AMD Vega 10, Vega 11, Vega 12 and Vega 20 Rumors and Discussion

Perhaps. But from what I'd seen Pascal is no champion regarding the overclocking. Maxwell was.
An GTX 1080 has an overclocking potential of ~ 10-15% - not many cards goes over 2100Mhz (up from 1860Mhz- the 'real' boost clock); And Vega will probably be overclockable too ( up too a certain point...)

 

Ummm... you're already comparing it to an overclocked Vega.

 

And whats the point in overclocking Vega, unless one needs to warm his room?

 

That guy has done OCing, undervolting and more. Also doing a 1.1v @ 1600

 

Later on in the thread he got 1692/1100 (not at 1.1v though).

Don't know if he just got an especially good one or not.

Regards,
SB

 

So, that looks like about a 20% frequency increase at the same voltage compared to Polaris (Polaris seems to usually hit 1.1V@1350Mhz or so). If you use the stock numbers, the picture doesn't really change. I suppose that's pretty ok, though by comparison Pascal chips easily reach 1.0V@2Ghz (yeah I know different architecture and all, but surely low voltage is key for good energy efficiency).

 

Wonder what the variation on HBM2 looks like, considering all the samples we've seen reach 1100MHz, and yet the base clock is 945.

 

AMD and Nvidia have routinely released drivers that have given double digit gains in the past. Why do you have such a hard time believing?

Furthermore, RX Vega isn't out yet. For all we know it could be a quantum graphics card powered by fairy dust and hershey bars...

 

http://www.gamersnexus.net/guides/2988-vega-fe-watercooled-tear-down-and-internals

So it's not wasted space

 

Yes it was interesting. It seems pretty costly too. Maybe it's ok for a "semi pro" card. I wonder if they will go back to a "simpler" solution like on the Fury X for the RX versions.

 

In single applications or games, sure. If something's amiss, a directed fix can help. But across the board? I have a hard time remembering drivers that achieved this, if any. When the respective miracle drivers were released, each IHV compared their performance gains not to the immediate predecessor, but to many month old sets of drivers, thus accumulating individual fixes for games that were not out when the baseline driver was released.

WRT to „believe“: I believe, that whoever (and thad includes all IHVs, see my debate over GV100's supposed 50% perf/watt improvement) has nothing to hide, does not throw up smoke screens and mirrors. Otherwise, believing is for churches.

 

Seems like it would cost about the same as Fury X. The only meaningful difference appears to be that the coolant storage tank is metal (I can't find a breakdown of the Fury X storage tank though - maybe that's metal?). The hoses are longer, which is good as Fury X's hoses are a bit short. Vega's power supply circuitry covers more board area, so that metal coolant channel in Vega is also longer.

The parts count is pretty much identical from what I can see.

 

eek2121, I live just down the road from Hershey, PA and yum those bars are good BUT we are talking gpus and from what I see, RX Vega is going to be a work in progress for quite sometime.

 

Maybe AMD overestimate their man power to develop this architecture(too short dev. time) or maybe at the middle they switch the focus to Navi or whatever reason but Vega doesn't seem ready to be out. And tbh for the N time Navi must be a big hit because Nvidia even with its avaricious policies is not doing the Intel it is innovating in every generations and AMD needs to make a ryzen if they really want to compete with them in equal conditions(performance/watt/area).

Maybe now that ryzen is out and doing well AMD finally have the relief in pressure, focus and money to put into Navi and make it their ryzen in GPUs.

 

Nice! At the risk of going off topic, I lived in Blairstown NJ and Newton NJ for 10 years. Both of those towns aren't that far from there.

 

http://www.gamersnexus.net/news/1387-nvidia-drivers-337-vs-mantle-aggressive

There is also the issue that it seems like many of the new hardware / architecture features don't appear to be enabled / working properly, as tests have shown the tiled rasterization doesn't appear to be working and memory bandwidth is much worse than Fury X while it should be about the same w/o the features, and much faster with them working.

Anyway, its only a few weeks away now, and AMD might even give us a sneak peak during their tour: https://radeon.com/en-us/rx-community-meetup/

 

Are you saying that memory bandwidth is driver dependent?

 

Everything on hardware is driver (or bios) dependent, because it doesn't know how to function. There is something wrong with Vega's memory bandwidth from the testing done which might be caused by some of the hardware not being properly utilized.

Why do you think the memory bandwidth is much worse even though the architecture was supposed to have better memory bandwidth than Fiji?

 

Raw memory bandwidth isn't, but effective memory bandwidth certainly is.

 

I'd hope raw bandwidth too, considering how absolutely pitiful it is on Vega. 250GB/s is around 100GB/s less than Fiji.