AMD Vega Hardware Reviews

[ieldra]

Yeah I can monitor HBM temps, since I already talk about them in my post By the way, I move from blockchain to beta 17.8.2 for FE (which are 17.8.1 in reality I believe), and I can control frequencies and voltages too now with wattool.

You mean this benchmark ? https://www.guru3d.com/files-details/final-fantasy-xiv-heavensward-benchmark-download.html
It won't really be "stock" for me, since I've already a waterblock on it. You just want scores ? Or temps too ?

I'm curious to see how it performs after your tweaks, and sorry I must have missed the HBM temps in your post going back to it now

 

[Rootax]

So, with 17.8.2 beta, 1677mhz core @ 1.150v , +50 power target, 945mhz for HBM2 (the gpu peaked et 41c, the hbm2 50c, everything is maxed out at 1440p) :

Spoiler: results :

Benchmark FINAL FANTASY XIV : Heavensward
Résultats mesurés le 10/09/2017 16:47:10
Score : 12019
Nombre moyen d'images par seconde : 91.049
Évaluation : Excellent
-Votre configuration matérielle vous permettra de faire tourner le jeu de façon optimale avec des performances graphiques excellentes à tous points de vue. Vous pourrez utiliser tous types de paramétrages sans aucun problème.
Temps de chargement :
Scène 1 0,992 sec
Scène 2 2,762 sec
Scène 3 3,249 sec
Scène 4 2,372 sec
Scène 5 4,052 sec
Scène 6 0,981 sec
Total 14,409 sec

DAT:s20170910164710.dat

Taille d'écran : 2560x1440
Mode d'écran : Plein écran
Version DirectX : 11
Préréglages graphiques : Enregistrés 1
Qualité graphique
-Afficher les effets des surfaces mouillées : Activé
-Élimination des objets cachés : Désactivé
-Réduire le niveau de détail des objets distants (LOD) : Désactivé
-Reflets en temps réel : Maximums (DirectX 11)
-Anticrénelage : FXAA
-Qualité lumière transparente : Élevée
-Détail des herbes : Maximum
-Détail du décor : Élevés
-Reliefs de l'eau : Élevés
Affichage des ombres
-Soi : Afficher
-Autres : Afficher
Ombres
-Réduire le niveau de détail des ombres (LOD) : Désactivé
-Résolution des ombres : Élevée : 2048 pixels
-Cascade des ombres : Optimale
-Adoucissement des ombres : Fort
Qualité des textures
-Filtrage des textures : Anisotrope
-Filtrage anisotrope : x16
Mouvements physiques
-Soi : Activés
-Autres : Activés
Effets de l'écran
-Assombrir naturellement les bords de l'écran (centre-bord) : Activé
-Ajouter un flou autour des objets en mouvement (flou radial) : Activé
-Occl. ambiante d'espace d'écran (SSAO) : HBAO+ : Élevée (DirectX 11)
-Reflets lumineux : Normaux
Scènes cinématiques
-Activer la profondeur de champ : Activé

Configuration système :
Windows 10 Professionnel 64 bits (6.2, version 9200) (15063.rs2_release.170317-1834)
Intel(R) Core(TM) i7-5820K CPU @ 3.30GHz
16296.500MB
Radeon Vega Frontier Edition (VRAM 3072 MB) 22.19.0666.0001

Ce logiciel ne garantit pas que votre système vous permettra de jouer aux versions Windows de FINAL FANTASY XIV : A Realm Reborn et de FINAL FANTASY XIV : Heavensward.

Site officiel de FINAL FANTASY XIV : Heavensward http://fr.finalfantasyxiv.com/pr/
(C) 2010-2015 SQUARE ENIX CO., LTD. Tous droits réservés.

With 17.8.2 beta, everything "stock" clock, voltage and pt wise, 945mhz for HBM2 : (the gpu peaked et 40c, the hbm2 48c, everything is maxed out at 1440p) :

Spoiler: results

Benchmark FINAL FANTASY XIV : Heavensward
Résultats mesurés le 10/09/2017 17:00:29
Score : 11019
Nombre moyen d'images par seconde : 83.715
Évaluation : Excellent
-Votre configuration matérielle vous permettra de faire tourner le jeu de façon optimale avec des performances graphiques excellentes à tous points de vue. Vous pourrez utiliser tous types de paramétrages sans aucun problème.
Temps de chargement :
Scène 1 0,891 sec
Scène 2 2,845 sec
Scène 3 3,296 sec
Scène 4 2,408 sec
Scène 5 2,354 sec
Scène 6 0,943 sec
Total 12,737 sec

DAT:s20170910170029.dat

Taille d'écran : 2560x1440
Mode d'écran : Plein écran
Version DirectX : 11
Préréglages graphiques : Enregistrés 1
Qualité graphique
-Afficher les effets des surfaces mouillées : Activé
-Élimination des objets cachés : Désactivé
-Réduire le niveau de détail des objets distants (LOD) : Désactivé
-Reflets en temps réel : Maximums (DirectX 11)
-Anticrénelage : FXAA
-Qualité lumière transparente : Élevée
-Détail des herbes : Maximum
-Détail du décor : Élevés
-Reliefs de l'eau : Élevés
Affichage des ombres
-Soi : Afficher
-Autres : Afficher
Ombres
-Réduire le niveau de détail des ombres (LOD) : Désactivé
-Résolution des ombres : Élevée : 2048 pixels
-Cascade des ombres : Optimale
-Adoucissement des ombres : Fort
Qualité des textures
-Filtrage des textures : Anisotrope
-Filtrage anisotrope : x16
Mouvements physiques
-Soi : Activés
-Autres : Activés
Effets de l'écran
-Assombrir naturellement les bords de l'écran (centre-bord) : Activé
-Ajouter un flou autour des objets en mouvement (flou radial) : Activé
-Occl. ambiante d'espace d'écran (SSAO) : HBAO+ : Élevée (DirectX 11)
-Reflets lumineux : Normaux
Scènes cinématiques
-Activer la profondeur de champ : Activé

Configuration système :
Windows 10 Professionnel 64 bits (6.2, version 9200) (15063.rs2_release.170317-1834)
Intel(R) Core(TM) i7-5820K CPU @ 3.30GHz
16296.500MB
Radeon Vega Frontier Edition (VRAM 3072 MB) 22.19.0666.0001

Ce logiciel ne garantit pas que votre système vous permettra de jouer aux versions Windows de FINAL FANTASY XIV : A Realm Reborn et de FINAL FANTASY XIV : Heavensward.

Site officiel de FINAL FANTASY XIV : Heavensward http://fr.finalfantasyxiv.com/pr/
(C) 2010-2015 SQUARE ENIX CO., LTD. Tous droits réservés.

--Texte pour réseaux sociaux--
Type 1
http://sqex.to/ffxiv_bench_fr #FFXIV Score : 11019 2560x1440 Enregistrés 1 DX11 Intel(R) Core(TM) i7-5820K CPU @ 3.30GHz Radeon Vega Frontier Edition
Type 2
http://sqex.to/ffxiv_bench_fr #FFXIV Score : 11019 2560x1440 Enregistrés 1 DirectX11 Plein écran Radeon Vega Frontier Edition
Type 3
http://sqex.to/ffxiv_bench_fr #FFXIV 2560x1440 Enregistrés 1 DirectX11 Score : 11019 Excellent
Type 4
http://sqex.to/ffxiv_bench_fr #FFXIV 2560x1440 Enregistrés 1 DirectX11 Plein écran Score : 11019
Texte long
Benchmark FINAL FANTASY XIV : Heavensward
Score : 11019 Excellent
2560x1440 Enregistrés 1 DirectX11 Plein écran
Intel(R) Core(TM) i7-5820K CPU @ 3.30GHz
Radeon Vega Frontier Edition
http://sqex.to/ffxiv_bench_fr #FFXIV

With stock, the gpu was switching a lot between 1348 , 1440 and 1600 mhz. With the "custom" setting, it was 1677mhz all the time.

There is not a big différence in temps either. My watt-o-meter is not plugged in right now, so I can't provide power consumption at the wall for now. Oh and, I've a 5820k@4.2ghz 1.2v in the same loop.

 

[Rootax]

Has anyone tested how the Adam demo from unity runs on Vega? I've tried searching but found nothing so far, could anyone with an rx vega or fe give it a run out of curiosity?

Download link to Adam demo.

Stock or after tweaking ?

 

[doob]

Stock is fine but if you got time for both even better

Although you might need some monitoring tool running in background like afterburner or wattman to log the frame times or fps, as i don't know if it that demo has a performance result at the end of it, can't remember, but i think it doesn't .

 

[Rootax]

Aaah, I though it was a bench, not "just" a demo :/

 

[Entropy]

Vega 56 here.
Anyone that cares about quiet computing would do well to avoid the stock cooler option.
However, depending on game and settings, the card may not run at full tilt, and may then fall into a power/thermal bracket where the cooler isn't obnoxious. Undervolting doesn't do a lot to help peak noise, as it typically allowes the card to run within the same power envelope but without throttling. In order to consistently run quiet, you need to reduce peak power draw to 100W or so, (as measured by GPU-Z) which takes some doing.

 

[entity279]

But switching BIOSes? The power saver BIOS helps?

 

[Lightman]

But switching BIOSes? The power saver BIOS helps?

Yes, as it limits peak fan speed. Instead of limiting power to 100W it's much better to limit fan profile and let card manage power. That gives you desired noise levels at best performance, especially if combined with undervolting. Modern power management is really sophisticated and there are many knobs you can turn to make your card run as you want it to.

I'm planning on going under water with my Vega at some point, but for now I tolerate a bit of noise. Card is a lot quiter and better than old 5870 and 290X stockers so not all is bad, but coming from 3rd party triple fan Fury, there is substantial gap in percived noise.

 

[itsmydamnation]

The noise doesn't bother me that much because the noise is low pitched.

I am also wondering just how good/bad the llp process is. I see similarities with Zen and Vega in terms of voltage wall even when power/heat isn't an issue. Is this just advanced power Management at play or does llp process just suck on the high clocks/voltage side? What was polaris like in this regard?

 

[Lightman]

The noise doesn't bother me that much because the noise is low pitched.

I am also wondering just how good/bad the llp process is. I see similarities with Zen and Vega in terms of voltage wall even when power/heat isn't an issue. Is this just advanced power Management at play or does llp process just suck on the high clocks/voltage side? What was polaris like in this regard?

I have similar observations. This pricess likes coldness much more than volts to clock better. Proper cooling is really important if you plan to OC to high clocks.
I've lowered target GPU temp. by 5C and the card is much happier to stay stable at low volts, but obviously it made it more noisy.

 

[Entropy]

Yes, as it limits peak fan speed. Instead of limiting power to 100W it's much better to limit fan profile and let card manage power. That gives you desired noise levels at best performance, especially if combined with undervolting. Modern power management is really sophisticated and there are many knobs you can turn to make your card run as you want it to.

Wouldn't changing the fan limits in WattMan achieve the same results? I'll test it tonight. I'm already undervolting a lot, so letting it throttle to keep within thermal limits might work well without introducing an irritating on/off/on/off behaviour. Or not. Interesting experiment either way!

 

[Deleted member 13524]

I thought the 2nd bios was a "safe mode" bios, hence why it can't be re-written and the clocks are somewhat lower.

 

[Grall]

2nd BIOS isn't flashable? Hurm, well, yes that makes sense...

 

[Rootax]

Holly crap FE doesn't like HBM2 OC.

I can have 1050mhz in benches / stress tests, like futurmark, Unigine Heaven, valley, saturation, for hours... But it crashes in Xcom2, even at 985. I guess the 8gb modules are not that oc friendly, or, I have a "bad batch", or, well, If I only do benches like some guys for testing the stability, I would say it's perfect at 1050, which is wrong in the end...

For the core, I'm at 1677 (stable, with +60%PT. I can do "just" 50% with 1.15v, but I put 60% when I was at 1.175, and I didn't change that later) @1.150v. I can do 1.125 in most benches, but Heaven crash under 1.150. So appart from the hbm2 thing, I'm pretty happy right now.

 

[Rootax]

Another Vega vs Fury clock for clock :

https://www.hardocp.com/article/2017/09/12/radeon_rx_vega_64_vs_r9_fury_x_clock_for/15 (conclusion page).

"When you look at all the above architecture changes and benefits over previous high-end AMD GPUs you have to ask yourself where are those and why aren’t they making a bigger difference? Either the features are broken, turned off, not working, or its advantages were simply over-marketed."

 

[Entropy]

Yes, as it limits peak fan speed. Instead of limiting power to 100W it's much better to limit fan profile and let card manage power. That gives you desired noise levels at best performance, especially if combined with undervolting. Modern power management is really sophisticated and there are many knobs you can turn to make your card run as you want it to.

Worked out nicely limiting fan speed in WattMan, in conjunction with undervolting. It's not totally transparent what the power target slider actually affects and how much, but the fan speed slider directly affects maximum noise, so that is probably the preferable place to adjust. So now power draw is decent, noise is under control, and it still benchmarks almost 10% over stock Vega56, whereas I was prepared to drop back 10% from stock if necessary. I expected worse than I got.

 

[Picao84]

Another Vega vs Fury clock for clock :

https://www.hardocp.com/article/2017/09/12/radeon_rx_vega_64_vs_r9_fury_x_clock_for/15 (conclusion page).

"When you look at all the above architecture changes and benefits over previous high-end AMD GPUs you have to ask yourself where are those and why aren’t they making a bigger difference? Either the features are broken, turned off, not working, or its advantages were simply over-marketed."

To be fair, more than half of the performance difference between GTX980 and GTX1080 is also explained by clock speed. Pascal clocks 40-50% higher than Maxwell and delivers around 70% more performance (GM204 vs GP104). The remaining 20% can also be explained by GP104 having 25% more cores than GM204. Did not see anyone making questions about its architecture. Additionally, not every single new feature its targeted at performance. I have a hard time getting the sense of these analysis and questions.

 

[xEx]

To be fair, more than half of the performance difference between GTX980 and GTX1080 is also explained by clock speed. Pascal clocks 40-50% higher than Maxwell and delivers around 70% more performance (GM204 vs GP104). The remaining 20% can also be explained by GP104 having 25% more cores than GM204. Did not see anyone making questions about its architecture. Additionally, not every single new feature its targeted at performance. I have a hard time getting the sense of these analysis and questions.

Whats the diference in area between those? and tbh Nvidia could simply argure "it was enough to claim the performance crown".

in AMD case they marketed its "all new arquitecture" its all of its new feautre aimed to improve performance and then clock for clock it was equal than last(2 years old) gen. Not the same thing.

 

[Picao84]

Whats the diference in area between those? and tbh Nvidia could simply argure "it was enough to claim the performance crown".

in AMD case they marketed its "all new arquitecture" its all of its new feautre aimed to improve performance and then clock for clock it was equal than last(2 years old) gen. Not the same thing.

AMD also said that a lot of transistors were used to make sure that Vega would clock as high as it did. That is also part of "all new architecture" since Fury was not capable ot clocking nearly as high, while Maxwell was already a nice overclocker. You cannot just make a GPU and hope it will clock high. It has to be engineered for it and AMD succeded in that.

Regarding the diference in area between GP104 and Vega, dont forget Vega has features for compute that GP104 does not.

 

[Rootax]

In the comments they said they will do the same thing between 980 and 1080.

Saying the Vega improvments were not for "performances" is not right... You can even make the argument that Vega is a regression compared to Polaris. Right now a 14nm Fiji with only the tweak to go faster would do as good as Vega. You can't tell me it's normal and something isn't wrong.