liekly true, there is already rumours on Vega20 that ill need to slot in before Navi so Vega 10 might not be around for long
-
Competitor rules
Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.
AMD VEGA confirmed for 2017 H1
- Thread starter Doom112
- Start date
- Status
liekly true, there is already rumours on Vega20 that ill need to slot in before Navi so Vega 10 might not be around for long
looks like you are right. makes me curious how they will market the Mi25 in that case.
the half-precision is very useful for deep learning, the double precision useful for HPC. The Tesla Gp100 does both very well, at a price.
Aye, it is rather odd to say the least; all they've shown and mentioned is FP16 use cases and performance.
If they don't have the FP64 performance to compare with Tesla, it might mean they try and compete on price and FP16 alone.
I haven't been paying any attention to developments with gfx cards for a good while now so I'm wondering if there is a feeling that these new cards might be worth paying attention to? I'm currently using an R9 390, but am seriously considering jumping ship to the green team because I feel like AMD's driver support is painful (still no support for asynchronous reprojection)
I think if right then it will be all about the price, I would take 1080 performance if the price is right
I mean, if this is the case, then no, gamers are not going to go for it. What AMD needs is fresh and impressive reviews on Day 1. Them implementing tech that may or may not come good later on has been the cause of many of their problems in the recent past. As always, their problem is that by the time the tech does come good, there will likely be much more powerful hardware available from a competitor. This is great for existing customers who want something that remains viable, but not so great for getting people to buy in the first place.
GTX1080 performance for around £400 would be awesome.
I just dont know that this leaves enough room underneath for a cut down model, and especially not a lesser GPU.
I think have to imagine that Vega will reach higher than that. Maybe not quite 1080Ti, but it's possible it can get quite close if it really is 500mm² or so as estimated. Maybe even a bit better depending on how good the Vega-specific improvements are. This for £500-550 would probably go over pretty darn well, with a cut down version more along the lines of a regular 1080 for like £400-430 or so. I dont think anybody could complain in that case.
I would just imagine they'd want something to directly rival the GTX1070, which is probably a much bigger seller than the 1080 or 1080Ti.
So I dunno. I'm a bit confused as to what Vega will be, and what a potential Vega 11 would be and where exactly it would fit in. I hope AMD have planned all this out really well.
I would genuinely be surprised if Vega and the top vega for this year only comes out just a bit faster than a 1080. The MI25 card is clocked at 1526 MHz base clock. And from what ive seen looks passively cooled (not sure as it's supposedly around 300 watt TDP???) That is about 12.5 Tflops of performance. So with a actively cooled card id expect it to hit higher clocks with boost. The 1080 is what 9 Tflops so there seems to be a massive difference in theoretical performance. So id expect it to land closer to the 1080ti which is 11.3 Tflops but probably not beating it. Well not when the 1080Ti is overclocked anyways. As the 1080ti overclocked must be hitting near 12Tflops
At current prices, cutdown Vega needs to be something like 5-10% faster than a GTX 1070 for under £350. Otherwise it'd be a big disappointment, considering the time difference in release date.
I still find it hard to believe large Vega won't be pretty beastly. Polaris is already around 10% IPC better than previous GCN (i.e. the FuryX).
So even if Vega was only another 5% IPC on Polaris, you're looking at a FuryX with 50% more clockspeed, and 15%+ more IPC. That's going to be very good by most people's standards, and the touted upgrades to the Vega architecture should bring much more than a couple % on Polaris.
Overclocked 1080 Ti is actually over 14 Tflops. 2050 MHz x 2 x 3584 cores.
Dennis? What you doing in here?
When GTX 1080 Ti is OC than it is around 13.5-14 tflops
However, Tflops comparison is useless because GTX 980 Ti is around 6 tflops and beats Fury X ,which is at 8.6 tflops.
Now there is a post of yours I can agree with sir
They'll have to improve a whole lot in maximizing potential, especially in DX11, if we want to predict performance based on FLOPS.
I mean, a GTX970 is 3.5TF, while RX480 is 5.8TF but only a fraction faster in most games.
980 Ti is more like 7.5 Tflops running at real boost clocks. 6 Tflops would only be running at about 1060 MHz
Also much of the disparity in Fury X performance is part of the whole point in the Vega architectures upgrades (so AMD tout anyway). The FuryX couldn't properly utilise all its 4096 cores, due to inefficiencies in the design, which Vega is meant to fix.
Also the Fury X is starting to beat the 980 Ti in some new games with the new APIs anyway.
You can compare GTX 980 with R9 390X.
No matter AMD does with GCN still it will still require around 20% more raw performance to compete with Nvidia alternative. They need to change the whole design from top to bottom not just bumping and few new stuff.
Which, assuming these improvements result in 1:1 performance gains, would give a healthy lead over a GTX1080 and would rival the 1080Ti in titles that heavily favor AMD cards, which there are more and more of nowadays.
Pretty much what I'm *realistically* expecting, so yea, I agree it should be quite good, so long as the price is right. £500 is my guess for an 'aggressive' price given HBM will probably limit how aggressive they can actually go.
Still puts them in a bit of a pickle as to what to do with the larger £300-350 market that Nvidia is dominating with the GTX1070. A cut down Vega at £350 would potentially be a relative bargain.
Which exactly the what the NCU powering the Vega is. Brand new architecture. Not another GCN iteration.
It is not brand new. It is GCN. NCU is computing unit not shader core.
Permabanned
- Joined
- 15 May 2006
- Posts
- 4,107
- Location
- London
- Status