Digital Foundry Article Technical Discussion Archive [2014]

[Deleted member 13524]

You're right, I thought the 7900GTX was clocked lower than 500MHz (that would be G70, not G71 after the die shrink), and wikipedia says the G70 in the RSX does 5 ALU operations per pixel pipe per clock and 2 ALU operations per vertex pipe per clock, so I thought the card could only do below 68 GFLOPs. And anandtech claims 74,8 GFLOPs.

Regardless, GFXBench in T-Rex Offscreen shows the Shield to have over twice the performance of a Geforce Go 7900 GS, which is a 375MHz G71 with 4 pixel pipes disabled, so actual performance between the Shield and a 7900 GTX should actually be quite similar (besides bandwidth-limited scenarios, of course).

 

[pjbliverpool]

My memory is a bit fuzzy, but for G71, ignoring the mini-alu/scalar unit, I thought the raw flops was

24 PS* vec4 * 2 ALU * 2 flops/ALU *650Mhz = ~250 GFLOPs
8 VS * vec4 *1ALU * 2flops/ALU * 650MHz = ~42 GFLOPs)

I thought the vertex shaders were vec5 giving 52 GFLOPs, not that it makes a lot of difference!

 

[AlNom]

I thought the vertex shaders were vec5 giving 52 GFLOPs, not that it makes a lot of difference!

There's a mini-ALU (scalar) that goes alongside the vec4, which I ignored in the calc.

Regardless, GFXBench in T-Rex Offscreen shows the Shield to have over twice the performance of a Geforce Go 7900 GS, which is a 375MHz G71 with 4 pixel pipes disabled, so actual performance between the Shield and a 7900 GTX should actually be quite similar (besides bandwidth-limited scenarios, of course).

Maybe it's faster due to FP20 in the pixel shaders in T4 Odd anyway.

 

[Arnold Beckenbauer]

Yes, I'm an asshole:
Because it's a port from the X360, there is no mixed mode for T4, which can improve the performance on NV30 devices. (yes, T4 is more NV40 generation)

http://www.anandtech.com/show/1144/6

They should make a comparison between the shield version and HL2 running on one if those 8" windows 8.1 tablets.

HL2 runs good on my W4-820 (1280x800, Shadows on Medium (no high available), Water reflections on Reflect World, Textures on high (no very high), Shaders an high, HDR on, Motion Blur on, 1xAA and 2xAF. But if there are somewthere transparent textures (dust, fire, smoke...), the fps becomes red from green (under 20 fps).

 

[Deleted member 13524]

What's "1XAA"?

 

[Nesh]

Non existent Anti Aliasing

 

[Arnold Beckenbauer]

What's "1XAA"?

NoAA for geeks and nerds.

 

[Deleted member 13524]

Maybe it's faster due to FP20 in the pixel shaders in T4 Odd anyway.

Here's a comparison between 2005/2006 discrete GPUs and smartphones/tablets that anadntech took about a year ago, using 3dmark and GFXBench:
http://www.anandtech.com/show/6877/the-great-equalizer-part-3

If you picture a Tegra 4 in those tests, it should beat the 7900GTX in almost every test.

 

[AlNom]

Here's a comparison between 2005/2006 discrete GPUs and smartphones/tablets that anadntech took about a year ago, using 3dmark and GFXBench:
http://www.anandtech.com/show/6877/the-great-equalizer-part-3

If you picture a Tegra 4 in those tests, it should beat the 7900GTX in almost every test.

hm... from an earlier article from Anand:http://www.anandtech.com/show/7508/nvidia-tegra-note-7-review/4


Filltest, offscreen
7900GTX - 4647
Tegra Note 7 - 1164

Tri-throughput, fragment lit, offscreen
7900GTX - 227
Tegra Note 7 - 38.5

Egypt HD, offscreen
7900GTX - 67
Tegra Note 7 - 55

T-Rex, offscreen
7900GTX - 24
Tegra Note 7 - 21


Curious.

3D Mark Graphics Test 1
7900GTX - 12
Tegra Note 7 - 59.9
Tegra Note 7 (Extreme) - 56

3D Mark Graphics Test 2
7900GTX - 263
Tegra Note 7 - 52.7
Tegra Note 7 (Extreme) - 36


----

This test [Graphics Test 1] is very heavy on the vertex shaders, but the 7900 GTX and friends should do a lot better than they are. These workloads however were designed for a very different set of architectures.

Hm...

 

[Deleted member 13524]

hm... from an earlier article from Anand:http://www.anandtech.com/show/7508/nvidia-tegra-note-7-review/4http://www.anandtech.com/show/6787/...re-deep-dive-plus-tegra-4i-phoenix-hands-on/6

Old results and that's Tegra Note, not the Shield.
On Egypt 2.5 Offscreen the Shield scores 65 FPS, and in T-Rex 2.7 Offscreen it scores the same 24 FPS as the 7900GTX. And the offscreen tests should be quite favorable to the discrete card because they're made in 1080p where the large gap in memory bandwidth could make quite a difference.


Looking at all the metrics we've seen so far, I can't see how a Shield couldn't run Half Life 2 maxed out at 720p (where a 7900GTX runs the same game at 1080p comfortably).
From the videos, the game seems to be running with much lower visuals than for example Killzone Mercenary, and the Shield should run circles around a PS Vita.

 

[AlNom]

And the offscreen tests should be quite favorable to the discrete card because they're made in 1080p where the large gap in memory bandwidth could make quite a difference.

Wasn't that the point of comparing the HW? Just trying to figure out what's going on here with the speed differential.

I suppose Tegra4 isn't wholly G70 verbatim. Not sure what the architectural improvements are. The article had mentioned something funny about the vertex shader performance they were seeing on G70 tests.

Are the Egypt & T-Rex benchmarks heavy in VS?

HL2 doesn't strike me as heavy in VS, so the advantage in the benchmarks may not apply.

---

So if we're just looking at simple VS & just raw pixel throughput, Tegra 4 is still a different performance bracket (differences between Note7 & Shield aside). But like I said earlier, there's also some funny business with # of objects, larger view distances/"complex" scenery (comparing views outside of the station)/hallways with only a few characters that make it seem there's other weird stuff going on in the port.


---
[strike]

Old results and that's Tegra Note, not the Shield. Vita.

Actually, the Tegra Note 7 article was November 2013.

(And yes, TN7 is 600MHz GPU vs 672MHz)
[/strike]

nevermind!

----

 

[Deleted member 13524]

Wasn't that the point of comparing the HW? Just trying to figure out what's going on here with the speed differential.

Yes, but I was just mentioning the performance comparison considering that the tests are being made in 1080p whereas the Shield could only ever do 720p max.

Actually, the Tegra Note 7 article was November 2013.

¬¬ yes, I was quoting the first article from Februrary 2013 you referenced to, before you edited your post..
That or I may be getting crazy. Am I?

 

[AlNom]

Yes, but I was just mentioning the performance comparison considering that the tests are being made in 1080p whereas the Shield could only ever do 720p max.

Ah...

hmm... One other thing that came to mind was texturing (since we only see bilinear used in the port) as Tegra 4 only has 4 texture units.

edit: bleh, I'm just going to go with funky port job.

¬¬ yes, I was quoting the first article from Februrary 2013 you referenced to, before you edited your post..
That or I may be getting crazy. Am I?

It was supposed to be a quick fix.

 

[Rangers]

DF Wolfenstein preliminary comparison is up. Current gen versions/PC seem mostly identical.

However the X1 had one instance of running at 1440X1080 out of a cutscene. Second time, as the same thing (running at reduced res after a cutscene) happened on Tomb Raider X1.

 

[Wynix]

Tech Analysis: Half-Life 2 and Portal on Android



It's more a port from the Xbox360's version (Orange Box):
http://www.youtube.com/watch?v=tYLvzCGJ2jM
The Orange Box
We pit EA's port against PC and 360.

...

They jumped the gun with these ports, they should have waited for the Shield 2.

 

[chris1515]

http://www.eurogamer.net/articles/digitalfoundry-2014-hands-on-with-project-morpheus?page=comments

DF and Project Morpheus

 

[Ike Turner]

DF Wolfenstein preliminary comparison is up. Current gen versions/PC seem mostly identical.

However the X1 had one instance of running at 1440X1080 out of a cutscene. Second time, as the same thing (running at reduced res after a cutscene) happened on Tomb Raider X1.

DF updated the story noting that the PS4 does the same thing (but less often than the X1)..

 

[Scott_Arm]

http://www.eurogamer.net/articles/digitalfoundry-2014-wolfenstein-the-new-order-face-off

Never played Rage enough to know how well the dynamic framebuffer worked. How do they do it without adding latency? Both X1 and PS4 version are locked 60fps using dynamic framebuffer, but how do they keep the controls snappy? I expect more dynamic framebuffer implementations this gen.

 

[HTupolev]

How do they do it without adding latency?

They probably drop to a lower resolution if the timing for maintaining a particular refresh rate starts nearing its limits at the current resolution.

Why would dynamic resolution add latency?

 

[jlippo]

http://www.eurogamer.net/articles/digitalfoundry-2014-wolfenstein-the-new-order-face-off

Never played Rage enough to know how well the dynamic framebuffer worked. How do they do it without adding latency? Both X1 and PS4 version are locked 60fps using dynamic framebuffer, but how do they keep the controls snappy? I expect more dynamic framebuffer implementations this gen.

Basic idea is to keep framebuffer full resolution and do the scaling during post phase.
This article from intel explains it nicely.
https://software.intel.com/en-us/articles/dynamic-resolution-rendering-article