2018 15" MacBook Pro (RP560X) [8th,6C,H] + GTX 1080 Ti @ 32Gbps-TB3 (ASUS XG Station Pro) + Win10 [OliverB] // Radeon VII vs GTX 1080 Ti  

1. Workload and computing benchmarks.

a) GPGPU

b) LuxMark, GeekBench and SPECviewperf13

	Radeon VII	GTX 1080 Ti
GeekBench OpenCL

c) Comments:
- The raw calculation numbers of Radeon VII are much better than the GTX 1080 Ti. Especially the double float performance is impressive. Scientific programs like numeric simulations will profit immensely by this.
- Luxmark profits most of this "computing power".
- Amazingly in "application benchmarks", the GTX1080Ti looks equal, if not better.
- Interesting note for benchmark "SNX-03" of SPECviewperf13: This is very slow on both cards, nVidia Quadro K6000 scores with 150 (vs. 14 or 16) here. See also here.
- I haven't posted any Video codec benchmarks, because I haven't solved the mystery yet, why AMD eGPU don't use full GPU power when encoding. See also here. nVidia with NVENC is the winner here in any case.

Tie: 1 - 1

2. Synthetic Benchmarks

	Radeon VII	GTX 1080 Ti
Unigine Valley Extreme HD
Unigine Superposition 1440p Extreme
3dMark Time Spy
3dMark Firestrike
3dMark 11

C0mments:
- This is another tie: 1 - 1

Hi @OliverB, before running gaming benchmarks. I’ve questions, I saw this also on several benchmarks that MacBook Pro physical score is pretty low. Is that Apple EC firmware setting or cpu thermal throttling happens during benchmarks. 8850 that has higher frequency clock than  8750H theoretically should perform better. Maybe by undervolting it’ll be better?
Good side as always MacBook Pro 15 graphics score is decent compared Laptops with dgpu, I’ve headache catching that
https://www.3dmark.com/spy/6594762
https://www.3dmark.com/fs/16420889
Edit: for dx12 update i think i saw in nvidia Forum that Radeon VII will update for better performance like on timespy, maybe you want to check that as well.

3. Gaming Benchmarks

External display: Resolution 2560 x 1440, Max details, typical AA
Internal display: Resolution 2880 x 1800, Max details, No AA

.

Comments:

- There is a clear winner: The GTX 1080 Ti
- It has to been noted that the EVGA GTX 1080 Ti SC2 is an overclocked card, while the Radeon VII is reference.
- I removed "Alien Isolation" and "Resident Evil 6" from the Average. With them the result was 86.8 to 100.
- The results often do not reflect benchmarks of Desktop PC. It looks as Radeon VII has more FPS drops in eGPU setups.
- The first five titles a major and current, so benchmarks on the internal display have been made, too.
- All titles have built-in benchmarks, I am looking to add some others.

Great posts @oliverb ! Unless I’m reading them wrong (hey it’s possible!) the VII is pretty much on par with a 1080ti, lagging a bit in gaming but ahead in other tasks. I wonder how much extra gaming performance Wattman could get out of the card? My Vega 56 once tweaked gained 10+ FPS in all the games I tried it in.

Posted by: Eightarmedpet

Great posts @oliverb ! Unless I’m reading them wrong (hey it’s possible!) the VII is pretty much on par with a 1080ti, lagging a bit in gaming but ahead in other tasks. I wonder how much extra gaming performance Wattman could get out of the card? My Vega 56 once tweaked gained 10+ FPS in all the games I tried it in.

@eightarmedpet, thank you. I haven't tried a lot tweaking with Wattman yet, once I click "Auto undervolting" and the system crashed, aha. lol
Don't have the time, but I will keep one Radeon VII for at least 14 days and want to try it. (I am having two RadeonVIIs at this moment, running perfectly in a Asus XG Station Pro, which will be for sale in the next weeks, does Asus give me a bonus for doing this?) 

To be fair, in brand-new titles the Radeon VII looks much better than in older title. I tested some new titles like Resident Evil 2 Remake and Dreamfall Chapter with MSI Afterburner and that was amazing:. Highest Settings 1440p more than 135 FPS.
But: Everytime the Radeon VII is performing great, it's getting very loud on load, about 57 dbA on 20 cm distance and open enclosure. That is said just for information, because I always said I don't care about some noise on load when the thing is silent on idle work and it is silent then.

@oliverb I think for testing sake you should send me one Asus enclosure and one Vega VII so I can independently verify it’s performance. 

Im happy to do this for free. I’ll pm you my details 😉

(on a serious note- still interested in the enclosure).

@eightarmedpet,  hehe,

the second Asus enclosure will be for sale, that is 100%, the day Asus service, aka fools, is returning the one where they have to change the power cables.

Are you planning to buy a Radeon VII? How much are they in UK? Here you won't get nothing under 750 EUR. :/ (That was a double negative, hehe)

I am tempted by the VII but realistically it would be overkill for my 2017 dual core MacBook so I’m going to try and hold off for Navi and see what that brings. 
I have just bought a node lite to put a tb3 card in to accelerate my 5k directly which will keep me busy for a bit...

It's all about programming and efficiency: Radeon VII makes in newest games double of FPS compared to older ones. The performance with Vulkan is amazing.

It is tempting... the sapphire model is 650 on Uk amazon. I’ll see if I can resist. From what itsage says the AMD drivers are better optimised for internal displays too, which is how my ultrafine is handled.

The gaming benchmarks have no finished yet. There are still some games without built-in benchmark, and in 6 out of 10 games the Radeon VII is doing great: Here are already the numbers for the Radeon VII, the GTX 1080 Ti scores are yet to be made.

.
The issue "Wolfenstein The Old Blood" looks like an TB3-issue, too.  (Edit: It's not) Only when you hold still and thus no new data is sent to the GPU the FPS and GPU usage rises to over 60(%). (Edit: That is something, the GTX 1080 Ti doesn't do).

Edit: It doesn't look good for Radeon VII.. GTX 1080 Ti is stronger in every game of this list I have tested so far. 🙁 Furthermore, it never has those "only 50% of GPU is used" issue. It's always well used.

I am not seeing the Radeon VII perform any better than the ROG Strix Vega 64.  In fact, it runs a little worse.  I will admit that I am running with my internal LCD, only.  I have a 2018 MBP with the Vega 20 GPU.  My guess would be that the internal display is acting as a bottleneck at this point.  I tried Heaven, Superposition, Shadow of Mordor, Ashes and Civ VI.

It's surprising how similarly the games play, even when I overclock the Radeon VII up to 2GHz.  One other thing to comment on, the card is running rather quiet, with the junction temperature at 88 degrees, at most.  I think if the read-back to the internal display is the bottleneck, that could explain why the GPU is quiet, too... it's never pushed hard enough.

@brian_williams, it's a big disappointment. I was getting some hope after seeing big numbers over 100 in some new games, but as a matter fact of fact my GTX 1080 Ti has even bigger numbers there, plus there are not those horrible performance like with Doom: Radeon VII = 37 FPS, GTX 1080 Ti = 160 FPS... No questions anymore, your honour.

Could you, @brian_williams, please make some benchmarks with your eGPU Vega 64 in 1440p, Max Details in
- Doom
- Shadow Warrior
- Wolfenstein The Old Blood

Best to measure with MSI Afterburner, and some running around in the first Level.

3. Gaming Benchmarks

External display: Resolution 2560 x 1440, Max details, typical AA
Internal display: Resolution 2880 x 1800, Max details, No AA

(Note: Shadow Warrior New Order and Old Blood are the same engine. New Order was measured indoors, Old Blood outdoors)

@oliverb did you test The Division 2? I'm having great performance running it in 3440x1440 Ultra settings.

My Vega 56 performed far better than my 1070 with the Division one, so wonder if the Division games just run better on and. 
I set up my tb3 add in card last night and did a super quick test, looks like a massive improvement in FPS but I need to double check that (slim possibly the res changed automatically).

I did some testing with The Division 2, tonight.  Finally a real measurable improvement, especially once DX12 was enabled.  With my Vega 64 I was measuring 51 to 55 fps at the starting area, the yard behind the house.  With the Radeon VII I was measuring 55 to 63 fps.  I was running with the internal display at 2560x1600, vsync disabled.

With the Radeon VII I was able to overclock to 2GHz at 1075mv, but it only gave me an additional 2 to 3 fps.  I think I would choose power, temperature and silence over just a few frames per second, especially within a GPU enclosure.  The Radeon VII was still very quiet compared to even the ROG Strix's custom fan controller.

@juan_c_yunis, @eightarmedpet, @brian_williams,

Thank you for your contribution. I would like to add Division 1 and 2 to the Testing. Those are online games, arent' they? So I have to join them? Are there demos, where I can test and so do some benchmarks?

@oliverb I think they have offline versions/modes too - I don't think I have ever played The Division online, but then again I have never played for more than 5 mins to test it (its on my "to play" list).

I have finished my benchmarks of 25 games in Resolution WQHD, 4k, 5k and interal display (3k) and this the result:

Radeon VII: 4 games
GTX 1080 Ti: 21 games

Some of the 21 titles were not even close, 5 of the 21 games the GTX 1080 Ti had double or more FPS compared to the Radeon VII.

I can say with certainty: The Radeon VII may be a decent GPU, but it's far away from the performance of a GTX 1080 Ti, at least in gaming. Further, it has hardly any advantage in 4k or 5k, what you would expect for the larger memory.

What's even worse than the fact that the nVidia GPU won in most games: The horrible performance of the Radeon VII in some titles. It has some serious issues in large areas of some games, frames drop to 10 FPS. Needless to say that the GTX1080Ti renders the exact same situations fluently. It may be a driver issue, but this is already bad enough. It's really a pity for this otherwise most interesting card.

@itsage, if you want I can update the numbers with all results in this FPS table, but I guess most people are only interested in the final result rather than a lot of numbers.

@oliverb great job man, but don't forget to mention that this performance degradation only applies when the VII is used in a eGPU box.

Posted by: Juan C. Yunis

@oliverb great job man, but don't forget to mention that this performance degradation only applies when the VII is used in a eGPU box.

This may be true, but the same applies for the GTX1080Ti. It was also "captured" in exactly the same eGPU box and so suffered a performance degradation, too.

It would be interesting to compare in MacOSX, but the cards don't share one MacOSX version were both are supported. I could compare it with my GTX 780 Ti in Mojave...hehe

@oliverb thanks for running the benchmarks! Wow the numbers look really bad for the Radeon vii for gaming in egpu. Do you think it could be because Xconnect is not optimised for Radeon vii yet?

Been mulling to revert to high Sierra + 1080ti and put the Radeon vii in the desktop (which is connected to a 4K60 tv).

Posted by: Kelvin Tan

@oliverb thanks for running the benchmarks! Wow the numbers look really bad for the Radeon vii for gaming in egpu. Do you think it could be because Xconnect is not optimised for Radeon vii yet?

I don't think that is has something to do with Xconnect as it's only very bad with particular games. Some major games run really good (e.g. Resident Evil 2 Remake, Far Cry 5 and Wolfenstein 2), but some are a bad beyond belief (e.g. Doom in general, Crysis 3, Wolfenstein Old Blood). The problems look special (Doom OpenGL FPS are less than 1/3 than expected and crushes in Vulkan, Crysis and Wolfenstein Old Blood the performance drops to 10FPS in open areas. etc...)

Been mulling to revert to high Sierra + 1080ti and put the Radeon vii in the desktop (which is connected to a 4K60 tv).

As I said I tested most games in 4k and 5k, too (I just have discovered VSR/DSR), no difference. Radeon VII is owned by the GTX1080Ti in most titles. :/

Yeah just that on my TV i'm capped at 60fps anyway, but actually have a gsync ultrawide (Acer Predator X34P) so been mulling to just switch over. Problem right now is... I can't seem to download High Sierra for the 2018 MBP any more. 😥

Posted by: Kelvin Tan

Yeah just that on my TV i'm capped at 60fps anyway, but actually have a gsync ultrawide (Acer Predator X34P) so been mulling to just switch over. Problem right now is... I can't seem to download High Sierra for the 2018 MBP any more. 😥

My monitor has Max 72FPS and I am not sure if more FPS really improve the experience. For eGPU setups the FPS is limited anyway.
I think it just not possible to revert from Mojave to High Sierra, that's why I still stick to High Sierra even though it badly wants an upgrade :/

@kelvin_tan Try using installinstallmacos.py. The build shown should be uniform for all Macs (due to security updates that end up usually making the build numbers for Macs congruent).

@mac_editor thank you! If it works I’m gonna just hoard the high Sierra installer USB like crazy haha. Will try tmr, much appreciated mate!

Edit: It works! Downloading the packaged files right now. Thanks!

Edit2: no-go, the building of the image fails on my MBP2018 (2.6/32GB/1TB/560X) as it's not compatible. Looks like we are out of luck on the 2018 MBPs to revert to High Sierra.

Some addition about overclocking: I tried to undervolt and overclock and Radeon VII, I could win some performance by overclocking from 1810 Mhz to 1920 Mhz. A bigger increase made the system instable. I retested the games where Radeon VII lost closely to GTX1080 Ti and there was just one game where it could close the gap:
Return of the Tomb Raider, 1440p MSAA, Preset Very High: 87,3 FPS. But on 4K, it lost again by a clear margin. So still the GTX 1080 Ti performs better in that game, too.

Last word on this: Radeon VII is still a powerful card, but it's not as good in gaming as a fast GTX 1080 Ti (e.g. EVGA). This is a marketing hoax. The Radeon VII would have great value if it was fully supported in MacOSX (like the Vega 64), but this is not the case. Yet.

Here are the complete results of 26 games tested in 1440p (with AA), 20 games in 4k and 5k, and 10 games internal display (3k).

The final results are: (Radeon VII  vs EVGA GeForce GTX 1080 Ti SC2)
.

Graph

Comment.
It Is still a good result for the Radeon VII. Despite of issues in some games it yields a good average for 60,2 in 4k Ultra settings. GTX 1080 Ti SC2 is 20% better though, but not in synthetic benchmarks. There both cards are much closer which brings the hope that with improvement of drivers the Radeon VII could close the gap between them.

In case somebody cares about the Radeon VII performance:

I could reproduce the issues I am having with some games with a complete other card:

In a benchmark comparison between GTX Titan Black (6GB) and GTX 780 Ti (3GB), which only differ in memory (when gaming is concerned), I got the same effects with the GTX 780 Ti on some games at high settings @1440p. It's identical to the" Radeon VII Effect": You hold still, FPS are ok, you move, the FPS go down to 10 FPS.

Here it is expected as 3GB is simply to low, but the Radeon VII with 16GB? Looks for me a driver problem who misuses/overuses the memory.