• 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.

Poll: ** The AMD VEGA Thread **

On or off the hype train?

  • (off) Train has derailed

    Votes: 207 39.2%

  • (on) Overcrowding, standing room only

    Votes: 100 18.9%

  • (never ever got on) Chinese escalator

    Votes: 221 41.9%
  • Total voters
    528
Status
Not open for further replies.
"
What this means is that basically all of the pessimistic rumors about Vega have basically been confirmed. You know, the ones we shrugged off on this sub as unbelievable for months since they implied a performance regression from Polaris and 28nm architectures:

  • Gaming performance between a 1070 and 1080
  • 28nm-level power consumption
  • 550mm2 die size
I had no idea it was possible to create a new architecture that regressed in perf/mm2, perf/watt, and perf/FLOP. Something has to be wrong here since these results defy all intuition.

"

From reddit, AMD FIX THIS1!1!111!!!
 
AllBodies said:
Yeah I think all we can do now is assume it's a driver issue (and/or a stepping issue on the first run of cards), and wait for the RX gaming release.

If that isn't the explanation, then it means AMD has managed to make around 8 Billion transistors (or 45%+) of the chip absolutely useless. And that just doesn't sit right with me.
Click to expand...


Some of those are down to that hbc thing they have on the chip as well, or possibly quite a lot of them. Remains to be seen if it can actually do what theyre preaching, fury x drivers had to be hand tuned for high data loads on a game by game basis and that never worked out well at all. This seems to be more of an automated system from the looks of it.
 
Gerard said:
Some of those are down to that hbc thing they have on the chip as well, or possibly quite a lot of them. Remains to be seen if it can actually do what theyre preaching, fury x drivers had to be hand tuned for high data loads on a game by game basis and that never worked out well at all. This seems to be more of an automated system from the looks of it.
Click to expand...

True but that specific issue you're raising was only in situations where the FuryX's 4GB vRAM wasn't enough.

And HBCC is also meant to kind of be an automated solution to that. They've claimed it's a driver thing, doesn't have to be directly implemented in the engine, and all it does is hugely assist in performance in vRAM-limited scenarios. But Vega shouldn't run into that very often as 8GB of vRAM is plenty for now.
 
five8five said:
On this link from Anandtech they're saying that tiled rendering is working correctly -

https://forums.anandtech.com/thread...on-benchmarks-updated-june-30.2508928/page-19
Click to expand...

The people "confirming" it are not PcPer staff, or anandtech staff, or AMD though.

PcPer themselves said it's not working; and they're waiting for a response from AMD still.

In the videos and tests, it's done normal rendering like other GCN cards.

Here's FIJI rendering exactly like vega.
https://gfycat.com/InsecureEagerKingbird <------

1vUrQ2K.gif


qi4xRkJrQuaF6XrJu6n22g.png
 
Last edited:
five8five said:
On this link from Anandtech they're saying that tiled rendering is working correctly -

https://forums.anandtech.com/thread...on-benchmarks-updated-june-30.2508928/page-19
Click to expand...

zornyan said:
yeah looks like it, obviously different implantation than Nvidia which is why it's less efficient.

so, literially all rx vega has to look forward to is maybe 5% or so better performance from drivers?
Click to expand...

No they're misunderstanding tiled-rendering.

All they're seeing is it's 'different' to Tonga/Hawaii etc.

All it's doing is rendering the triangles from the opposite side first, and each section is larger (due to larger L2), but it's still the old method.

TBR is where it'll render an entire section first, all colours, before moving on to the next bit. So by the time it's finished the whole triangle, all layers are completed. What it's doing now is completing a complete triangle in one colour, and moving onto the next colour and doing a whole triangle.

Watch this whole video for a very complete explanation, and proper examples of Nvidia doing it correctly: https://www.youtube.com/watch?v=Nc6R1hwXhL8&feature=youtu.be&t=6m50s
 
I just love How people here after Fury X still belive it will magicly gain 40% speed :D:D:D:D:D:D


Its like Drivers will fix it OFC they will in 2019 :D

You guys do know that many people still not passed 3000mhz on Ryzen right ?? Using DDR4 rated ABOVE 3000 :D
 
Lord_ZED said:
I just love How people here after Fury X still belive it will magicly gain 40% speed :D:D:D:D:D:DD:D::D
Click to expand...

So you're 100% certain that Vega with 50% clocks, 50% more TFlops, claimed improved IPC on compute units, and a 70% smaller process than Fiji; should only be able to barely match an overclocked Fury X?
 
five8five said:
On this link from Anandtech they're saying that tiled rendering is working correctly -

https://forums.anandtech.com/thread...on-benchmarks-updated-june-30.2508928/page-19
Click to expand...

zornyan said:
yeah looks like it, obviously different implantation than Nvidia which is why it's less efficient.

so, literially all rx vega has to look forward to is maybe 5% or so better performance from drivers?
Click to expand...
If you had kept reading you would have come across this

https://gfycat.com/InsecureEagerKingbird
https://www.reddit.com/r/Amd/comments/6kdwea/vega_fe_doesnt_seem_to_be_doing_tiled/djleqwq/
Click to expand...
 
zornyan said:
yeah looks like it, obviously different implantation than Nvidia which is why it's less efficient.

so, literially all rx vega has to look forward to is maybe 5% or so better performance from drivers?
Click to expand...
It doesn't make sense though. How can it be lower IPC than Fury ffs. If it really is this bad then AMD have massively screwed up here.
 
DragonQ said:
It doesn't make sense though. How can it be lower IPC than Fury ffs. If it really is this bad then AMD have massively screwed up here.
Click to expand...

Well I asked PcPer if they can do some clock to clock tests against the Fury X; they obliged me after asking for FE tear downs; so hopefully we get that.
 
DragonQ said:
It doesn't make sense though. How can it be lower IPC than Fury ffs. If it really is this bad then AMD have massively screwed up here.
Click to expand...

Doesn't Pascal have slightly lower IPC compared with Maxwell to allow for the really high 2000mhz+ clocks? Maybe the same thing is going on here.
 
Jono8 said:
Doesn't Pascal have slightly lower IPC compared with Maxwell to allow for the really high 2000mhz+ clocks? Maybe the same thing is going on here.
Click to expand...

AMD claimed improved IPC though; it was on their slides for their "new" compute units compared to old GCN.

Also at the moment it looks like vega is nearly 30% slower IPC wise than Fiji. :/
 
Status
Not open for further replies.
Back
Top Bottom