Nvidia Pascal Reviews [1080XP, 1080ti, 1080, 1070ti, 1070, 1060, 1050, and 1030]

If they changed the profile (increase power limit and fan speed) power consumption and noise would obviously be less favorable.
68% fan speed seems quite high to me. Don't have much experience with blowers though.
He says it is quieter than previous models, but it is a perception thing and people have different sensitivities/tolerances.
Cheers
 
Why was Anandtech not given a review card? Y'all are amongst the biggest sites in the biz...

They were given a review card. They put up some performance preview numbers in the NDA-expiring day as everyone else. Back then, they stated they would need more time to write a proper in-depth review (because it's the first Pascal card, they'll go deep on the architecture) and that's why it's still coming.
 
Comparison
Titan x vs gtx 1080
transistors 8bilion vs 7.2bilion (1.11x)
sp 3072 vs 2560 (1.2x)
rop 96 vs 64 (1.5x)
tmu 192 vs 160 (1.2x)
l2 cache 3mb vs 2mb (1.5x)

And my question is, what nvidia did with transistors in pascal because titan x which have only 1.11x more transitors have much more rops, cache, sp, tmu. Did they somehow use large quantity of transistors to increase clock speed ?
 
Did they somehow use large quantity of transistors to increase clock speed ?
This is likely. If your processor frequency is limited by interconnect delay (simple wire transmission over the multiple layers of metal) then you try to minimize that delay by moving circuit elements around to minimize the distance and/or choose "faster" layers of metal especially optimized for it. But the complementary solution is to break especially long paths in two (or more) shorter hops and use a transistor as a repeater. You now can clock data at higher frequencies at the expense of one or more clock ticks of delay, like a pipeline. On a GPU this is very acceptable since it's not latency sensitive unlike a CPU. There's also possible power savings by using a repeater like this since you're charging up less metal wire at a time, requiring less drive voltage and current.
This optimization is half physics simulation and half voodoo art. A poor interconnect delay model can cripple a design, which in fact happened to the first spin of Fermi, delaying its release by multiple months.
 
Interesting they did not use any of the new SLI bridges but the system identified both cards.
Both of these tests were run at 4K resolution, with max image quality settings. Scaling was excellent, though we should point out we did NOT use one of the new high-bandwidth SLI bridges. NVIDIA's drivers didn’t throw any errors, however, and properly identified the system as SLI capable when the second card was installed.
 
Also aftermarket parts vary in cooling performance, I've found Gigabyte to have the best ones recently (actively cooling VRM too).
 
Comparison
Titan x vs gtx 1080
transistors 8bilion vs 7.2bilion (1.11x)
sp 3072 vs 2560 (1.2x)
rop 96 vs 64 (1.5x)
tmu 192 vs 160 (1.2x)
l2 cache 3mb vs 2mb (1.5x)

And my question is, what nvidia did with transistors in pascal because titan x which have only 1.11x more transitors have much more rops, cache, sp, tmu. Did they somehow use large quantity of transistors to increase clock speed ?

That is an interesting question I also wondered about that.
And if you do the math for the 15.3B transistor P100 it's even much more pronounced.
But then this is the real Pascal.

GM200 has 20% more compute/texture resources and 50% more ROP / L2 / rasterizers compared to GP104 for 11% more transistors.
You can spend more transistors to increase the clock, by introducing more pipelining, but going back that route is rather unlikely IMHO. (Paths are already significantly shorter due to more dense transistors and 300mm die vs 600mm die and of course there is finfet and perhaps better routing)

There must be some hidden goodness in the GP104 that Nvidia hasn't talked about.
 
The ACX cooler was on the 980Ti was pretty awful so I went with the Gigabyte G1 which is amazing. Curious to see if EVGA dramatically improved the ACX or not. Looks cool?
Do the Gigabytes cards lack coil whine?. I have also heard good things about MSI cards.
 
Also aftermarket parts vary in cooling performance, I've found Gigabyte to have the best ones recently (actively cooling VRM too).

Asus' Direct CU II coolers are rather nice, though there's like a full version for highest end cards and a version that is named the same for middle end that isn't as high specced and not able to cool as effectively(Why they use the same name instead of like Direct CU II++ for the higher end cards I don't friggin' know). I have one on my GTX 660, and I've never gone above 70 degrees outside of torture tests like Furmark, and I live in a very hot area with no AC in my room.

These coolers are amazing.
 
The ACX cooler was on the 980Ti was pretty awful so I went with the Gigabyte G1 which is amazing. Curious to see if EVGA dramatically improved the ACX or not. Looks cool?

not sure if they improved it but should find out soon enough though.
 
The ACX cooler was on the 980Ti was pretty awful so I went with the Gigabyte G1 which is amazing. Curious to see if EVGA dramatically improved the ACX or not. Looks cool?

There was an issue on some G1 cards with coil whine but I guess I got lucky.

I generally run it at 1480ish for games that need the extra horsepower. Otherwise running it stock, it just runs without much noise.
 
Back
Top