EVGA GeForce GTX 1070 SuperClocked 8 GB

[the54thvoid]

I want to make very clear, the GTX 1070 appears to be a very nice video card. It smokes the GTX 970 and GTX 980. I just believe that reviews should help people with their purchasing choices and a large number of GTX 980Ti owners are selling their cards at huge losses to purchase a video card that is about the same as they already had and in some cases not as good.

You're being pedantic. It's pragmatically ill founded to bench a custom card against custom cards from each generation. The fact is if you look back, TPU benches a 980ti custom card against a reference 980ti card in that same generation (much as this review benches against a reference 1070). This allows you to gauge how far above reference that custom is. So each generation @W1zzard provides inter family comparisons. Next architectural leap he will then bench that family against it's custom variants. The previous generation reference is used as a baseline to see how well the custom new gen works against the previous baseline.

I know my Kingpin is about 15-20% faster than a stock 980ti (based on TPU reviews of cards running at high 1400's). So I know to use my knowledge to base my purchases against that. TPU isn't here to hold your hand through exacting benchmarks on every custom card to make a 99% probability purchase. It's here to give you insight into how well a card does above baseline and the competition. It's easy enough to back pedal through reviews to find your own card and see how it fares against the reference model.

doesn't the lack of async modules has an impact on that though?
on topic: solid card, but that price is a bit on the expensive side imo.

It's not about Async modules. It's about compute. You don't need AMD's ACE's to do Asynchronous scheduling, you need to be able to do compute. Nvidia dropped compute after Fermi (would be interesting to see a driver update for Vulkan on a Fermi card, GTX480). That dropping of compute is what gives Nvidia it's massive power efficiency edge. As I read elsewhere, it's a double edged sword - Nvidia is efficient because it is already fully utilised and efficient without it's compute. AMD's GCN is woefully under used in DX9-11 but shines in DX12 as it can use it's ACE's to throw a ton of compute at the graphics tasks.

Although going on a mini tangent here - Nvidia isn't Async crippled, it's just already got a fully utilised architecture and using Async doesn't help it at all. AMD on the other hand gets to run at it's potential under DX12. So the playing field is quite even now - which is good, if only AMD released a Fury X update. As long as DX11 is still the major API (and it is) Nvidia will stick to low compute, high efficiency against AMD's high compute, low efficiency.

 

[$ReaPeR$]

It's not about Async modules. It's about compute. You don't need AMD's ACE's to do Asynchronous scheduling, you need to be able to do compute. Nvidia dropped compute after Fermi (would be interesting to see a driver update for Vulkan on a Fermi card, GTX480). That dropping of compute is what gives Nvidia it's massive power efficiency edge. As I read elsewhere, it's a double edged sword - Nvidia is efficient because it is already fully utilised and efficient without it's compute. AMD's GCN is woefully under used in DX9-11 but shines in DX12 as it can use it's ACE's to throw a ton of compute at the graphics tasks.

Although going on a mini tangent here - Nvidia isn't Async crippled, it's just already got a fully utilised architecture and using Async doesn't help it at all. AMD on the other hand gets to run at it's potential under DX12. So the playing field is quite even now - which is good, if only AMD released a Fury X update. As long as DX11 is still the major API (and it is) Nvidia will stick to low compute, high efficiency against AMD's high compute, low efficiency.

that was what i was thinking, it will be interesting to see how nvidia will fare when it inevitably makes compute modules part of its arch again, there will be another 4xx generation "hot" issue imo. and yes i agree that dx11 is king right now, but i predict that dx12 will take hold in a shorter time period than dx11 did since M$ is pushing very hard with their win10 platform. imo in a year or so the race will get very hot, literally and figuratively. interesting times indeed.

 

[kilenfl]

What about RX480 ?

 

[Caring1]

What about RX480 ?

What about it!
Look at the thread title.

 

[bug]

It's not about Async modules. It's about compute. You don't need AMD's ACE's to do Asynchronous scheduling, you need to be able to do compute. Nvidia dropped compute after Fermi (would be interesting to see a driver update for Vulkan on a Fermi card, GTX480). That dropping of compute is what gives Nvidia it's massive power efficiency edge. As I read elsewhere, it's a double edged sword - Nvidia is efficient because it is already fully utilised and efficient without it's compute. AMD's GCN is woefully under used in DX9-11 but shines in DX12 as it can use it's ACE's to throw a ton of compute at the graphics tasks.

I would like to add that, besides removing compute capability which didn't make much sense on a consumer card, Nvidia also did it so they can more easily sell Titan and Tesla cards. Because those still have full compute capability.
Just for more context, I'm not trying to say anything else here.

 

[the54thvoid]

I would like to add that, besides removing compute capability which didn't make much sense on a consumer card, Nvidia also did it so they can more easily sell Titan and Tesla cards. Because those still have full compute capability.
Just for more context, I'm not trying to say anything else here.

Don't know if the Titan cards have better compute? The original might have but the follow ups (black and X) I think were gimped.

 

[Dethroy]

@W1zzard Could you please confirm that the EVGA GTX 1070 SC's power target is really set to 150W?
According to EVGA it is set to 170W. And according to an EVGA tech support employee it is set to 180W/215W at normal/OC settings (source: http://de.evga.com/forums/FindPost/2440897).

 

[Joss]

Good Lord, have they dumped the whole tube of thermal paste on that thing?

Indeed
It doesn't build confidence does it?
Excess thermal paste may bring no harm, but why wouldn't that carelessness translate into other areas of the manufacturing process?

 

[mcraygsx]

Very well written article as always, the part that really caught my attention was

"Currently, the EVGA GTX 1070 SC is listed online for $439, which is similar to other custom-design GTX 1070s, but still way too high; remember, we were told $379. Today, one month after launch, not a single card is priced like that. Rather, it seems as though pricing for custom designs is gravitating toward the Founders Edition price of $449; some cards above, some below. However, if you compare the EVGA GTX 1070 SC to the Founders Edition, it's definitely the better deal. You will save 10 bucks and gain amazing noise and temperature levels and no thermal throttling. On the other hand, compared to the MSRP of $379, a price increase of $60 is too much. Given the huge demand and limited availability right now, companies get away with this and cash in on people who want their new graphics card now."

Now I cannot wait for the day when I can buy 1070 for merely $379.

 

[Nelly]

A score of 9.8 looks abit silly now, considering the findings of three other review sites for the EVGA ACX 3.0 pascal cards not having sufficient cooling in place to cool the VRMs compared to all other brands.

I guess it could be time consuming and costly, I read on the Guru3D website that the thermal imagining equipment cost 10,000 Euros.

 

[bug]

A score of 9.8 looks abit silly now, considering the findings of three other review sites for the EVGA ACX 3.0 pascal cards not having sufficient cooling in place to cool the VRMs compared to all other brands.

I guess it could be time consuming and costly, I read on the Guru3D website that the thermal imagining equipment cost 10,000 Euros.

What? The SC is perfectly fine, only the FTW line was affected (solutions have been deployed already).

 

[R00kie]

What? The SC is perfectly fine, only the FTW line was affected (solutions have been deployed already).

It's the other way around.
EDIT: nevermind, they're all affected.

 

[bug]

It's the other way around.
EDIT: nevermind, they're all affected.

Wth are you talking about? This was only about the FTW line, see here: https://www.techpowerup.com/227849/...bios-updates-gtx-1080-70-60-ftw-line-of-cards

 

[Ungari]

Well even Maxwell offers better peformance per watt still.

Nvidia are far ahead of AMD in this regard for sure.

Yes, they are pennies per month ahead.

 

[bug]

Yes, they are pennies per month ahead.

Will you stop with this nonsense?
Performance per watt is really important, it's not about a few pennies saved. It means less heat for the same level of performance and less heat means simpler cooling and more silence. For the current generation, it also means Nvidia can scale up to 2GHz while AMD is left out of the high end because it can't.

 

[Nelly]

Wth are you talking about? This was only about the FTW line, see here: https://www.techpowerup.com/227849/...bios-updates-gtx-1080-70-60-ftw-line-of-cards

Any non-reference EVGA ACX 3.0 card with a shipping date prior to September don't have any thermal pads installed.

On the instructions on how to install the thermal pads it includes Classified, FTW and SC / stock versions, the methods differ due to the screws on the PCB being different for the backplates.

See here for details >> http://www.evga.com/thermalmod/thermal_pad_mod_installation_guide.pdf

 

[Ungari]

Will you stop with this nonsense?
Performance per watt is really important, it's not about a few pennies saved. It means less heat for the same level of performance and less heat means simpler cooling and more silence. For the current generation, it also means Nvidia can scale up to 2GHz while AMD is left out of the high end because it can't.

Clockspeed is important for DX11, but not so much for DX12 and Vulkan.
Apparently, due to yields and the early manufacturing process there aren't enough low TDP Polaris chips that can reach high OCs on Air, to sell to the consumer market. We look to see this rectified in the refresh of this arch, but AMD won't be in the "high end" until Vega anyway.

 

[bug]

Any non-reference EVGA ACX 3.0 card with a shipping date prior to September don't have any thermal pads installed.

On the instructions on how to install the thermal pads it includes Classified, FTW and SC / stock versions, the methods differ due to the screws on the PCB being different for the backplates.

See here for details >> http://www.evga.com/thermalmod/thermal_pad_mod_installation_guide.pdf

Ok, there's the misunderstanding then. Not all cards are affected as my 1060 SC definitely doesn't come with ACX 3.0, but neither is the problem apparent only on FTW line.

Clockspeed is important for DX11, but not so much for DX12 and Vulkan.
Apparently, due to yields and the early manufacturing process there aren't enough low TDP Polaris chips that can reach high OCs on Air, to sell to the consumer market. We look to see this rectified in the refresh of this arch, but AMD won't be in the "high end" until Vega anyway.

Believe what you want. Maybe you'll get Vega running at 300MHz since apparently that is not an issue going forward.
But you still more potent cooling (and possibly PSU) to tame more inefficient designs

 

[Ungari]

Ok, there's the misunderstanding then. Not all cards are affected as my 1060 SC definitely doesn't come with ACX 3.0, but neither is the problem apparent only on FTW line.



Believe what you want. Maybe you'll get Vega running at 300MHz since apparently that is not an issue going forward.
But you still more potent cooling (and possibly PSU) to tame more inefficient designs

At 300Mhz, the power consumption and thermals should be outstanding!

 

[R00kie]

Wth are you talking about? This was only about the FTW line, see here: https://www.techpowerup.com/227849/...bios-updates-gtx-1080-70-60-ftw-line-of-cards

He wasn't talking about the BIOS updates (although that also affected all other models mind you), he was talking about the thermal issues the cards having, that were fixed by issuing the thermal pads to customers.