NVIDIA GeForce GTX 1080 3DMark Firestrike Benchmark Results Leaked - beats 980ti and Fury X

[TidaLWaveZ]

4 minutes ago, ThinkWithPortals said:

2.1GHz? Much speed! Very wow!

  Reveal hidden contents

 

Seriously though, 2.1GHz OC? That's mental.

It's bonkers mental.

[Inkz]

BREAKING NEWS!: The new flagship card is faster than the old one!

[zMeul]

who said there aren't drivers for GTX1080?

[Bouzoo]

28 minutes ago, zMeul said:

who said there aren't drivers for GTX1080?

Saw that earlier, they said the memory will run @2.5 GHz (for memory bandwidth) while it's clearly 1.25 here so I'm not sure what's with these numbers.

[zMeul]

3 minutes ago, Bouzoo said:

Saw that earlier, they said the memory will run @2.5 GHz while it's clearly 1.25 here so I'm not sure what's with these numbers. 

isn't it double pumped ? 1.2*2=2.5

this is GDDR5X, so I'm guessing it's different base clock than we're used to GDRR5

[Bouzoo]

Just now, zMeul said:

it's double pumped: 1.2*2=2.5

That crossed my mind, for total multiplier of 8, but I never figured it might have a problem with showing 2.5 GHz.

[Bouzoo]

5 hours ago, MageTank said:

I am seeing a lot of $400 980 Ti's on ebay. If they drop to about $300, I might still buy one over the GTX 1080, simply because that performance for that price just can't be beat. The only concerning factor i have, is whether or not i can get a GTX 980 Ti to run on a 500w platinum PSU, lol. Stupid SFF power restrictions x.x

You can always start bidding under $200, you might get lucky. unlikely

[Zaryab]

HOLY SHI888888~!!!!! 65% PERFORMANCE INCREASE FROM A GTX 980 AND A 35% INC FROM A 980TI!!!! AHHHHHHHHHH FUUUUUUUUUUUUUTUUUUUUUUREEE

[Master Disaster]

8 hours ago, Spectrez said:

they said the same thing about maxwell, but still havent delivered.

 

i fear that Nvidia didnt aim for Async computing, and focused on a gpu architecture that still only support software async computing  

Unverified at this point but yeah, if true its a huge chunk of upgrade that's for sure.

[i_build_nanosuits]

19 minutes ago, Zaryab said:

Why do people like you even exist? You moron. Its not just another flagship card... its an amazing flagship card... its 65% improvement over the 980... that DOESNT HAPPEN EVERY GENERATION OF CARDS. 35% better than a 980ti and is cheaper??? yeah its a big deal.

...and it has more memory, and faster memory as well...and it's early drivers for Pascal, it will only get better from now on!

...but, at stock it's 32% faster than the stock GTX 980ti ...(76% x 1.32 = 100.32) so not 35..yet.

...stock boost clock for GTX 980ti is 1075mhz...my 980ti is running at 1460mhz...therefore i have a 36% overclock in place...now do the math...

 

but i have to agree...a 300mm² chip that can do that, at that price...at 180W TDP, with more memory, and better memory...is just fantastic.

Not worth the upgrade for me personally...but great none the less.

[Coaxialgamer]

12 hours ago, Thony said:

According to Jay2Cents, nope. He said we are getting to the point where heat is not the limit anymore but other factors are, hence why Nvidia demo had it running at 67C.

what else could affect it ? Process node ?( surely 14nm is enough for 3ghz ) Power ? ( does nvidia restrict power limits to keep it's TDP low ?)

[Coaxialgamer]

2 hours ago, zMeul said:

who said there aren't drivers for GTX1080?

I'm kind of dissapointed with the memory bandwith though ... About the same as a 290x , for a card that should be almost twice as fast. And considering this is first gen gddr5X ( 10gb/s ) , it probably  won't overclock very well. I fear there might be a bottleneck .

[Thony]

20 minutes ago, Coaxialgamer said:

what else could affect it ? Process node ?( surely 14nm is enough for 3ghz ) Power ? ( does nvidia restrict power limits to keep it's TDP low ?)

Here: 

 

 

 

[MMKing]

2 hours ago, Bouzoo said:

Saw that earlier, they said the memory will run @2.5 GHz (for memory bandwidth) while it's clearly 1.25 here so I'm not sure what's with these numbers.

The reason why Vram clock speeds fluctuate so wildly between different tech sites, journalists and software is because of the fuckery that is ''effective clock speed''. A bullshit made up term that makes it pretty much impossible for software to correctly report he clock speed 100% of the time.

 

First of all, before i explain, there is so such thing as effective clock speed. 3500MHZ GDDR5 does NOT have an effective clock speed of 7000MHZ. Just because each cycle transfers two bits, does NOT mean that it has twice the number of cycles. The 3500 million cycles per second, delivers 7000 million bits of data per second.

 

Now the situation got even worse with GDDR5X. Now the stupid assholes are reporting effective clock speeds of 10 000MHZ. Not only is the Vram name incorrectly named. GDDR5X is quad data ram. ''Graphics Quad Data Rate'' or ''GQDR'' is the correct non-misleading name. But the reported cycle rate is also wrong.

 

The Vram in the 1080 cycles 2500 million times per second, delivering 4 bits of data per cycle, and it transfers data on 8 chips with 32 lanes each for a total of 256 lanes. This equals 10Gb of data per lane. per second for a total of 2560Gb per second or 320GB per second.

 

As of now i am declaring war on the ''effective clock speed'' term. Either you say the correct clock speed, which is 2500MHZ on the 1080. If you really want to say 10 000, call it MbPS (Megabit per second).

[Zaryabb]

19 hours ago, MMKing said:

The reason why Vram clock speeds fluctuate so wildly between different tech sites, journalists and software is because of the fuckery that is ''effective clock speed''. A bullshit made up term that makes it pretty much impossible for software to correctly report he clock speed 100% of the time.

 

First of all, before i explain, there is so such thing as effective clock speed. 3500MHZ GDDR5 does NOT have an effective clock speed of 7000MHZ. Just because each cycle transfers two bits, does NOT mean that it has twice the number of cycles. The 3500 million cycles per second, delivers 7000 million bits of data per second.

 

Now the situation got even worse with GDDR5X. Now the stupid assholes are reporting effective clock speeds of 10 000MHZ. Not only is the Vram name incorrectly named. GDDR5X is quad data ram. ''Graphics Quad Data Rate'' or ''GQDR'' is the correct non-misleading name. But the reported cycle rate is also wrong.

 

The Vram in the 1080 cycles 2500 million times per second, delivering 4 bits of data per cycle, and it transfers data on 8 chips with 32 lanes each for a total of 256 lanes. This equals 10Gb of data per lane. per second for a total of 2560Gb per second or 320GB per second.

 

As of now i am declaring war on the ''effective clock speed'' term. Either you say the correct clock speed, which is 2500MHZ on the 1080. If you really want to say 10 000, call it MbPS (Megabit per second).

SO TLDR; is gddr5x a LOTTT better than gddr5 or no? 

[MMKing]

12 hours ago, Zaryabb said:

SO TLDR; is gddr5x a LOTTT better than gddr5 or no? 

Assuming the buss width is identical. You're looking at a about 40% improvement over the previous generation.

 

Luckily, for easy comparison. The 980 and the 1080 has an identical buss width, and the 1080 has bandwidth of 320GB/s while the 980 has 224GB/s or a 42.8% improvement.

 

As for overclocking. We know GDDR5 easily adds 200-400MHZ extra on the memory for 7.4-7.8 Gb/S per lane. If GDDR5X overclocks very poorly or worse, not at all, the technology is more or less just an incremental increase in bandwidth. Though 3-4GHZ clocked memory is allegedly coming in 2017/2018 so it's possible that the OEM partners release 1080s with higher clocking memory down the road.

 

But in the end, it doesn't really matter how you look at it. GDDR5X is a positive.

[Lays]

On 5/13/2016 at 2:08 PM, Coaxialgamer said:

I'm kind of dissapointed with the memory bandwith though ... About the same as a 290x , for a card that should be almost twice as fast. And considering this is first gen gddr5X ( 10gb/s ) , it probably  won't overclock very well. I fear there might be a bottleneck .

Why? It's already been shown over the past few years that memory bandwidth doesn't make a big difference in most scenarios. 

 

We've already seen during the tech demo it running at 1 ghz over stock specs, and they say it was just a random sample they grabbed.

[SurvivorNVL]

8 minutes ago, Lays said:

Why? It's already been shown over the past few years that memory bandwidth doesn't make a big difference in most scenarios. 

 

We've already seen during the tech demo it running at 1 ghz over stock specs, and they say it was just a random sample they grabbed.

It would have been great if they had not gone with Hynix's HBM, and just went with HMC(Sped up that train!), that way we'd have the high bandwidth of HBM2, and the glory of ultra low-latency, the issue HBM/HBM2 has.  Raw bandwidth?  Hell yeah.  Latency?  Not as great as GDDR5, if I recall for HBM.  Hopefully we get HMC in 2018.

[Lethal Seraph]

I really hope so. Why would they release a new architecture and only to be beaten by the predecessor? 

[Xsilent(X)]

buying 2 of these