I wouldn't take it as "we can seamlessly connect two GPU's and make them act as one", but rather that they can have far higher bandwidth and lower latency crossfire, which would make implementation easier. But making it transparent to the devs? That would require next level driver work.I liked this comment:
Note: They didn't mention it. Not they didn't do it.
When he says "flavors of RX Vega will be faster" I take that to assume certain models will in fact have a higher clockspeed.. otherwise if its just optimization for RX then we would probably assume all RX would be faster?
But yes.. we will find out in 12 days.
You truncated the complete answer: "RK: Consumer RX will be much better optimized for all the top gaming titles and flavors of RX Vega will actually be faster than Frontier version!"
That doesn't mean that that RX version is higher clocked. I think there's a chance that 40% is because of better gaming drivers, 40% higher clocked HBM (512GB/s) and 20% that FE is cut down chip. But I think it's kinda silly to think that a cut down chip is more expensive than a full one and AMD would be ballsy to do that. But recently they have been pretty ballsy.
So they're seriously using the same fe moniker.... Lol....
They're basically adopting the FE moniker but without the gouged price or inferior cooling solution nvidia decided to use.So they're seriously using the same fe moniker.... Lol....
Sent from my Nexus 5X using Tapatalk
Yeah I think that's a great guess. However I think we will see 16GB cards much sooner than people are saying. Since they are already making 16GB Vega's it will be a non issue to simply release that sku as an RX with RX drivers.It's not just going to be drivers, because they've let "Pro" users use either the Pro or the "Gaming" drivers.
I'm guessing it's going to be higher clocked 8GB cards vs lower clocked 16 "Pro" versions.
They're basically adopting the FE moniker but without the gouged price or inferior cooling solution nvidia decided to use.
I suspect that the consumer chip will have higher clocks, which has been true historically. It will probably have a higher clock on the memory as well as I'm assuming that the 8-Hi stacks can't be pushed as far, or not if AMD wants to hit production targets, so it may be silicon lottery on memory OC of FE cards. With the 4-Hi stacks they can probably be more picky, or just pair the memory that can't clock as well with the cut chips. Pro cards have always been about the drivers. The silicon itself isn't going to be much different, other than the pro cards may be binned for efficiency.
I suspect AMD isn't fully finished with the gaming drivers yet so there's more performance to be gained there. They may also not wish to tip their hand to NVidia too much either. The benchmark against P100 is probably Vega in the best light possible, but there's no reason that a game optimized towards Vega's strengths couldn't also see something like 30% better performance over a 1080 Ti. Either way I suspect that in 2 years that's where Vega will probably be, more because NVidia is going to let Pascal drivers rot once Volta comes out than anything AMD does.
Personally, I'm more excited about what AMD will be able to do with Vega in their APUs. If Vega is capable of hitting 1600 MHz without breaking the bank in terms of power draw, imagine how efficient it would be running at > 1200 MHz in an APU. Even an 8 CU design would offer a substantial amount of performance (> 1 TFLOP) within a small power budget. That would put it around the RX 550 in terms of performance which can pull 100 FPS at 1080p in a lot of popular "e-sport" titles. Add Freesync to the display and you'd have a pretty damn sexy notebook.
And a phenomenal little micro box for your TV that can also handle a nice range of game content.
I don't know much about AMD's APUs and their pricing, so where do you think it would land in pricing compared to Intel?
So Fudzilla reported there's a delay and the performance is not on par with 1080Ti: http://fudzilla.com/news/graphics/43666-amd-to-announce-vega-tonight
I remember this disturbingly similar to Fiji launch, ct say I'm happy to see this happens again. I just want less cringe inducing PR campaign and solid products.
How will HBC effect content creation to 3d artists?
Does it help in any way in 3d programs? How so?
A: Having 16GB of HBC on board will allow 3d artists to work on larger and even more complex models than ever before. Depending on the workload we have seen scenarios where 16 GB of HBC is effectively same performance as having 32 GB or 64 GB of regular VRAM
Q: Awesome tech with the HBCC, one thing i want to ask about it is if its needed for it to be cooded for each new game to function? Like DX 11 Crossfire profiles? Or does it work by itself no matter game/program and do not require extra cooding?
A: To realize the full potential of HBCC, yes we will need to see content from game developers use larger datasets. But we have seen some interesting gains even on current software, particularly in min frame rates. Part of the goal of launching Radeon Vega Frontier edition, is to help speed up that process.
It's really depends on what the specs end up looking like. My thinking has always been a 4C/8T Zen CPU and 8 CU (512 SP) Vega GPU part. That strikes a good balance and probably lets them span $125 - $250 range depending on bins. They could have some as high as $350 if they get some nice low leakage parts that can run all CPU/GPU cores and hit the same ~45W TDP that the best Intel mobile quad cores fall into. They'd have similar CPU performance, but could probably easily double or triple the GPU performance over what a lot of the Intel mobile chips offer.
Some posters here have thought as many as 12 CUs, which to me seems like overkill because it pretty much obsoletes the 560 and anything below it if the clocks are reasonably good. On the other hand, having a wider GPU part allows them to drop the clock speeds by 33% while maintaining the same overall performance. I don't know to what extend they're willing to trade die space for power use, but supposedly Ryzen yields have been really good so they may see it as worthwhile.
Vega APU will probably do more for AMD in terms of market and mindshare than desktop Vega does, even if it is a Titan killer. Every fall there's a lot of new college students buying notebooks and a lot of them want to have something that can do casual gaming as well. AMD has a big opportunity here to get new buyers attached to their brand.
They're basically adopting the FE moniker but without the gouged price or inferior cooling solution nvidia decided to use.
I think we will see 16GB cards much sooner than people are saying. Since they are already making 16GB Vega's it will be a non issue to simply release that sku as an RX with RX drivers.
I wouldn't take it as "we can seamlessly connect two GPU's and make them act as one", but rather that they can have far higher bandwidth and lower latency crossfire, which would make implementation easier. But making it transparent to the devs? That would require next level driver work.
Is 16 GB even necessary in a consumer card? From the demo's that were shown, the HBCC makes a fairly large difference even at 2 GB. Unless this card offers some monstrous 4k gaming potential where it's 60+ FPS for a majority of titles, I don't see much benefit in trying to future-proof a purchase with more than 8 GB of memory.
So Fudzilla reported there's a delay and the performance is not on par with 1080Ti: http://fudzilla.com/news/graphics/43666-amd-to-announce-vega-tonight
I remember this disturbingly similar to Fiji launch, ct say I'm happy to see this happens again. I just want less cringe inducing PR campaign and solid products.
Aren't mobile and desktop APUs the same, just different packaging and power limits? I.e. they'll more than likely have the same number of cores and CUs, cache and the like, just configured differently.Zen/Vega APU performance numbers were already given by AMD:
+100% perf/watt
+50% CPU performance
+40% GPU performance
-50% power
Thats mobile or are you talking about desktop APUs?
I wouldn't take it as "we can seamlessly connect two GPU's and make them act as one", but rather that they can have far higher bandwidth and lower latency crossfire, which would make implementation easier. But making it transparent to the devs? That would require next level driver work.
Nice comparison. The savings arent huge in terms of die size but still better than the convential GDDR technology.
This is very ignorant especially when the buisness case is there for nVIDIA to produce these GPUs even with horrible yield rates (we dont know for sure but assuming it will be very low). They have massive super computer contracts to meet.
Just take a look into their latest datacenter revenue numbers. They aren't stupid.
But this is purely off topic
This is good, but AMD is going to need to be proactive. It's not enough to just throw the tools out there and hope they get used, they have to actually send engineers out to customer sites to help them port their code. As the underdog, they've got to try harder if they want to pick up market share.
Yes, pretty much my thoughts.I don't know. I think developing a multi chip solution that's transparent to the system and game developers is exactly where they're heading.
I'm just not sure Vega or even Navi will be capable of that however.
Vega will be very fast, but compared to what? Fiji? Polaris? or GP102?