GTX TITAN X 4-Way SLI + 5K Benchmarks

Baasha

Baasha

Limp Gawd
Joined
Feb 23, 2014
Messages
249
GTX Titan X 4-Way SLI & 5K Benchmarks:


https://youtu.be/XOTCTdzTPwE

Finally got around to doing some benchmarks - the skinny on it is that 4-Way SLI scaling sucks balls - even at 5K.

VRAM usage is insane - well over 6GB in most games at 5K.
 
Even on benchmarks like GPU-bound Unigine Heaven? I saw you ran it, but not sure of the interpretation of results.
 
Seems like he was running out of CPU horsepower: he kept mentioning how the cards were so little used.
 
So where is the data? I just see some jerk playing a game.

I bet the CPU was chocked with 4 of those cards.
 
The Benchmark screens go by pretty quickly - but if you pause it, you can see the Afterburner OSD on the top left for GPU info.

The actual results are pretty good for Heaven 4.0 - @ 1440P & 4K.

3DMark Fire Strike Ultra is also good - >15K.

The scaling in games suck ass - you can see the graph toward the end of the video.

Here are the benchmarks of the games:

hJq93Inh.jpg
 
Why 4xMSAA on BF4? in 5k on a 27"... ???

with 217.57 PPI ????

my eyes might not be all that stellar.. but i can use my 28" 4k monitor without scaling.. but i cant see the difference of anything above 2xmsaa and this is on a 157.35 PPI monitor..

So overkill much??

you know... nVidia might just get DSR to work in sli.. then you can really use it at the max
 
Something tells me we're just not quite where we want to be when it comes to 4K/5K yet and gaming. LOL
 
Wow Crysis 3 loves 4-way actually. I suppose because the frame rates are low to start with. Though Sniper Elite 3 scaled very nicely to 350 fps ... as if 105 from a single card wasn't enough, lol.
 
This explains what is wrong Baasha.
http://www.littletinyfrogs.com/article/460524/DirectX_11_vs_DirectX_12_oversimplified

"It’s all about the cores

Last Fall, Nvidia released the Geforce GTX 970. It has 5.2 BILLION transistors on it. It already supports DirectX 12. Right now. It has thousands of cores in it. And with DirectX 11, I can talk to exactly 1 of them at a time.

Meanwhile, your PC might have 4, 8 or more CPU cores on it. And exactly 1 of them at a time can talk to the GPU.

Let’s take a pause here. I want you to think about that for a moment. Think about how limiting that is. Think about how limiting that has been for game developers. How long has your computer been multi-core?

But DirectX 12? In theory, all your cores can talk to the GPU simultaneously. Mantle already does this and the results are spectacular. In fact, most benchmarks that have been talked about have been understated because they seem unbelievable. I’m been part of (non-NDA) meetings where we’ve discussed having to low-ball performance gains to being “only” 40%. The reality is, as in, the real-world, non-benchmark results I’ve seen from Mantle (and presumable DirectX 12 when it’s ready) are far beyond this. The reasons are obvious."
 
Iklos said:
Why 4xMSAA on BF4? in 5k on a 27"... ???

with 217.57 PPI ????

my eyes might not be all that stellar.. but i can use my 28" 4k monitor without scaling.. but i cant see the difference of anything above 2xmsaa and this is on a 157.35 PPI monitor..

So overkill much??

you know... nVidia might just get DSR to work in sli.. then you can really use it at the max
Click to expand...

Dsr does work with sli
 
cageymaru said:
This explains what is wrong Baasha.
http://www.littletinyfrogs.com/article/460524/DirectX_11_vs_DirectX_12_oversimplified

"It’s all about the cores

Last Fall, Nvidia released the Geforce GTX 970. It has 5.2 BILLION transistors on it. It already supports DirectX 12. Right now. It has thousands of cores in it. And with DirectX 11, I can talk to exactly 1 of them at a time.

Meanwhile, your PC might have 4, 8 or more CPU cores on it. And exactly 1 of them at a time can talk to the GPU.

Let’s take a pause here. I want you to think about that for a moment. Think about how limiting that is. Think about how limiting that has been for game developers. How long has your computer been multi-core?

But DirectX 12? In theory, all your cores can talk to the GPU simultaneously. Mantle already does this and the results are spectacular. In fact, most benchmarks that have been talked about have been understated because they seem unbelievable. I’m been part of (non-NDA) meetings where we’ve discussed having to low-ball performance gains to being “only” 40%. The reality is, as in, the real-world, non-benchmark results I’ve seen from Mantle (and presumable DirectX 12 when it’s ready) are far beyond this. The reasons are obvious."
Click to expand...

I was reading the article and had to stop at "It's not magic." Ha, bullshit. Of course it is.

Good article for us simpletons. But of course DX12 is just like airplanes. We don't really know how they work, it's just magic.
 
cageymaru said:
This explains what is wrong Baasha.
http://www.littletinyfrogs.com/article/460524/DirectX_11_vs_DirectX_12_oversimplified

"It’s all about the cores

Last Fall, Nvidia released the Geforce GTX 970. It has 5.2 BILLION transistors on it. It already supports DirectX 12. Right now. It has thousands of cores in it. And with DirectX 11, I can talk to exactly 1 of them at a time.

Meanwhile, your PC might have 4, 8 or more CPU cores on it. And exactly 1 of them at a time can talk to the GPU.

Let’s take a pause here. I want you to think about that for a moment. Think about how limiting that is. Think about how limiting that has been for game developers. How long has your computer been multi-core?

But DirectX 12? In theory, all your cores can talk to the GPU simultaneously. Mantle already does this and the results are spectacular. In fact, most benchmarks that have been talked about have been understated because they seem unbelievable. I’m been part of (non-NDA) meetings where we’ve discussed having to low-ball performance gains to being “only” 40%. The reality is, as in, the real-world, non-benchmark results I’ve seen from Mantle (and presumable DirectX 12 when it’s ready) are far beyond this. The reasons are obvious."
Click to expand...

The dx12 "preview" tests are nothing like real games though. At 4k with mantle on/off the performance difference on a 295x2 is like 4% on average so in practical gaming terms DX12 will allow more freedom in designing games but even if current games were running DX12 the performance gains won't actually be that noticeable.
 
cageymaru said:
This explains what is wrong Baasha.
http://www.littletinyfrogs.com/article/460524/DirectX_11_vs_DirectX_12_oversimplified

"It’s all about the cores

Last Fall, Nvidia released the Geforce GTX 970. It has 5.2 BILLION transistors on it. It already supports DirectX 12. Right now. It has thousands of cores in it. And with DirectX 11, I can talk to exactly 1 of them at a time.

Meanwhile, your PC might have 4, 8 or more CPU cores on it. And exactly 1 of them at a time can talk to the GPU.

Let’s take a pause here. I want you to think about that for a moment. Think about how limiting that is. Think about how limiting that has been for game developers. How long has your computer been multi-core?

But DirectX 12? In theory, all your cores can talk to the GPU simultaneously. Mantle already does this and the results are spectacular. In fact, most benchmarks that have been talked about have been understated because they seem unbelievable. I’m been part of (non-NDA) meetings where we’ve discussed having to low-ball performance gains to being “only” 40%. The reality is, as in, the real-world, non-benchmark results I’ve seen from Mantle (and presumable DirectX 12 when it’s ready) are far beyond this. The reasons are obvious."
Click to expand...

Yea, DX12 is going to be pretty awesome.

I did most of the benchmarks w/ AA maxed just to show how much VRAM etc. is being used.

Shadow of Mordor @ 5K w/ everything maxed uses ~ 9.6GB which is insane:

SoM_5K_2_Thumb.jpg


Hopefully Win 10 w/ DX12 will be fully released by this Fall.

Tgrove said:
Dsr does work with sli
Click to expand...

It does not work on 4K monitors w/ SLI yet. It most definitely does not work in 5K.

In fact, in 5K, even on a single GPU, DSR is disabled.

I also inquired about 5K Surround since I had 2x 5K monitors (returned one).

Was going to pick up three for 5K Surround but Nvidia confirmed that 5K Surround will NOT work even w/ 4x GTX TITAN X in 4-Way SLI (that's 12 DP 1.2 ports) and they don't have any plans to implement that in the near future.

Yea - so NO DSR @ 5K even on a single GPU.

In game, however, like Shadow of Mordor, allows 200% scaling which is just bonkers. Haven't tried it because I am almost 100% certain that even 12GB of VRAM is not enough for that. :p

I tried BF4 w/ no AA and no POST AA and am getting close to 60FPS on almost all maps w/ everything else maxed out (Ultra) at 5K which is sweet!

Just picked up Hardline today. Let's see how that one does.. hehe..
 
This is interesting. People thought NVIDIA was crazy going for 12GB of VRAM on the Titan X. Evidently not.
 
I still run my 2x titan with a 7680x1440 setup, makes me feel like I might need to upgrade. You can never have too much vram.
 
Thanks for the benches Baasha. I'm curious about a couple of things though after seeing your results:

  1. Have you double-checked that your GPU's are running in PCI-E 3.0 and not 2.0? Not sure if this is even still an issue for X99 chipsets but I know it was for me on x79 (and still is on z97). Every time I upgrade drivers I have to run the 'Force PCI-E 3.0' utility to keep my system from reverting back to 2.0.
  2. Is there any chance you can squeeze a few more mhz out of that 5960X? I suspect every little bit you can get will help with regard to your GPU usage.
Of course the biggest drawback is gonna be due to immature drivers (as you've mentioned) and there's not really much you can do about that but wait.

Anyway, looks like one heck of a setup!
 
Last edited:
Can somebody quantify what is the difference between the 4-Way Titan X and 3-Way? Typically the fourth card brings just little more than decoration of PC internals but is this also true for the Titan X?
 
I find it a bit strange that Tomb Raider only used less than 3GB maxed out including TressFX because I could have sworn that when I benchmarked it, it almost used 4GB on my setup.
 
Dan_D said:
This is interesting. People thought NVIDIA was crazy going for 12GB of VRAM on the Titan X. Evidently not.
Click to expand...

Are you looking at the same benches? You are running out of GPU grunt long before you are running out of memory.

To the OP can you do some benchmarks on Shadow of Mordor with playable settings, like where you are getting min framerates of at least 25/30. How much Vram does it use then?
 
wirk said:
Can somebody quantify what is the difference between the 4-Way Titan X and 3-Way? Typically the fourth card brings just little more than decoration of PC internals but is this also true for the Titan X?
Click to expand...

Check the benchmarks, a few titles actually get something from the 4th card, like Crysis 3 and Sniper Elite. Others show no gain or negative scaling like Heaven.
 
reaper12 said:
Are you looking at the same benches? You are running out of GPU grunt long before you are running out of memory.

To the OP can you do some benchmarks on Shadow of Mordor with playable settings, like where you are getting min framerates of at least 25/30. How much Vram does it use then?
Click to expand...

I just glanced over the memory usage. Nothing more. But I do see your point.
 
wirk said:
Can somebody quantify what is the difference between the 4-Way Titan X and 3-Way? Typically the fourth card brings just little more than decoration of PC internals but is this also true for the Titan X?
Click to expand...

4-Way helps at much higher resolutions (>4K).

The Titan X is such a strong card, however, that even 5K is handled beautifully by 3-Way SLI. 4K Surround (3x 4K monitors) is where 4-Way SLI will really shine IME.
 
Baasha said:
Just did a 3-Way SLI review at 5K and the results are... amazing.

Link: https://youtu.be/NQIc9MuP8ck
Click to expand...

Your sound levels are off - the volume level required for your voiceover means that the intro music is punishingly loud. And your Uber rig screen - the bright green bit is almost illegible - and this is a 3-card video not a 4-card one. Jebus but you go overkill on the transition sequences at the start. Okay, you've got 4 Titan X cards. Show them once and move on.

Once you get beyond that, you're giving lots of really useful info.
 
Quartz-1 said:
Your sound levels are off - the volume level required for your voiceover means that the intro music is punishingly loud. And your Uber rig screen - the bright green bit is almost illegible - and this is a 3-card video not a 4-card one. Jebus but you go overkill on the transition sequences at the start. Okay, you've got 4 Titan X cards. Show them once and move on.

Once you get beyond that, you're giving lots of really useful info.
Click to expand...

Thanks for the feedback. Yea, something is off w/ my mike levels. The system still has 4 cards and I just disabled one via the mobo (Asus RVE) so that's why I used the same pic - misleading as you say but didn't know what else to put there lol.

Will update the intro as well.
 
OP, I think you need to do some more work on your rig to make sure everything is in order and you have no bottlenecks.

I have been running 3 original titans at 7860x1600 which is almost as many pixels as you are pushing and I get much better performance than you're posting. Unless the 4th Titan X is giving no or negative scaling, something is off here.

To me the obvious culprits are:
-do you have the CPU needed to drive this many cards (6 core minimum)
-or even more of an issue, does your motherboard have the PCIE lanes needed?

I remember when I was expanding my rig, people used to swear all over the forum that PCIE lanes or PCIE 2.0 vs 3.0 doesn't make a big difference. They turned out to be dead wrong and going to a motherboard with big time PCIE bandwidth doubled my performance. If you are running over 3 top tier cards and over 10 million pixels, the motherboard can easily get bogged down if you don't have something top of the line.
 
anonmoniker said:
OP, I think you need to do some more work on your rig to make sure everything is in order and you have no bottlenecks.

I have been running 3 original titans at 7860x1600 which is almost as many pixels as you are pushing and I get much better performance than you're posting. Unless the 4th Titan X is giving no or negative scaling, something is off here.

To me the obvious culprits are:
-do you have the CPU needed to drive this many cards (6 core minimum)
-or even more of an issue, does your motherboard have the PCIE lanes needed?

I remember when I was expanding my rig, people used to swear all over the forum that PCIE lanes or PCIE 2.0 vs 3.0 doesn't make a big difference. They turned out to be dead wrong and going to a motherboard with big time PCIE bandwidth doubled my performance. If you are running over 3 top tier cards and over 10 million pixels, the motherboard can easily get bogged down if you don't have something top of the line.
Click to expand...

5960X, X99, 32MB DDR4 - I think his specs are fine.

Perhaps you can post proof that you're getting "much better" performance?

Let's face it - 3-way and 4-way SLI are a joke. No one is getting good scaling or performance.
 
Hi Baasha, Is there anyway you can redo all these benchmarks running at 4K resolution? Not trying to be rude but don't think anyone cares about 5K just yet considering there's only 1 27" monitor that does 5K while there's dozens of 4K computer monitors and hundreds of 4K TV's. Even 3x 4k monitors in surround would have been a more informative. I have 3 Titan X's but curious what the performance increase would be going with 4 way SLI. Unfortunately your 5K benchmarks don't help me and there's only a few other vague 4 way sli benchmarks out there. Any chance this could happen? I know the new drivers significantly improved performance as well like you said so maybe the benchmarks should be redone anyway. I also think this would significantly increase your YouTube subscribers as well if you're trying to go that route since there's probably a thousand+ 4k owners for every one 5K owner. Thanks!
 
Last edited:
Seyumi said:
Hi Baasha, Is there anyway you can redo all these benchmarks running at 4K resolution? Not trying to be rude but don't think anyone cares about 5K just yet considering there's only 1 27" monitor that does 5K while there's dozens of 4K computer monitors and hundreds of 4K TV's. Even 3x 4k monitors in surround would have been a more informative. I have 3 Titan X's but curious what the performance increase would be going with 4 way SLI. Unfortunately your 5K benchmarks don't help me and there's only a few other vague 4 way sli benchmarks out there. Any chance this could happen? I know the new drivers significantly improved performance as well like you said so maybe the benchmarks should be redone anyway. I also think this would significantly increase your YouTube subscribers as well if you're trying to go that route since there's probably a thousand+ 4k owners for every one 5K owner. Thanks!
Click to expand...

Hey Seyumi,

Thanks for the feedback.

Yes, I will do a comprehensive 4K benchmark video.

The 5K is so addicting that every time I want to do 1440P or 4K benches, I keep switching to 5K and playing! :D

I will do it though - I just did 1440P, 4K, and 5K benchmarks of GTA V w/ everything maxed out including AA - will release the video soon.
 
anonmoniker said:
OP, I think you need to do some more work on your rig to make sure everything is in order and you have no bottlenecks.

I have been running 3 original titans at 7860x1600 which is almost as many pixels as you are pushing and I get much better performance than you're posting. Unless the 4th Titan X is giving no or negative scaling, something is off here.

To me the obvious culprits are:
-do you have the CPU needed to drive this many cards (6 core minimum)
-or even more of an issue, does your motherboard have the PCIE lanes needed?

I remember when I was expanding my rig, people used to swear all over the forum that PCIE lanes or PCIE 2.0 vs 3.0 doesn't make a big difference. They turned out to be dead wrong and going to a motherboard with big time PCIE bandwidth doubled my performance. If you are running over 3 top tier cards and over 10 million pixels, the motherboard can easily get bogged down if you don't have something top of the line.
Click to expand...

The original 5K benchmarks w/ 4-Way SLI turned out to be quite bad because of the horrendous 347.88 drivers.

If you see the 3-Way SLI review at 5K (here: https://youtu.be/NQIc9MuP8ck), the performance is way better.

Now that the drivers are updated, the performance overall is much better but I've been playing GTA V almost exclusively and so the 4-Way SLI does pretty well but there's a lot of room to improve. Here are my 'first impressions' of GTAV on the PC: https://youtu.be/s1DQF9fX9uM

One thing to note, and I'm not nitpicking here, is that resolution scales exponentially, not linearly. So your resolution of 12.3MP compared to 5K (14.7MP) is not "almost" the same. It is significantly higher. The same can be said of 4K Surround and 5K - 4K Surround (my old setup) is 24.8MP - that's a HUGE step-up from 5K.

The performance needed to get 60FPS at 4K, 1600P Surround (your setup), and 5K are significantly different.

The 4-Way review I did first in this thread, as I said before, is not a great representation of actual performance due to poor drivers for the Titan X. The 3-Way review shows much better performance and scaling (as does 4-Way now).

4-Way SLI Titan X w/ 347.88:
hJq93Inh.jpg


3-Way SLI Titan X w/ 350.05:
lKO5szx.jpg


Oh, and a little something else :D

BJVB6Tl.jpg
 
You must log in or register to reply here.
Back
Top