TR Forums

auxy wrote:

^{full version here}

You can do better than this A hint: make sure your Titan is running at 16x PCIe 3.0 mode (remove other cards if the BIOS drops you to lower modes due to shared PCIe lanes), it does make difference with this benchmark. A second hint: raise the Titan's temp target to max value - it "likes" to throttle down as soon as temps start to approach 80C. You should be getting around 130 FPS max with just these adjustments (I couldn't get this but I have a lame i7-2600 with old PCIe standard). Last - if you're feeling courageous, there's a couple of BIOSes floating around the OCN that should get you up to 1202MHz on air, just look around at "Nvidia GeForce GTX TITAN Owners' club" (don't forget to post your results there too!)

Randomness wrote:

You can do better than this :wink: A hint: make sure your Titan is running at 16x PCIe 3.0 mode (remove other cards if the BIOS drops you to lower modes due to shared PCIe lanes), it does make difference with this benchmark. A second hint: raise the Titan's temp target to max value - it "likes" to throttle down as soon as temps start to approach 80C. You should be getting around 130 FPS max with just these adjustments (I couldn't get this but I have a lame i7-2600 with old PCIe standard). Last - if you're feeling courageous, there's a couple of BIOSes floating around the OCN that should get you up to 1202MHz on air, just look around at "Nvidia GeForce GTX TITAN Owners' club" (don't forget to post your results there too!) :wink:

Yah, I'm only running PCIe x8, but I did raise the temperature target as high as it would go. It was at ~1200Mhz the whole time during the benchmark, hehe. I made a video of the benchmark on Extreme HD; you can see it remains at 1200Mhz; obviously, the score is a little lower because I was recording. (^∇^)

Randomness wrote:

You can do better than this A hint: make sure your Titan is running at 16x PCIe 3.0 mode

Really? I haven't noticed any increase in performance ever with PCIe 2.0 vs 3.0 with my HD 7970. I just run on PCIe 2.0 because it seems to offer a bit more stability. Yes, I know the Titan requires a bit more bandwidth, but it shouldn't enough to make much of a difference. Maybe PCIe 3.0 is broken on Sandy bridge-E.

Has anyone else ever noticed a difference between setting PCIe to 2.0 vs 3.0 in the bios?

cynan wrote:

Maybe PCIe 3.0 is broken on Sandy bridge-E.

I realize Sandy Bridge-E is not Sandy Bridge, but Sandy Bridge doesn't support PCIe 3.0; it would surprise me a little if SB-E did.

cynan wrote:

Has anyone else ever noticed a difference between setting PCIe to 2.0 vs 3.0 in the bios?

I'll test it when I get home, with PCIe 3.0 x8 vs. PCIe 3.0 x16, but I doubt it will matter.

auxy wrote:

cynan wrote:

Maybe PCIe 3.0 is broken on Sandy bridge-E.

I realize Sandy Bridge-E is not Sandy Bridge, but Sandy Bridge doesn't support PCIe 3.0; it would surprise me a little if SB-E did.

Well, there's the option to switch between PCIe 2.0 and 3.0 in my Sandy-E board's bios. I don't think PCIe 3.0 was "officially" supported in the end. I think it was supposed to be (just like Sandy-e was suppose to have more USB 3.0 and SATA-III ports natively). At least there are quite a few enthusiast boards with the bios option. GPU-Z detects that PCIe 3.0 is working when I enable it in the bios (and not when I disable it) - don't know if that is reliable or not.

cynan wrote:

GPU-Z detects that PCIe 3.0 is working when I enable it in the bios (and not when I disable it) - don't know if that is reliable or not.

Huh. It accurately reports the expected values on my machine; I specifically tested that at one time. Does PCIe not connect directly to the CPU on SB-E?

auxy wrote:

cynan wrote:

GPU-Z detects that PCIe 3.0 is working when I enable it in the bios (and not when I disable it) - don't know if that is reliable or not.

Huh. It accurately reports the expected values on my machine; I specifically tested that at one time. Does PCIe not connect directly to the CPU on SB-E?

As far as I know, PCIe 3.0 is the same for SBE as for IB. So yes. Except of course that SBE has 40 PCIe 2.0/3.0 lanes while IB only has 16 lanes. It's just that for some reason Intel didn't get around to fully validating PCIe 3.0 for SBE. I did read something about Nvidia disabling PCIe 3.0 on SBE, but I think this was fixed at some point. AMD HD 7900 cards have always worked at PCIe 3.0 on SBE (at least according to GPU-Z).

cynan wrote:

I did read something about Nvidia disabling PCIe 3.0 on SBE, but I think this was fixed at some point. AMD HD 7900 cards have always worked at PCIe 3.0 on SBE (at least according to GPU-Z).

Well, Nvidia didn't even support PCIe 3.0 until Kepler series, hehe. I'm sure somebody just assumed they were running at 2.0 on their Fermi cards because Nvidia are jerks.
^{I mean, they totally are jerks, but that's not why. Hee.}
Anyway, that's neat to learn about SB-E and PCIe 3.0. My motherboard is actually a Z68 (Sandy Bridge) chipset, but it was built specifically with support for IVB/22nm and PCIe 3.0. °˖✧◝(⁰▿⁰)◜✧˖°

Following up the previous post:

Here's an Anandtech article from way back to the HD 7970 release (Dec 2011!) testing an HD 7970 on SBE. It says that PCIe 3.0 vs 2.0 did nothing for gaming, but increased performance in computing benchmarks by 9%.

You can run this nifty tool with AMD cards to check the card's pcie speed. It shows you the actual data transfer speeds. Neat tool: http://www.softpedia.com/get/System/Ben ... Test.shtml

ClickClick5 wrote:

You can run this nifty tool with AMD cards to check the card's pcie speed.

It doesn't work on Geforces? 「(°ヘ°)

JustAnEngineer wrote:

[img]http://i230.photobucket.com/albums/ee120/JustAnEngineer/Games/UnigineValley.png]

. Was it running the integrated graphics? I ran with exact same settings and got only a few points different on my surface pro.

auxy wrote:

ClickClick5 wrote:

You can run this nifty tool with AMD cards to check the card's pcie speed.

It doesn't work on Geforces? 「(°ヘ°)

I don't think so..........I remember trying it on a 260 and it did not, but try it.

thecoldanddarkone wrote:

Was it running the integrated graphics? I ran with exact same settings and got only a few points different on my surface pro.

GT620 probably isn't any faster than HD4000, eheh... (‘～`；)

ClickClick5 wrote:

I don't think so..........I remember trying it on a 260 and it did not, but try it.

I will! As soon as I get home! I'm curious to know too! (⊙△⊙✿)

cynan wrote:

Randomness wrote:

You can do better than this A hint: make sure your Titan is running at 16x PCIe 3.0 mode

Really? I haven't noticed any increase in performance ever with PCIe 2.0 vs 3.0 with my HD 7970. I just run on PCIe 2.0 because it seems to offer a bit more stability. Yes, I know the Titan requires a bit more bandwidth, but it shouldn't enough to make much of a difference. Maybe PCIe 3.0 is broken on Sandy bridge-E.

Has anyone else ever noticed a difference between setting PCIe to 2.0 vs 3.0 in the bios?

I cannot test it, but...
See the score here: http://www.overclock.net/t/1363440/nvid ... t_19486448
And then here: http://www.overclock.net/t/1363440/nvid ... t_19486935
by the same person, who basically changed the PCIE type (and did a re-install of EVGA's Precision (but I doubt that Precision's clock settings can really produce such difference).

Well, I tested it -- removing the second graphics card did indeed raise me (per GPU-Z) to PCIe 3.0 x16 and it raised my Valley score to... ─=≡Σ((( つ•̀ω•́)つ

...well, no, actually, it's almost exactly the same. It was 2560 before, now it's 2619. I guess that's a small improvement, but 3% is nothing to worry about. ┐(￣ー￣)┌

2619 is still a lot lower than those guys are scoring in that thread; the one guy who saw a large improvement had a large OC in the second post, and also is running LGA2011, although I'm not sure if that matters much.

Dunno! ┐(‘～`；)┌

One last effort for the lowly Radeon. I pushed it pretty hard. It's now in the fetal position sobbing quietly in the corner...

cynan wrote:

One last effort for the lowly Radeon. I pushed it pretty hard. It's now in the fetal position sobbing quietly in the corner...

"Lowly"? (･｀ｪ´･)つ
My GPU costs a thousand dollars! You hush.

Radeon HD6970:

JustAnEngineer wrote:

Radeon HD6970: (image)

I find myself continually struck by how horrible the Radeon 5xxx/6xxx series is. Why did people hate on Fermi so much again? (◕︿◕✿)

Those frame rates with the 2-year-old Radeon HD6970 are at twice the resolution that your display has.

JustAnEngineer wrote:

Those frame rates with the 2-year-old Radeon HD6970 are at twice the resolution that your display has.

FYI, for comparison purposes, the default setting for benchmarking with Valley (here and in the other mentioned thread) has been to use the Extreme HD preset. This sets resolution to 1080p, all settings to maximum (including 8x AA - the real performance sapper).

cynan wrote:

FYI, for comparison purposes, the default setting for benchmarking with Valley (here and in the other mentioned thread) has been to use the Extreme HD preset.

Ah, this is obviously some strange usage of the word "default" that I wasn't previously aware of.

I let it run at the default settings that it defaulted to when it installed. I didn't realize that there was some other "default" setting. My six-year-old monitor doesn't support 1920x1080.

JustAnEngineer wrote:

Those frame rates with the 2-year-old Radeon HD6970 are at twice the resolution that your display has.

Yes, I can see that. Am I supposed to be impressed? ┐(￣ー￣)┌

My 3-year-old GTX460 scores ~950 in 1920x1080 with 8x MSAA, which is inarguably more demanding both in terms of fillrate and bandwidth than 2560x1600 no AA. That's a three-year-old mid-range card with much more terrible specs getting 2/3 the performance of a newer, top-end card, on a more challenging test.

You can also go back and look at my friend Emily's results on her 6870, which I think is a 5770? They're also pretty bad.

I guess Fermi ran hot and was power-thirsty, but in terms of raw performance, it was pretty awesome.

JustAnEngineer wrote:

Ah, this is obviously some strange usage of the word "default" that I wasn't previously aware of.

cynan wrote:

FYI, for comparison purposes, the default setting for benchmarking with Valley (here and in the other mentioned thread) has been to use the Extreme HD preset.

Wiktionary wrote:

(often attributive) a tentative value or standard that is presumed.

JustAnEngineer wrote:

My six-year-old monitor doesn't support 1920x1080.

Really? Isn't it 2560x1600? I think you've just said something nonsensical.

Well, maybe "default" isn't the right word. Anyway, using the Extreme HD preset lets people easily know what settings you are using. If you don't pick a preset, all you get is "custom" which isn't very useful for comparison purposes. The selection of Extreme HD was inspired by this thread at overclock.net. I'm not sure that your monitor needs to support 1080p for the benchmark to run at 1080p. For instance, I've played games at 1920x1200 on a Dell 3007wfp that also doesn't support 1080p (is that what you have?)...

Edit: Woops, just noticed it says 3007wfp in your sig

auxy wrote:

JustAnEngineer wrote:

My six-year-old monitor doesn't support 1920x1080.

Really? Isn't it 2560x1600? I think you've just said something nonsensical.

No, it's true. The Dell 3007wfp only supports 2560x1600 and 720p. That's it. But the video card should scale 1080p or whatever to run at 2560x1600. You will only be rendering at 1080p density, which is the important thing for the benchmark.

Bordom is kicking in....I'll make a TR bench spreadsheet for us.

cynan wrote:

No, it's true. The Dell 3007wfp only supports 2560x1600 and 720p. That's it. But the video card should scale 1080p or whatever to run at 2560x1600. You will only be rendering at 1080p density, which is the important thing for the benchmark.

Yah, that's what I was talking about. Both Radeons and Geforces can scale the outputted image on the GPU; as a result, any monitor supports any resolution that fits inside its native res entirely. I don't even know or care what resolutions my monitors support besides native.

auxy wrote:

JustAnEngineer wrote:

Radeon HD6970: (image)

I find myself continually struck by how horrible the Radeon 5xxx/6xxx series is.

Eh, they were somewhat acceptable from a hardware point of view and price/performance point of view back at their day... Of course, the "performance" part didn't apply to some games because of half-assed (which was typical for this company) driver optimizations and the long, long time it usually took for those optimizations to materialize - I remember having multiple issues with BF:BC2 and my HD6950 card long time ago, where some driver release was causing constant crashes of this game after about an hour (or slightly less) of playing it, and also something about slow level loading times due to specific shader compiling in DX10 mode (switching to DX9 made maps load faster)... Anyways, was my last ATI/AMD card and I'm glad I removed it from my PC