Nvidia GeForce GTX 1000 Series (Pascal) MegaThread: FAQ and Resources

[renz496]

Based on the updated review of GTX 1070, Founder's Edition GTX 1070 uses surprising amount of power from PEG; the power consumption from PEG hovers around 75W WITHOUT overclocking. That's is above the limit of 65W.

Would the card consume more power from PEG when the card is overclocked and the power limit is lifted? Would there be any long-term stability issue of the motherboard when it is paired with a 1070FE? Why no one ever talks or tests this?

I have seen virtually no any other reviews showing separate power consumption from 6/8-pin connectors and PEG until after RX480 reported the issue two weeks ago.

Does any owners of 1070FE have any experience related to this issue over overclocked and non-overclocked 1070FE and the stability of the motherboard?

I have a workstation with Xeon E3-1276v3 on a ASUS P9D WS motherboard with 32GB of ECC memory. I want to use just this workstation instead of building another one just for gaming, and I have to make sure this PC for work is as stable and durable as possible. I can't afford it to go wrong in any ways. So I have not installed the card since I received it two weeks ago.

Power Consumption Results
http://www.tomshardware.com/reviews/nvidia-geforce-gtx-1070-8gb-pascal-performance,4585-7.html

the issue mostly on older mobo or one with cheap component. also from the article itself:

They're not particularly dangerous, but can cause audible artifacts if you're using on-board audio on a poorly-designed motherboard.

in case of RX480 reference they excessively drawing more power from the slot. this guy pairing reference RX480 with very old mobo and experiencing failure while the same mobo able to handle more power hungry 980TI just fine

https://www.youtube.com/watch?v=rhjC_8ai7QA

 

[manleysteele]

Lot's of possibilities ran through my head when I saw the leak. The first was the life of a processor meme. I don't know how long it takes NVidia to go from a design concept to first silicon, but Intel is on record as saying it takes 5 years, more or less.

If a processor is ready, nothing good happens from waiting. I think Volta was always a 14nm part. That's the first thing I though of. NVidia would have had to have identified the need for Pascal very early in the Volta process to release it first. I certainly don't think NVidia is above some light subterfuge with processor roadmaps and schedules.

My actual opinion is that the present roadmap was always the real thing concerning a successor to Fermi and Kepler.

Once a design is finalized and is ready for taping, every day you wait is a fool's bargain. Waiting on a process that is not ready, when the process that is ready is a sufficing solution makes no business sense. In real business, design dollars and man-hours are no different than production dollars and man-hours. It all has to be paid for by the product. A product that is not for sale makes no money. Business 101.

i think after the 40nm disaster nvidia start making back up plan. nvidia can design the processor how they want but it is useless if there is no fab can manufacture their design. and for them TSMC is the only fab that capable to manufacture their gpu. they did try looking other such as samsung but samsung doesn not have the expertise and experience making large gpu they need. and after 40nm nvidia also dedicated more of their engineer to work closely with TSMC. i think nvidia know early on that 20nm will be no go for gpu. i think that's how they able to re-tweak maxwell and still releasing maxwell in 2014.

I agree that TMSC has turned out to be a better partner for NVidia than Global Foundries has for AMD. With Maxwell, NVidia had the luxury of running it as the 800 series in the mobile space for a while before bringing it to the desktop as the 900 series. Obviously, they and TSMC learned a lot during that period. Whether it's something NVidia learned or something TSMC learned, it's all grist for the mill. AMD really needs to find a similar windfall before releasing Vega.

 

[BurgerandChips66]

'Pascal is merely Maxwell on speed'

Here is an interesting video comparing the performance of Maxwell vs. Pascal when running at the same rate of Flops, which is not the same clock speed.

https://youtu.be/nDaekpMBYUA

When Volta lands with significant architectural improvements (so not just Async Compute or the addition of HBM) there is likely to be another huge bump in performance heading our way.

 

[George Phillips]

Based on the updated review of GTX 1070, Founder's Edition GTX 1070 uses surprising amount of power from PEG; the power consumption from PEG hovers around 75W WITHOUT overclocking. That's is above the limit of 65W.

..............

in case of RX480 reference they excessively drawing more power from the slot. this guy pairing reference RX480 with very old mobo and experiencing failure while the same mobo able to handle more power hungry 980TI just fine

https://www.youtube.com/watch?v=rhjC_8ai7QA

The first drivers for RX480 distribute power draw from PCI-Express slot more than 75W and someitme even higher. In contrast, many customized 980Ti draw very little or even just 5 w of power from PCI-E slot. GeForce GTX 1070FE is drawing the top limit of what can be provided by PCI-E slot. So I wonder if anyone with 1070FE has similar issues especially after overclocking.

 

[TehPenguin]

'Pascal is merely Maxwell on speed'

Here is an interesting video comparing the performance of Maxwell vs. Pascal when running at the same rate of Flops, which is not the same clock speed.

https://youtu.be/nDaekpMBYUA

When Volta lands with significant architectural improvements (so not just Async Compute or the addition of HBM) there is likely to be another huge bump in performance heading our way.

Very interesting. So it's like have a CPU from the same family with the same IPS and just clock one higher. Very informative.

So is it just the power efficiency on pascal that gives it the edge and Volta will be the next step for better silicon? Or maybe a 3GHz GTX 1180 :lol:

 

[Bstraw53]

I'm starting to get pretty bummed out. I don't want to pay more than MSRP for one of these cards but I check every single day on 3-4 of the most popular sites and it seems like they sell out within minutes of getting them in.

I'd love to find either a ROG Strix 1080 or EVGA FTW.. Anyone have any advise on locking one down?

Thanks!

 

[James Mason]

I'm starting to get pretty bummed out. I don't want to pay more than MSRP for one of these cards but I check every single day on 3-4 of the most popular sites and it seems like they sell out within minutes of getting them in.

I'd love to find either a ROG Strix 1080 or EVGA FTW.. Anyone have any advise on locking one down?

Thanks!

Supply and demand.
http://www.investopedia.com/university/economics/economics3.asp

 

[Bstraw53]

I understand Supply and demand. Just thought maybe someone here would know something I don't. I'm not very "hardcore" when it comes to PC hardware so I thought maybe someone here knew of a better way that checking the sites every day. Seemed logical since most everyone in here actually has one of the newer cards.

 

[James Mason]

I understand Supply and demand. Just thought maybe someone here would know something I don't. I'm not very "hardcore" when it comes to PC hardware so I thought maybe someone here knew of a better way that checking the sites every day. Seemed logical since most everyone in here actually has one of the newer cards.

Stop checking and wait till demand is lower than supply.

The people here who have the new cards caved and payed over the MSRP.

 

[Ambular]

@Bstraw53: Or bookmark this page. http://www.nowinstock.net/computers/videocards/nvidia/gtx1080/

 

[manleysteele]

'Pascal is merely Maxwell on speed'

Here is an interesting video comparing the performance of Maxwell vs. Pascal when running at the same rate of Flops, which is not the same clock speed.

https://youtu.be/nDaekpMBYUA

When Volta lands with significant architectural improvements (so not just Async Compute or the addition of HBM) there is likely to be another huge bump in performance heading our way.

I see you a tortured methodology for making a result come out the way he wants and raise you a yes, but a 1080 will not only knock a 980Ti to the ground, it will kick it while it's down. It'll be much worse when the 1080Ti hits the street.

 

[renz496]

'Pascal is merely Maxwell on speed'

Here is an interesting video comparing the performance of Maxwell vs. Pascal when running at the same rate of Flops, which is not the same clock speed.

https://youtu.be/nDaekpMBYUA

When Volta lands with significant architectural improvements (so not just Async Compute or the addition of HBM) there is likely to be another huge bump in performance heading our way.

still they did not make comparison to GP100. pascal is more than just a simple refinement from maxwell. if you follow gpu news quite closely you should know that nvidia did make GK210 especially for tesla only. pascal added many compute related stuff that did not exist in maxwell. but some of this is simply useless in gaming card. maxwell is more like the ideal design for a pure gaming card. with their growth in HPC and professional application nvidia finally decided to make two specific design for two different market instead one design that will rule them all like they did in the past with GF100/GF110/GK110. that's why GP102 exist. and for that reason we will never going to see GP100 in a gaming card.

 

[manleysteele]

'Pascal is merely Maxwell on speed'

Here is an interesting video comparing the performance of Maxwell vs. Pascal when running at the same rate of Flops, which is not the same clock speed.

https://youtu.be/nDaekpMBYUA

When Volta lands with significant architectural improvements (so not just Async Compute or the addition of HBM) there is likely to be another huge bump in performance heading our way.

still they did not make comparison to GP100. pascal is more than just a simple refinement from maxwell. if you follow gpu news quite closely you should know that nvidia did make GK210 especially for tesla only. pascal added many compute related stuff that did not exist in maxwell. but some of this is simply useless in gaming card. maxwell is more like the ideal design for a pure gaming card. with their growth in HPC and professional application nvidia finally decided to make two specific design for two different market instead one design that will rule them all like they did in the past with GF100/GF110/GK110. that's why GP102 exist. and for that reason we will never going to see GP100 in a gaming card.

^This.

 

[renz496]

The first drivers for RX480 distribute power draw from PCI-Express slot more than 75W and someitme even higher. In contrast, many customized 980Ti draw very little or even just 5 w of power from PCI-E slot. GeForce GTX 1070FE is drawing the top limit of what can be provided by PCI-E slot. So I wonder if anyone with 1070FE has similar issues especially after overclocking.

i don't think it will be a problem. ever since 780Ti nvidia has implement something like intelligent power delivery for their card to optimize overclocking. i don't remember all the details but when overclocking instead of drawing more power from the slot and the additional power connector the card only draw more power from the power supply power connector instead giving more burden on the PCI-E slot. hence in case of 1070FE there most likely no more than 75w being pull from the slot even when the card being OCed.

 

[Bstraw53]

@Bstraw53: Or bookmark this page. http://www.nowinstock.net/computers/videocards/nvidia/gtx1080/

And THIS is the EXACT type of thing I was looking for.

Thank you, you are the MVP.

 

[TechyInAZ]

http://www.newegg.com/Product/Product.aspx?Item=N82E16814487261

I just noticed something funny about the accessories of this GTX 1060, why does it come with an 2x 6-pin to 8-pin PCIE adapter? The card only has a single 6-pin.

 

[Ambular]

And THIS is the EXACT type of thing I was looking for.

Thank you, you are the MVP.

\o/

 

[TehPenguin]

@Techy

I think it went something like this :
"Hey, boss, we need to make the 1060 bundle, what do we throw in? "
"Are some of those mouse pads still there? "
"No."
"What about the sli bridges? Oh, right... Some nVidia pens maybe? "
"I gave those to the boys at the office... But...uh...We have some 6to8 pin power adapters. "
"Yeah, throw that in. That'll add some weight."

 

[msproject251]

I have a setup with an i7 4790 and a gtx 750 and i want to upgrade to the 1070 or 1080, which one should I get?

 

[renz496]

depends on how much you willing to spend. of course looking from performance per dollar aspect the cost increase going to 1080 from 1070 are not worth it. but that is nothing new to flagship gpu. even 1070 already a few time faster than 750.

 

[almostdecent]

yah plenty of custom cards available now. stock is still low and price gouging is in full effect from many stores. but there are lots of options

see page 1 of this thread for a list of all the models, specs and reviews.

A good source for what is in stock and where is:

https://www.nowinstock.net/computers/videocards/nvidia/gtx1060/

 

[Math Geek]

yah plenty of custom cards available now. stock is still low and price gouging is in full effect from many stores. but there are lots of options

see page 1 of this thread for a list of all the models, specs and reviews.

A good source for what is in stock and where is:

https://www.nowinstock.net/computers/videocards/nvidia/gtx1060/

good enough that it has been linked on page one of this very thread for weeks now 😀 but thanks for the heads up. some folks may have forgotten about the site.

 

[melenie]

So I've managed to push my msi 1080 gamin x beyond that 2Ghz clock sweet spot. Just overclocking the old fashioned way. Not messing with GPU boost 3.0.

Currently running 2.125Ghz clock with 5.39Ghz ( 10.78Ghz effective) mem clock.

Doing all this whilst sitting at a very very cool 36degrees!

What do you guys think of these numbers? Would I be better off scaling the core clock down a little to increase the mem clock?

 

[TechyInAZ]

So I've managed to push my msi 1080 gamin x beyond that 2Ghz clock sweet spot. Just overclocking the old fashioned way. Not messing with GPU boost 3.0.

Currently running 2.125Ghz clock with 5.39Ghz ( 10.78Ghz effective) mem clock.

Doing all this whilst sitting at a very very cool 36degrees!

What do you guys think of these numbers? Would I be better off scaling the core clock down a little to increase the mem clock?

I think core clock still trumps memory clock, but don't quote me on that.

 

[renz496]

it is. don't think that there will be much benefit to the performance if the card is not that bandwidth limited to begin with.