AMD Radeon VII Linux Benchmarks - Powerful Open-Source Graphics For Compute & Gaming

[artivision]

NVIDIA:

DLSS fraud, bad visuals and low performance gain with half to one year training:

https://www.youtube.com/watch?v=Jpd5j5W1NZw

DXR fraud, PetaFlops are needed for Full Effect so only for some reflections with a lot of noise and performance hit wile not always present:

https://www.youtube.com/watch?v=jaUP4LucmZM

HW fraud, extra non shader units are outdated today just see "Primitive Shaders" and:
https://www.nextbigfuture.com/2018/0...any-cases.html
https://www.nextbigfuture.com/2016/0...aster-for.html

D3D11 fraud, GameWorks call's low overhead extra Nvapi extensions, AMD and DXVK can pick up some of them.

SALES fraud, Pascal had 2xFP16 for servers but not for gamers.

Make your selves a favor and do not buy.

[Brisse]

and washers to tighten the grip

While I've actually done this myself on a card that was clearly defective and had literally zero mounting pressure, as an engineer I still have to point out that doing this on a correctly designed card can potentially have destructive consequences. Gaining a couple of °C in return for a cracked GPU or PCB is not worth it.

[finalzone]

I’m actually impressed that AMD has a very competitive product! Further with this being launch day testing on fresh drivers we could potentially see improvements down the road.

Micheal
The OpenCL numbers are a surprise, especially considering release day drivers. In my mind this warrants further focused testing for people concerned with compute performance. Part of that focus should be reliability testing and even some attempts at underclocking.

Remind that the test is done with open source driver which is impressive.

[Templar82]

I certainly wasn't expecting it to compete with the 1080ti/2080. It's been a long time since AMD were this competitive in pure performance and I imagine it will only improve with further driver tweaks.

[caligula]

The OpenCL numbers are a surprise, especially considering release day drivers. In my mind this warrants further focused testing for people concerned with compute performance. Part of that focus should be reliability testing and even some attempts at underclocking.

Not really surprising. The hardware is really powerful. Nvidia uses some algorithmic tricks to boost perf. The cooling can be improved though. They say they managed to improve tjunc by 10°C with better paste and washers to tighten the grip. In addition, under voltage operation used 50W less power.

[theriddick]

Sometimes optimisations won't help if a game heavily favours nvidia hardware over AMD,
i.e. F1-2017 game runs a bit slow for AMD under Linux, oddly so.

[caligula]

I really hope they could still optimize the drivers. Would be fair to compare this with 2080 ti & nouveau. After all nvidia drivers are proprietary blobs.

[nightmarex]

I was expecting to read this and be like, "meh" and ended up "Oh crap... now I want one".

[theriddick]

Some nice results, a real shame no board partner cards are coming because the PCB of the RAD7 can certainly be shrunk down to a mini size to fit in a ITX system much like what happens with NVIDIA's cards.

Hehe, its even a little more efficient then my 1080TI Mini card, oh how I want a Rad7 MINI card now... if only AIB existed for this launch... sad

[TheYoshiGuy]

Will be interested to see if; like the Vega56, Vega64 and the Vega APUs, this card will also hard-lock the system when trying to run console emulators.