NVIDIA Pascal, GP100

[Arkin]

Αναλυτικο testing απο φιλο στο reddit για τις pascal. Aπο οτι φαινεται ο τελευταιος driver εδωσε boost στα περισσοτερα παιχνιδια και μεγαλυτερα mins.

418.81 WHQL Early Driver Performance test

Hello, nVidia fellows.

I've finally completed my Early Performance Benchmark for the new 418.81 WHQL driver release.

Benchmark PC is a Windows 10 v. 1809 October Update, (latest patches applied) custom built desktop, 16Gb DDR3-1600 Ram, Intel i7-4790k with one Asus Strix GTX 1070Ti Advanced Binned, on a single BenQ 1080p 60hz. monitor with no HDR nor G-Sync. Stock clocks on both CPU and GPU.

Unless explicitly stated otherwise, games run borderless windowed, using the built in Benchmarking tool, with available 'cinematic' options disabled when possible, (Motion Blur, Chromatic Aberration, Film Grain, Vignette effects, Depth of Field, and such, not due to performance but for my own preference and image quality reasons).

The usual disclaimer: This is NOT an exhaustive benchmark, just some quick numbers and my own subjective impressions; and I can only judge for my own custom PC configuration. Any other hardware setup, different nVidia architecture, OS version, different settings... may (and will) give you different results.

 

Numbers below:

---

First one. Tom Clancy's: The Division using Snowdrop Engine with Dx12. 1080p resolution, almost maxed settings (just lowered a bit Extra Streaming Distance and Object Detail), Neutral Lightning, Dx12 enabled.

The Division - three runs with 417.71:

• Avg. FPS: 85.9 / 86.1 / 86.0

• Typical FPS: 86.3 / 86.7 / 86.5

• Avg. CPU: 67% / 63% / 62%

• Avg. GPU: 96% / 96% / 96%

The Division - three runs with 418.81:

• Avg. FPS: 86.6 / 86.3 / 86.5

• Typical FPS: 87.1 / 86.8 / 86.9

• Avg. CPU: 67% / 62% / 63%

• Avg. GPU: 96% / 96% / 96%

Once again, seems DX12 performance is stable or even a tiny bit better than the previous driver. It's not by much, and may be within the error margin of the test, but the trend seems consistent.

Actual gameplay feels in line with the previous driver. Smooth and stable.

Sadly frame times are not available, as FRAPS refused to work whenever I started The Divsion under Dx12 mode.

---

Next one. A Dx11 game on the AnvilNext engine: Ghost Recon: Wildlands on 1080p, mostly V.High but no Gameworks options enabled.

GR: Wildlands - three runs with 417.71:

• Avg FPS: 78.03 / 77.54 / 77.95

• Min FPS: 66.87 / 70.58 / 70.58

• Max FPS: 89.91 / 87.74 / 87.74

• Avg CPU: 52.1% / 51.9% / 51.4%

• Avg GPU: 95.8% / 95.1% / 95.7%

• Frame times (3-run average): Avg. 12.8 - Lower 1% 16.8 - Lower 0.1% 20.1

GR: Wildlands - three runs with 418.81:

• Avg FPS: 79.46 / 78.75 / 78.55

• Min FPS: 67.06 / 69.31 / 67.66

• Max FPS: 90.10 / 89.29 / 89.46

• Avg CPU: 50.2% / 51.0% / 49.0%

• Avg GPU: 96.2% / 96.4% / 96.5%

• Frame times (3-run average): Avg. 12.7 - Lower 1% 15.4 - Lower 0.1% 17.4

This time, the DX11 Wildlands benchmark is better and more stable than the previous driver. Raw FPS numbers are a tiny bit better, nothing stellar, but what is more important, the Lower 1% and Lower 0.1% Frame Times are noticeably better, indicating an improvement on game stuttering and smoother gameplay.

Actual gameplay outside the benchmark run seems also smoother during fast action sequences.

---

Next is FarCry 5, a Dunia Engine game (a heavily modified fork of the original CryEngine). Settings are 1080p, maxed Ultra settings with TAA and FoV 90.

FarCry 5 - three runs with 417.71:

• Min FPS: 63 / 63 / 59

• Avg FPS: 78 / 80 / 79

• Max FPS: 104 / 103 / 102

• Frame times (3-run average): Avg. 12.9 - Lower 1% 20.6 - Lower 0.1% 24.5

FarCry 5 - three runs with 418.81:

• Min FPS: 68 / 71 / 67

• Avg FPS: 84 / 87 / 85

• Max FPS: 108 / 108 / 108

• Frame times (3-run average): Avg. 12.0 - Lower 1% 17.5 - Lower 0.1% 22.1

FarCry 5 have got a huge performance improvement in this driver. Whatever happened with 417.35 seems for the most part, fixed (lost about 10% performance back on that driver, for what I heard related to a DirectX update done by Microsoft on a Win10 update).

Whatever the cause, the performance may not be as good as it was on 417.22, but it is much much better now than all other drivers in between.

Not only FPS averages are better, but also all Frame Time values are noticeably better, indicating a smoother gameplay with less stuttering. Yay!!!

---

Now an Unreal Engine game: Batman: Arkham Knight on 1080p, maxed settings and all Gamework options enabled (thus, heavily using nVidia PhisX engine).

Batman: AK - three runs with 417.71:

• Min FPS: 41 / 43 / 43

• Max FPS: 122 / 123 / 120

• Avg FPS: 82 / 84 / 82

• Frame times (3-run average): Avg. 12.4 - Lower 1% 21.4 - Lower 0.1% 27.5

Batman: AK - three runs with 418.81:

• Min FPS: 41 / 42 / 41

• Max FPS: 121 / 120 / 123

• Avg FPS: 85 / 85 / 85

• Frame times (3-run average): Avg. 12.3 - Lower 1% 21.3 - Lower 0.1% 26.3

While the Max FPS numbers are a bit down, the Average values and Minimum are up by a couple of extra FPS.

Not only that, but following the same trend the average Frame Time variance is somewhat better all around, and the Lower 0.1% improved by quite a chunk, indicating less lag spikes during the game.

---

Next one is Middle Earth: Shadow of Mordor, a LithTech Engine game. Settings are 1080p, maxed Ultra with FXAA antialiasing.

Shadow of Mordor - three runs with 417.71:

• Avg FPS: 135.17 / 134.74 / 134.65

• Max FPS: 232.05 / 234.95 / 231.00

• Min FPS: 52.35 / 54.06 / 52.48

• Frame times (3-run average): Avg. 7.90 - Lower 1% 16.3 - Lower 0.1% 18.9

Shadow of Mordor - three runs with 418.81:

• Avg FPS: 131.87 / 131.24 / 131.23

• Max FPS: 189.68 / 191.90 / 195.21

• Min FPS: 77.30 / 77.02 / 79.87

• Frame times (3-run average): Avg. 7.82 - Lower 1% 11.2 - Lower 0.1% 13.6

The huge changes which happened on Shadow of Mordor on 417.71 reverted back. This time the average FPS and maximum FPS are back to 417.22 numbers (lower than the previous driver), but the minimum FPS numbers are much better. At the same time, the Frame Times are substantially better, specially on the lower 1% and 0.1% brackets.

While the raw FPS average may be a bit lower, the game smoothness is now miles ahead of 417.71, with much less stuttering, and tbh, not a single lag spike can be perceived during gameplay at all.

---

Finally, Monster Hunter: World. It uses the Capcom propietary MT Framework engine, very common on other Capcom games (Lost Planet games, several Resident Evil games, Devil May Cry games, ...).

I'm using the Kushala Daora engine-rendered intro cutscene, which features two of the most troublesome scenarios performance-wise: the complex geometry of Elder's Recess and the heavy particle-driven whirlwind effects.

Settings are 1080p, maxed settings except for Volume Rendering set at Off and Subsurface Scattering Off, and FXAA antialiasing.

MH: World - four runs with 417.71:

• Avg FPS: 120 / 120 / 119 / 119

• Lower 1% FPS: 89 / 93 / 94 / 92

• Frame times (4-run average): Avg. 8.39 - Lower 1% 10.8 - Lower 0.1% 12.5

MH: World - three runs with 418.81:

• Avg FPS: 116 / 116 / 115

• Lower 1% FPS: 92 / 94 / 93

• Frame times (4-run average): Avg. 8.68 - Lower 1% 10.7 - Lower 0.1% 12.4

Monster Hunter World shows a small decrease in the average FPS numbers (and the average Frame Time), yet the lower 1% FPS and Frame Times are a bit better. Thus, game is performing a bit slower overall, but more smooth and with better frame pacing, improving the gaming perception during both the benchmarked run and during actual gameplay (I hunted a Xeno'jiva and the game went butter smooth during the whole combat, not a single hiccup).

 

---

 

Driver testing results

System stability is so far good with the new driver. The Division, Wildlands, FarCry4, FarCry5, XCOM2, EVE: Online, Terraria, World of Tanks Blitz, Batman Arkham Knight, BattleTech, the Mass Effect trilogy, Monster Hunter: World, Middle Earth: Shadow of Mordor, Star Trek Online and WoW (short testing game sessions) all ran fine without stability issues.

 

Performance-wise this driver is good for Pascal cards. Really good. In fact, given the recent trend, even surprisingly good. Not only all games got improvements in key areas (smoothness and even Raw FPS averages on most tests), but the massive drop in performance of FarCry 5 on 417.35 seems to have been corrected.

Performance seems a bit below 417.22 still on DX11 games, yet I'd say that it's better overall than any other 4xx branch driver release (except for the mentioned 417.22).

 

This time, I have no issue recommending this driver for Pascal owners which jumped into the 4xx driver branch. If you are still on 399.24, that's probably still a better performer for Dx11 games, and if you don't have issues you should keep it.

But if you need to update your drivers past 399 due to any bug, or if you are interested on DX12 performance improvements, Freesync compatibility, or any of the recent Game Ready profiles, then this new driver is probably the better choice (even more so for Laptop users who couldn't run 417.22 due to the low power state bug).

A note here though. I've read from a couple of different sources of very poor performance of this driver on Total War Warhammer 2 (and maybe other Total War titles). I cannot confirm it, and I haven't observed any drop in my tested games. But it may be a Total War engine specific issue, so have this in mind if you decide to update.

Επεξ/σία 5 Φεβρουαρίου 2019 από Arkin

[akoinonitos]

Δεν κολλαει στο νημα αλλα με τους τελευταιους drivers πηρα αρκετα στο firestrike wtf?

Επεξ/σία 7 Φεβρουαρίου 2019 από akoinonitos

[Arkin]

Οι νεοι drivers εδωσαν επιδοσεις που βλεπαμε στους 3xx drivers οπως ο 391. Πηραν απο turing μεχρι maxwell για 7xx kepler δεν ξερω. Και πιο stable fps και μεγαλυτερα min. Ποσο ρολο παιζει ενας driver ο ενας μπορει να ριμαξει το performance και ο αλλος να το απογειωσει.

[akoinonitos]

Μόλις τώρα, Arkin είπε

Οι νεοι drivers εδωσαν επιδοσεις που βλεπαμε στους 3xx drivers οπως ο 391. Πηραν απο turing μεχρι maxwell για 7xx kepler δεν ξερω. Και πιο stable fps και μεγαλυτερα min. Ποσο ρολο παιζει ενας driver ο ενας μπορει να ριμαξει το performance και ο αλλος να το απογειωσει.

Θα τεσταρω το βραδυ και κανα game να δω αν υπαρχει διαφορα και στην πραξη, αλλα 2k διαφορα στο fs ειναι too much

Επεξ/σία 7 Φεβρουαρίου 2019 από akoinonitos

[Arkin]

Μόλις τώρα, akoinonitos είπε

Θα τεσταρω το βραδυ και κανα game να δω αν υπαρχει διαφορα και στην πραξη, αλλα 2k διαφορα στο fs ειναι too much

Αρα τελικα η nvidia ριχνει τις επιδοσεις στις παλιοτερες καρτες για να προωθησει τις καινουργιες? Γιατι κατι τετοια διαβαζω και τρελαινομαι. Ενω βλεπω το ακριβως αντιθετο.

[BabisTs]

1 ώρα πριν, akoinonitos είπε

Δεν κολλαει στο νημα αλλα με τους τελευταιους drivers πηρα αρκετα στο firestrike wtf?

Στο Guru3d ποσταρει το total score bro.... και όχι το graphic score..... 

[akoinonitos]

25 λεπτά πριν, BabisTs είπε

Στο Guru3d ποσταρει το total score bro.... και όχι το graphic score..... 

Ρε φιλε εχεις δικιο ,11360 ειναι το grapchics score

[ultraex2003]

Θα συγκρίνετε το graphics σκορ πριν και μετά εάν έχει πάρει τίποτα από driver

[toyo73]

Σε kepler GTX 780  εχω μειωση 2 fps και 250 score στο benchmark του ACO.Οτιδηποτε πανω απο 399.24 εχω μειωση.Στο συγκεκριμενο παιχνιδι, μιας και μονο με αυτο ασχολουμε.

[Arkin]

Νεος driver. Anthem/dirt rally 2

 

https://www.nvidia.co.uk/Download/driverResults.aspx/143333/en-uk

[Arkin]

Ηρθε το τουμπανιασμενο Display Port της Ugreen σημερα οποτε εχουμε δοκιμες G Sync αργοτερα με Pascal κάρτα. Θα δοκιμασω και CRU να δω αν γινεται να πεσει το range και πιο κατω απο τα 40

 

 

Επεξ/σία 25 Φεβρουαρίου 2019 από Arkin

[noFEARgr]

2 ώρες πριν, Arkin είπε

Ηρθε το τουμπανιασμενο Display Port της Ugreen σημερα οποτε εχουμε δοκιμες G Sync αργοτερα με Pascal κάρτα. Θα δοκιμασω και CRU να δω αν γινεται να πεσει το range και πιο κατω απο τα 40

πιστευω οτι το 90% του κοσμου πλεον εχει αυτο το mousepad 😂

[Arkin]

4 λεπτά πριν, noFEARgr είπε

πιστευω οτι το 90% του κοσμου πλεον εχει αυτο το mousepad 😂

Ισχύει φίλε μου. Βγαίνει σε 3 διαστάσεις πήρα το μεγαλύτερο 90x40 το έχω ευχαριστηθεί. 

Επεξ/σία 25 Φεβρουαρίου 2019 από Arkin

[Arkin]

Πολυ καλα νεα για το G sync καρτα freesync οθονη.

 

Ενεργοποιησα το freesync στην οθονη εβαλα απο το control panel το gsync και εβαλα και το indicator για να ξερω αν ειναι ενεργο η οχι. Εβαλα 50% limit στην καρτα και εριξα τα ρολογια ωστε να πεσει απο τα 75 και να δω τι γινεται. Ιδου το αποτελεσμα. 

 

Το vsync το εχω ενεργο κανονικα παραλληλα με το gsync. Στη δευτερη εικονα που γραφει 65fps παει νερακι χωρις να καταλαβαινω οτι επεσα απο τα 75. Πριν με το vsync αν επεφτα στα 73-74 υπηρχε stuttering στα 37fps τωρα εξαφανιστηκε. Στο riva tuner εχω βαλει cap στα 75.

 

Οποτε δουλευει αψογα και ας μην ειναι gsync compatible το monitor. Πολυ ευχαριστημενος και απο το καλωδιο της Ugreen εξαιρετικη η ποιοτητα του. CRU αυριο. 

 

 

[noFEARgr]

1 ώρα πριν, Arkin είπε

Πολυ καλα νεα για το G sync καρτα freesync οθονη.

Ενεργοποιησα το freesync στην οθονη εβαλα απο το control panel το gsync και εβαλα και το indicator για να ξερω αν ειναι ενεργο η οχι. Εβαλα 50% limit στην καρτα και εριξα τα ρολογια ωστε να πεσει απο τα 75 και να δω τι γινεται. Ιδου το αποτελεσμα. 

Το vsync το εχω ενεργο κανονικα παραλληλα με το gsync. Στη δευτερη εικονα που γραφει 65fps παει νερακι χωρις να καταλαβαινω οτι επεσα απο τα 75. Πριν με το vsync αν επεφτα στα 73-74 υπηρχε stuttering στα 37fps τωρα εξαφανιστηκε. Στο riva tuner εχω βαλει cap στα 75.

Οποτε δουλευει αψογα και ας μην ειναι gsync compatible το monitor. Πολυ ευχαριστημενος και απο το καλωδιο της Ugreen εξαιρετικη η ποιοτητα του. CRU αυριο. 

το indicator πως το βαζεις ;