1070 vs 980 Ti

[scottyseng]

So I got a used 1070 (MSI Gaming X) yesterday in the mail, and got it overclocked to 2.1GHz and 2,052 MHz on the VRAM clock. I have a 980 Ti (Gigabyte G1 Gaming) which I also bought used over the summer, also overclocked to 1.5GHz and 1,952 MHz on VRAM.

 

I bought a 1070 to have a GPU for my NAS so it could render (Since it stays on all of the time while the PC doesn't) but iray doesn't have support for the 1070 yet (At least iray built into 3DS Max / Maya)...I didn't realize that.

 

While I was testing the 1070 to make sure it worked, my question is, the 1070 should be faster right? I kind of wonder if I messed up buying a 1070.

 

In my tests, the 980 Ti was faster. On Unigene Heaven, the 980 Ti wins by 5 fps. In working in 3DS Max / Maya / Unity / Revit, the viewport is faster on the 980 Ti. I don't play games much at all and mostly work on modeling / game development / programming, and the only real games I own are the Neptunia games, where the 980 Ti runs the game (VII at 4KHD via res mod) at 60fps (game locked) at 70% load while the 1070 is completely pinned at 90%.

 

The only test the 1070 won at was 3DMark Time Spy and Firestrike, but not by much. The 1070 also uses like 20-30 watts less under load.

 

If I were to get another bang for the buck CUDA render card though, should I get another 980 Ti or 1070?

[tarfeef101]

1 minute ago, scottyseng said:

So I got a used 1070 (MSI Gaming X) yesterday in the mail, and got it overclocked to 2.1GHz and 2,052 MHz on the VRAM clock. I have a 980 Ti (Gigabyte G1 Gaming) which I also bought used over the summer, also overclocked to 1.5GHz and 1,952 MHz on VRAM.

 

I bought a 1070 to have a GPU for my NAS so it could render (Since it stays on all of the time while the PC doesn't) but iray doesn't have support for the 1070 yet (At least iray built into 3DS Max / Maya)...I didn't realize that.

 

While I was testing the 1070 to make sure it worked, my question is, the 1070 should be faster right? I kind of wonder if I messed up buying a 1070.

 

In my tests, the 980 Ti was faster. On Unigene Heaven, the 980 Ti wins by 5 fps. In working in 3DS Max / Maya / Unity / Revit, the viewport is faster on the 980 Ti. I don't play games much at all and mostly work on modeling / game development / programming, and the only real games I own are the Neptunia games, where the 980 Ti runs the game (VII at 4KHD via res mod) at 60fps (game locked) at 70% load while the 1070 is completely pinned at 90%.

 

The only test the 1070 won at was 3DMark Time Spy and Firestrike, but not by much. The 1070 also uses like 20-30 watts less under load.

 

If I were to get another bang for the buck CUDA render card though, should I get another 980 Ti or 1070?

If (assuming all other things stay the same) the 980ti is faster in YOUR SPECIFIC APPLICATIONS, then go with it!!

Although yes, in terms of raw performance, yeah, 1070 is faster. But like you have discovered, it is far less integrated and optimized for all uses as of yet.

[kayotics]

58 minutes ago, scottyseng said:

So I got a used 1070 (MSI Gaming X) yesterday in the mail, and got it overclocked to 2.1GHz and 2,052 MHz on the VRAM clock. I have a 980 Ti (Gigabyte G1 Gaming) which I also bought used over the summer, also overclocked to 1.5GHz and 1,952 MHz on VRAM.

 

I bought a 1070 to have a GPU for my NAS so it could render (Since it stays on all of the time while the PC doesn't) but iray doesn't have support for the 1070 yet (At least iray built into 3DS Max / Maya)...I didn't realize that.

 

While I was testing the 1070 to make sure it worked, my question is, the 1070 should be faster right? I kind of wonder if I messed up buying a 1070.

 

In my tests, the 980 Ti was faster. On Unigene Heaven, the 980 Ti wins by 5 fps. In working in 3DS Max / Maya / Unity / Revit, the viewport is faster on the 980 Ti. I don't play games much at all and mostly work on modeling / game development / programming, and the only real games I own are the Neptunia games, where the 980 Ti runs the game (VII at 4KHD via res mod) at 60fps (game locked) at 70% load while the 1070 is completely pinned at 90%.

 

The only test the 1070 won at was 3DMark Time Spy and Firestrike, but not by much. The 1070 also uses like 20-30 watts less under load.

 

If I were to get another bang for the buck CUDA render card though, should I get another 980 Ti or 1070?

Well I think the 980ti is better because of it's architecture it makes it a very good card to overclock with the right cooling but I am guessing if you want to buy one of these cards you have enough to spend on water cooling for it. But on the other hand the 1070 is I think cheaper and is newer so it will be more energy efficient and  maybe give more performance then the 980ti of overcolcked which is good for a already high clocked card like the 1070. Sorry if this was useless I am trying to help.

 

Kayotics.

[ItsTheDuckAgain]

Also keep in mind power cost:

 

1070FE has a TDP of ~150W

  980Ti has a TDP ofs ~250W

so a 100W difference.

 

100W difference per h is 100W per 1h or 0.1kWh.

 

For me calculating power price then goes like this:

WattDifference * HoursUsedPerDay / 1000 * 365 * PriceFor1KWhAtYourPlace

 

For me it looks like this then:

100 * 10 / 1000 * 365 *0,25 =  91,25 CHF per year

 

Assuming 10h usage per day at MAX load. Of course that is all hypothetical but should yield a general idea.

 

 

To take it one step further.

My system pulls an average of ~247W from the wall and I use it for about 6,75h per day, again on average.

Usage daytime hours: 247x3,75/1000x365x0,25=84 CHF

Usage nighttime hours: 247x3/1000x365x0,17=45 CHF

(different prices for power depending on the  hour of the day)

Which results in a cost per year of ~130CHF to run my PC

[Acridtj]

At this point the 980ti will edge out the 1070 in alot of instances, but since the 1070 is a new chipset and they are still working on fully utilizing the power of it and the other Pascal cards it is a good choice, as it's cheaper and is barely beaten by the 980ti and even beats the 980ti in some cases so imo it was a good choice.

[scottyseng]

4 minutes ago, ItsTheDuckAgain said:

-snip-

That's TDP, that's not how many watts the card actually uses. I think for me, my whole PC uses 300W with the 1070 vs 350W under load. Both kind of use roughly the same wattage at idle.

 

I'm guessing the general idea is that I should probably buy another 1070 (Well, after iray adds pascal support) for the next assist GPU for rendering.

[ItsTheDuckAgain]

7 minutes ago, scottyseng said:

That's TDP, that's not how many watts the card actually uses. I think for me, my whole PC uses 300W with the 1070 vs 350W under load. Both kind of use roughly the same wattage at idle.

 

I'm guessing the general idea is that I should probably buy another 1070 (Well, after iray adds pascal support) for the next assist GPU for rendering.

yeah of course that's TDP - that's why I wrote: "Assuming 10h usage per day at MAX load. Of course that is all hypothetical but should yield a general idea."

 

you got to do the math yourself to figure it out :-)

[GTBTK]

The performance between the two different models are similar in most current situations. They will swap places in benchmarks as you have already observed. 

 

GPU boost 3.0 is different in pascal cards and you have a learning curve to maximize the benefits you can get from overclocking. You will find that through experimentation, your 1070 performance will improve as you discover better tweaks over time. You do have more OC performance available when you master the curve that you don't have access to with the 980TI. Maximum frequency at one point on the curve, while impressive for bragging rights, is less important for performance than maximizing the stable values of all points on the curve. I get better performance at 2088 with higher midrange levels than I can get with a default curve with 2151Mhz at 1.093v 

 

Benefits that the 1070 does give you though is lower power usage, lower heat management issues, Asynchronous Compute compatibility that you don't get with Maxwell cards. This will become more important over time as more Games are developed to run with DX12 natively. 

[scottyseng]

2 minutes ago, GTBTK said:

Benefits that the 1070 does give you though is lower power usage, lower heat management issues, Asynchronous Compute compatibility that you don't get with Maxwell cards. This will become more important over time as more Games are developed to run with DX12 natively. 

Yeah, I don't think I'll ever own any games that use DX12...or continue hoping that Autodesk adds DirectX 12 support for viewports.

 

If anything, I'm trying to learn how to add DirectX12 support to games. haha.

 

I guess for the future, I'll try to snag more used 1070s and turn them into a render server.

 

I haven't gotten to adjusting the curve for the 1070 yet, since I'm still debating if I want the 1070 in my PC or server. Then I'll go fetch precision X (Currently using MSI afterburner) and follow the video Jayz2cents posted up a while back for overclocking pascal cards (I've never seen GPU Boost at all before owning either of these cards...since my previous card was a FirePro V7900...which is equal to a HD5850).

[GTBTK]

Who knows what will be around in 12-18 months? I would be interested to see how the 980Ti compares to the 1070 in GPGPU rendering performance. I have never had the chance to do a comparison.

 

Make sure that you check and see what sort of memory you have. If you have Micron memory and the Bios is 86.04.26.00.XX then you need to do the bios update to fix the micron bug in that bios version

 

GPU boost 2.0 in the 980Ti is a bit more predictable as it involves less variables than GPU 3.0 in the 1070

 

Afterburner 4.3 is better than precision XOC for fine tuning and managing the curve. If you just use the sliders they are both pretty much the same. Remember Jay is a peripheral part of the EVGA marketing machine so the Precision is part of the subliminal marketing message. Precision add benefits if you are using EVGA cards because it enables extra features like Kboost that you wont get with the MSI card

 

Try this on the 1070 and see how your performance compares to the 980Ti.

 

Voltage +100

Power Limit +126

Temp +92

Core +50

 Memory to +500

Open the curve window and drag the .975 point up to 2012Mhz and the 1.093 point up to 2088 so you end up with a curve with a double hump.

 

Then see how it performs. If you crash, pull the .975 point back down to 1999Mhz and try again. if it still crashes pull the 1.093V point down to 2076Mhz

 

[scottyseng]

1 minute ago, GTBTK said:

-snip-

Oh, you can manage the curve on afterburner? I thought only Precision X was the only software to do that. I've been using afterburner for years (The FirePro stock fan curve runs the GPU way too hot).

 

hmm, I have a MSI Gaming X 1070 and managed to get +150 core, but the memory would not go fast 400...I actually backed it  down to +100 because +400 caused my system to crash and artifact on desktop. I also already had the power limit / temp / voltage maxed.

 

Though, where is the curve window in MSI afterburner? I don't seem to see it / find it in settings. I'll probably play with this more when I wake up in the morning. I haven't had much time to really push the OC on the 1070 quite yet.

[xAcid9]

The 1070 will be slightly faster in DX12 title that leverage AsynCompute if you have enough CPU

 power, other than that your 980 Ti will slightly faster in most case while consuming more power around 80w-100w more.

3 minutes ago, scottyseng said:

Though, where is the curve window in MSI afterburner? I don't seem to see it / find it in settings. I'll probably play with this more when I wake up in the morning. I haven't had much time to really push the OC on the 1070 quite yet.

It's in Fan tab inside setting. You need to tick "Enable User Define blablablabla".

[GTBTK]

Sure. You need ver 4.3 of After burner. 

 

Sounds like you have micron Ram and need to install the bios update you can get here https://www.msi.com/Graphics-card/support/GEFORCE-GTX-1070-GAMING-X-8G.html#down-bios

 

You can access the curve by clicking the little bar graph icon or by pressing Ctrl-F

[GTBTK]

1 hour ago, ItsTheDuckAgain said:

Also keep in mind power cost:

 

1070FE has a TDP of ~150W

  980Ti has a TDP ofs ~250W

so a 100W difference.

 

100W difference per h is 100W per 1h or 0.1kWh.

 

For me calculating power price then goes like this:

WattDifference * HoursUsedPerDay / 1000 * 365 * PriceFor1KWhAtYourPlace

 

For me it looks like this then:

100 * 10 / 1000 * 365 *0,25 =  91,25 CHF per year

 

Assuming 10h usage per day at MAX load. Of course that is all hypothetical but should yield a general idea.

 

 

To take it one step further.

My system pulls an average of ~247W from the wall and I use it for about 6,75h per day, again on average.

Usage daytime hours: 247x3,75/1000x365x0,25=84 CHF

Usage nighttime hours: 247x3/1000x365x0,17=45 CHF

(different prices for power depending on the  hour of the day)

Which results in a cost per year of ~130CHF to run my PC

MSI 1070 Gaming X has maximum bios power limits the range from 230W to 291W. The reality is that in a 4K test you can get to 230W but getting close to 291W is only really possible using furmark

[3 Lions]

. 

[3 Lions]

Get another GTX 1070. 

[scottyseng]

7 hours ago, xAcid9 said:

-snip-

Yeah, I'm primarily using the GPUs for CUDA rendering, where they don't really care if there's DX12 support or not. Though maybe in the distant future one of the games I play will use it...or if Autodesk adds DX12 for viewports.

 

Ah, I was looking for the voltage curve editor.

7 hours ago, GTBTK said:

-snip-

Ah, CTRL-F works. I couldn't spot the little graph icon, but CTRL-F is fine for me. Thanks though, that means I don't have to go get EVGA Precision X.

 

Though I never thought GPUs would need BIOS updates like Motherboards do. Interesting. I'll also update the BIOS and try playing with the voltage curve.

 

Yeah, in my testing, the 980 Ti vs 1070, I could only get a max of 50 watts difference unless I pulled out Furmark / MSI Kombustor, where the 980 Ti sucked 60-70W more.

7 hours ago, LinusTechTipsFanFromDarlo said:

Get another GTX 1070. 

Sounds like the general consensus. I'll keep hunting for more used 1070s for more CUDA rendering performance.

[GTBTK]

Before the Bios Update, I could only get the memory up to +400 before it started checkerboard artifacting and crashing the PC. Post bios update my card will run fine to just under +600 before I start seeing overclocking artifacts. I suspect that you will be pleased. In my experience, memory OC works wonders for gpgpu processing.

 

Graphics cards are effectively small super computers by themselves. GPUs need instructions on how to work just as much as the main CPU does.

[Guest]

won't the 980ti be better at rendering? regardless of 10% performance increase from drivers over the upcoming months?

 

anyways, the efficiency of the 1070 is enough to justify it for me. Looking to sidegrade the 390 because of it's power hungry nature. 

[scottyseng]

6 hours ago, GTBTK said:

-snip-

I've updated the card, and thanks to your tip, I also saw that my 980 Ti also had a BIOS update. I'll be sure to play with overclocking tomorrow after my calculus exam...I was busy moving my server and PC to a new server rack I got.

 

I do hope the memory can be pushed to even 400+.

 

Very interesting. Yeah, you sure learn something new every day. For kicks though, I ran both my PC and my server under full load....1000W for rendering...

6 hours ago, Alexokan said:

-snip-

Yeah, the 980 Ti is better for rendering, but after more thinking I'll probably get 1070s or 1080s used for the efficiency. The gap between the 980 Ti grows under higher loads, it really starts sucking down power when rendering.

 

I'm pretty sure a 1070 should give you a boost in speed from a 390. Usually a 1070 is compared to a Fury. You will see a drop in power usage for sure. I remember how hot a 390 was because I did a build for my friend with it a while back.