no realistic reason to factor it into a buying decision for a gaming system.
http://www.techpowerup.com/reviews/I...caling/23.html
Anandtech did a good review of plx mobo for haswell. http://anandtech.com/show/6170/four-...cs-and-evga/26
AMD CPUs behaved differently to NVidia with single AMD GPU configurations taking a 3-7% hit on fps versus non plx. Nvidia single 580 was "indifferent". For both AMD and NVidia multi GPU got a boost with plx though it varied depending on how it was implemented by the manufacturers. power consumption takes a hit, as does price as plx chips are $40+
There seems to be a couple of recurring bits of confusion here.
1/ Chipset has nothing to do with memory support any more. A DDR4 supporting FX CPU could happily use the existing chipset.
2/ To do an MCM of two CPUs you need a way of tying them together. The Pentium 4 used the front side bus which was designed to easily make 2 socket motherboards. Opterons use HyperTransport as that is what it is designed for. All an FM2 part has is PCIe, which people have used for an inter CPU network but you can't do NUMA over it AFAIK, it requires additional electronics to switch the PCIe, and latency sucks.
The only thing that would worry me about FM2+ being the way forward is the 100W limit on those motherboards.
There are currently 1 users browsing this thread. (0 members and 1 guests)