Watercooled GTX 1080s OC to 2.5 Ghz And Out Preforms P100

[awesomeness10120]

7 minutes ago, ShadowTechXTS said:

Imagine this on phase-change.

More like LN2

 

[SteveGrabowski0]

3 hours ago, super_skank said:

There will be a reason these cards have 1733mhz as there boost and not 2100mhz....i suspect only 50% will get to 2000 nevermind 2100....if not they would have been clocked higher from the get go.

I suspect the 1080 will have a lot of overclocking headroom on average. Nvidia knows they're going to be stuck on this 16 nm node for a few years, so it wouldn't surprise me to hear they were conservative with core clocks just like they were with the 680. Plus people expect low power consumption after the 980. Then they can do exactly what they did with Kepler, releasing a very cut down GP100 as the GTX 1180 and upping the TDP to 210-220W on the GTX 1080 so they can overclock and rebrand it as the GTX 1170. Nvidia was advertising the card being capable of 2GHz overclocks in the promotional video they showed at the launch event, so I imagine 2GHz will not be hard for it to hit.

[grayperview]

6 minutes ago, SteveGrabowski0 said:

I suspect this card will have a lot of overclocking headroom on average. Nvidia knows they're going to be stuck at this 16 nm node for a while, so it wouldn't surprise me to hear they were conservative with core clocks just like they were with the 680. Plus people expect low power consumption after the 980. Then they can do exactly what they did with Kepler, releasing a very cut down GP100 as the GTX 1180 and upping the TDP to 210-220W on the GTX 1080 so they can overclock and rebrand it as the GTX 1170. Nvidia was advertising the card being capable of 2GHz overclocks in the promotional video they showed at the launch event, so I imagine 2GHz will not be hard for it to hit.

Maybe, but 1700mhz is very conservative if they can easily do 2100mhz like is being speculated, i suspect it's closer to the limit on the majority of the yield they have had so far, if not why not clock them to boost to 1900mhz at least, i suspect 2100mhz is around the limit, a really good one with a custom VRM may get to 2250mhz...

 

2500mhz doesn't sound plausible to me no matter how good the power delivery is or how good the cooling is....Even watercooled 980ti had thermal headroom on water, and brilliant custom VRM's but the silicon still crapped out at not much more than 1550mhz...

 

It will be interesting to see, and i should have the money to pick one up in a few weeks, so lets just wait and see.

[iamdarkyoshi]

9 hours ago, DocSwag said:

Where's your source?

Oh. Right. That site.

 

 

K

[Lays]

31 minutes ago, super_skank said:

Maybe, but 1700mhz is very conservative if they can easily do 2100mhz like is being speculated, i suspect it's closer to the limit on the majority of the yield they have had so far, if not why not clock them to boost to 1900mhz at least, i suspect 2100mhz is around the limit, a really good one with a custom VRM may get to 2250mhz...

 

2500mhz doesn't sound plausible to me no matter how good the power delivery is or how good the cooling is....Even watercooled 980ti had thermal headroom on water, and brilliant custom VRM's but the silicon still crapped out at not much more than 1550mhz...

 

It will be interesting to see, and i should have the money to pick one up in a few weeks, so lets just wait and see.

It's 1733 "rated" boost clock, when gpu-boost kicks in, it will be more along the lines of 1800-1850, depending on ASIC value & temp of the card. 

[Sky Daddy]

10 hours ago, Coaxialgamer said:

That's insane. But i suspect the gdd5X ( 10gb/s out of the box ) will be a bottleneck and only hbm2 will be capable of handling that efficiently . That , and no cpu will be capable of handling it outside of 4k

well, if you have a 1080 and aren't playing at 1440p or higher or vr then your doing it wrong

 

still though, max details + stupid custom resolutions could bring the performance back down to earth 

[M4st4M1nd]

Looks like this will be the card I wait for. I just hope it doesn't take EVGA too long to release it.

[Coaxialgamer]

7 hours ago, niofalpha said:

... And why would you think that?

Saying as GDDR5 isn't a massive (if one at all) bottle neck, and GDDR5X is roughly 25% faster assuming no other changes are made to it's memory interface then It'll be the same (memory speed- performance) if the rumors are to be believed, It'll coincide perfectly with the performance of the 1080.

 

And if you're talking about Overclocking then you can Overclock the VRAM too.

Hopefully... Just assuming Nvidia's not pulling a Fury.

well , bandwith is already pretty limited on maxwell , and this should be quite a bit faster. The gtx 1080 has memory bandwith of ~320GB/s . That's roughly the same bandwith as an r9 290x... for a card that's quite literally twice as fast.

not to mention this is the first time gddr5X cards are hitting the market . You can bet gddr5X won't overclock as well as current gddr5. 

[ThinkWithPortals]

Both Jeremy Clarkson and JackSepticEye must now work at NVIDIA.

I can picture the meeting now:

 

Jen-Hsun Huang: So, what shall we work on with the 1000 series?

Jeremy Clarkson: POWAHHHH!!!!!!

JackSepticEye: SPEED IS KEYYY!!!!!

Jen-Hsun Huang: We'll get right on it!

 

Can't wait till the 1070 is launched... it will be mine, oh yes... my precious...

[Dabombinable]

39 minutes ago, Coaxialgamer said:

well , bandwith is already pretty limited on maxwell , and this should be quite a bit faster. The gtx 1080 has memory bandwith of ~320GB/s . That's roughly the same bandwith as an r9 290x... for a card that's quite literally twice as fast.

not to mention this is the first time gddr5X cards are hitting the market . You can bet gddr5X won't overclock as well as current gddr5. 

I actually found that with Maxwell, overclocking the memory does nothing for bandwidth at all-overclocking the core makes massive differences (even to the slow vRAM), so I think that Maxwell actually couldn't even make full use of the possible GDDR5 bandwidth. All I can say is that if that's the case for the 1080 and 1070, then GDDR5X is kind of a waste.

[Lays]

4 minutes ago, Dabombinable said:

I actually found that with Maxwell, overclocking the memory does nothing for bandwidth at all-overclocking the core makes massive differences (even to the slow vRAM), so I think that Maxwell actually couldn't even make full use of the possible GDDR5 bandwidth. All I can say is that if that's the case for the 1080 and 1070, then GDDR5X is kind of a waste.

The VRAM oc still helps, especially in synthetics. It definitely isn't as big of a difference as core OC though.

[Dabombinable]

1 minute ago, Lays said:

The VRAM oc still helps, especially in synthetics. It definitely isn't as big of a difference as core OC though.

I only had miniscule increase in actual (aka not from GPUZ) memory bandwidth with a memory OC to 8GHz on my G1 Gaming, the core clock on the other hand increased everything drastically including memory bandwidth. I honestly don't think there is a point in overclocking vRAM at all unless your running something with DDR3 vRAM or very low clocked GDDR5.

[patrickjp93]

9 hours ago, niofalpha said:

... And why would you think that?

Saying as GDDR5 isn't a massive (if one at all) bottle neck, and GDDR5X is roughly 25% faster assuming no other changes are made to it's memory interface then It'll be the same (memory speed- performance) if the rumors are to be believed, It'll coincide perfectly with the performance of the 1080.

 

And if you're talking about Overclocking then you can Overclock the VRAM too.

Hopefully... Just assuming Nvidia's not pulling a Fury.

Because the raw bandwidth of 10GHz GDDR5X on a 256-bit bus is only marginally higher than the 980TI's memory bandwidth, and we're already exceeding that level of performance. If the memory can't be overclocked much, it will doubtlessly become a bottleneck.

 

And as for overclocking, Micron is not well-known to make good overclocking memory. Rock-solid, lower heat memory, yes, but overclockable beyond 10-15%? Not really...

[Master Disaster]

16 hours ago, FirstArmada said:

 

This is crazy didn't expect clock speeds to reach this high ever on water,

 

TLDR : P100 has 10.6 TFLOPs of single precision performance and a GTX 1080 at 2.1 Ghz has 10.7 TFLOPs of single precision performance and at 2.5 Ghz the 1080 will have 12.8 TFLOPs

 

 

 

Source: http://wccftech.com/geforce-gtx-1080-blow-p100-gpu-single-precision-performance-variants-including-25-ghz-liquid-cooled-edition/

P100 on Water though...

 

BOOM!!!

[grayperview]

On 14/05/2016 at 6:51 AM, Lays said:

It's 1733 "rated" boost clock, when gpu-boost kicks in, it will be more along the lines of 1800-1850, depending on ASIC value & temp of the card. 

yes i know that, still very conservative considering they were wanking themselves silly over 2100...

[awesomeness10120]

1 hour ago, patrickjp93 said:

Because the raw bandwidth of 10GHz GDDR5X on a 256-bit bus is only marginally higher than the 980TI's memory bandwidth, and we're already exceeding that level of performance. If the memory can't be overclocked much, it will doubtlessly become a bottleneck.

 

And as for overclocking, Micron is not well-known to make good overclocking memory. Rock-solid, lower heat memory, yes, but overclockable beyond 10-15%? Not really...

At the reveal event it was shown to overclock to 11000 MHz.

[Commander Llama]

I wonder what this means for air coolers too. My G1 980Ti runs amazingly cool compared to my previous GPUs. Unlike my 4790k...

[patrickjp93]

5 hours ago, awesomeness10120 said:

At the reveal event it was shown to overclock to 11000 MHz.

Oh, so instead of 340 it's 374GB/s vs. the 980TI's 336, and you think it's enough of a difference to make up for the huge difference in core performance?

[RagnarokDel]

Notice how they dont talk about double precision performance? Inb4 it still gets outperformed by a 7970.

[awesomeness10120]

1 hour ago, patrickjp93 said:

Oh, so instead of 340 it's 374GB/s vs. the 980TI's 336, and you think it's enough of a difference to make up for the huge difference in core performance?

No, I'm just saying that the memory deficet can be solved. 

[Trik'Stari]

Gets excited, then sees the source

[MoonlightSylv]

17 hours ago, awesomeness10120 said:

More like LN2

 

5GHz GPU clock

[shermantanker]

14 hours ago, M4st4M1nd said:

Looks like this will be the card I wait for. I just hope it doesn't take EVGA too long to release it.

Same here. I want to have it before battlefield is out. 

[awesomeness10120]

14 minutes ago, ShadowTechXTS said:

5GHz GPU clock

I'll see what Kingpin brings to the table

 

[FirstArmada]

On 5/13/2016 at 4:32 PM, awesomeness10120 said:

Fix for dark theme users. Go to edit and click the remove format button.

 

Finally understood what you meant, fixed!