Nvidia GeForce GTX 1070 8GB Pascal Performance Review (Archive)

Page 4 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.

It isn't as if there were tons of mature DX12 games out there to choose from in the first place. It would not make much sense for Chris to spend the bulk of his limited hands-on time on DX12 when the bulk of games that are going to be played on it are still going to be DX11.
 
I used to think the same thing, and logically it makes sense, but 2k actually means 1440p. It's stupid (in my opinion), but for whatever reason the convention of 2k = 1440p has caught on.

[/quotemsg]

I think both 2K and 25 x 14 are stupid abbreviations. What is wrong with 2560 x 1440 or 1440p?
 
So, this is the successor to the GTX 970, but you didn't include benchmarks for that card? Am I missing something here? How are us 970 owners supposed to see if this is worth the money?

Look at the 980 on the charts..now imagine the 970 slightly below (10-20% below) that of the 980. 970 is good for 1080p but not so much for 1440p.
 


1080p is a waste with either a 1080 or 1070. that's why it is not included. if you are only playing 1080p, then stick with a 970 or wait for the 1060 to show up whenever that is. it's not worth the time t test just to say "look at that big number". if you REALLY need them, look for 980ti/titan x 1080p numbers and add a few to that. oh wait there are few of them as well, since they are also WAY overkill for 1080p.

So its only me here that considers less then 100fps playable? And optimal around 100 and above... I might have a eye sight above average that 60 fps in games scrathes my eyes.. tho as desktop usage and office does not matter.... Sorry but i dot see the new pascal as overkill for 1080p 144hz... but just the right cards depending on what you play and whoever plays fps will get those for 1080... Especially if as nvidia is claiming reduced imput lag over previous generation... tho it seems no one can test that imput lag.

Personnally i want one of those for 1080 144hz but not sure which yet so i will wait for non reference to surface and reviews for those to see which is better...
 


1080p is a waste with either a 1080 or 1070. that's why it is not included. if you are only playing 1080p, then stick with a 970 or wait for the 1060 to show up whenever that is. it's not worth the time t test just to say "look at that big number". if you REALLY need them, look for 980ti/titan x 1080p numbers and add a few to that. oh wait there are few of them as well, since they are also WAY overkill for 1080p.

Why would it be a waste? What if i want to use that 100Hz/100fps @ 1080p?
 
So, this is the successor to the GTX 970, but you didn't include benchmarks for that card? Am I missing something here? How are us 970 owners supposed to see if this is worth the money?

They include the 980; you can easily estimate where the 970 would fall from that (about 10% less performance than the 980).
 
I'm really interested to see how the polaris flagships will turn out (not the mainstream cards announced). For the first time in a long time, AMD is actually on a smaller die size (14nm) than Nvdia, and I'm betting the several-steps shrink from 28nm to 16nm is responsible for a huge portion of pascal's performance.

At the same time, I'm worried the narrative of the last couple generations will play out again: Nvidia releases a new X80. AMD announces their new generation and releases early benches showing the AMD flagship topping the X80 Nvidia card. Some time after the AMD announcement, but before the AMD cards hit the market, Nvidia releases the X80 ti and/or Titan cards in the lineup, which beat AMD's.
 


Actually if you upgrade to a 1070 you wont need to think about an upgrade even when you get a 1440p or 144hz monitor.
 
think I am going to buy another 970 for SLI or wait for the red team to come out with something. From the looks of it a 970 SLI setup would kill this card.

I have a 970sli setup and honestly I contemplated multiple times selling them for a 980ti instead. 980ti is = 970sli in 1080p gaming and in 1440p gaming the 980ti is on average about 4fps slower and 2ftps slower in 4k on average. I could easily see the 1070 being as fast as or faster than 970sli. I would reconsider if pricing for you to upgrade to a 1070 is equal or comparable to adding another 970. I personally will be getting a 1080 and selling my 970's.
http://www.pcgamer.com/nvidia-geforce-gtx-980-ti-review/
 
I am I the only one that wished they would come out with a 4Gig version of 1070 ? That could retail at $329 instead of the $379 something for us folk who game at 1080/1440p. I know not going to happen but I just feel unless your going 4k you don't really need 8 gigs of memory. Heck a 4gig 1080 using Gddr5x would be sweet I would be ok even if they removed the SLI option maybe 1080/1070 LE card. I guess I can dream.
 


Probably so. At 1440p and ultra settings + 4xAA, several of my games are pushing over 3GB of VRAM allocation with 970 SLI (FC4, Crysis 3, Shadow Of Mortar). That's fine for 1080p, but not for 2K and future gaming requirements. I would rather pay more for future proofing, especially at the sweet price/performance spot with the 1070.

The 1060 will fill that 4GB gap as a sub $300US card, and likely be performing somewhere between a current 970 and 980. Also the biggest impact on reduced VRAM is in minimum FPS performance. That was proven in tests of 2GB vs. 4GB GTX 680s four years ago. Of course, you need to have a wide enough memory bus to take advantage of that extra memory. A 4GB 960 is really a waste of money and memory since its rather narrow 128-bit bus doesn't take full advantage of the extra memory.

People have short memories. The 2GB GTX 680 was a $520 GPU, and the 2GB 670 was a $400 GPU. And the 670 destroyed the former top of the line $500 1.5GB GTX 580. This has been the pattern for a while: more VRAM with every new generation (or refresh) at the same GPU tier segment (275->470->570->670->970->1070).
 
Do I see a trend here? why is Ashes of the Singularity is the number one listed software? was it alphabetically listed maybe? but why? when it does not get close to all other games test results, when a game is specifically tuned for a video vendor of course it would be useless to use as a cage anything, as an owner of Ashes of the Singularity I would like you to not use it anymore
 


Because it's the only real DX12 game.
 

With its development roots in the Star Swarm demo as a showcase for AMDs Mantle, its hard to say definitively whether we're looking at DirectX 12 performance, or AMD optimizations. It's like if Project CARS happened to be a DirectX 12 game. We'd all be talking about how much ahead Nvidia is in DirectX 12. As it is, both of these games have vendor-specific optimizations baked deep into the mix, and really shouldn't be used for extrapolating or generalizing any conclusions outside of the specifics for that benchmark.
 
I haven't read all the comments, so I don't know if someone already pointed this out, but if you look at the two bare PCBs you can see where there are actually places for a sixth phase on the power supply, and a spot for a second 6-pin power in, as well as a spot for another capacitor next to that second 6-pin.

So... This is complete speculation, but it kinda looks like the PCB is ready to be used for another card a step above the 1080 even, which means that nicer card would also use the same GDDR5X ram (since HBM2 would need a completely different layout) and a little more power with a second plug.
So are we looking at the 1080ti using the same PCB and ram but with more power? Or the next Titan? I'm guessing the 1080ti, since the next Titan would be using the GP102 and probably need a different PCB. But if it were the same PCB for the Titan, then that would confirm that the Titan will be using the same GDDR5X ram instead of HBM2.

Thoughts?

Update: Right before posting I shared my thoughts with my brother (who shared this article) and he rightly reminded me that the 980ti is similar to the Titan, so the 1080ti will likely be using the same GP102 as the new Titan, and the extra pins are likely for headroom for custom boards. So now I'm thinking they are there for third-party factor OC-ed boards.
 


Custom cards already have more power phases. I think the Asus Strix has 8 (or is it 10, can't remember).
 
I wonder why they cut so much hardware this time.
980 ? 970 was 16 to 13 SMs 81.25% hardware left
1080 ? 1070 is 20 to 15 SMs 75% hardware left

There are 4 GPCs. Turn one off you have 3. That makes sense to me. It is why they turn off part of three of the clusters in the 970 instead of one complete one is what I don't understand...
Personally, I think they decided that the weird way they cut the 970 down wasn't worth it since it also affected the cache and RAM, so this time they cut out a whole GPC and left the rest of it alone so it wouldn't be so disruptive to the rest of the chip. Also, this likely has to do with binning. This way they can use more chips by simply disabling the one bad GPC if there is one. In the 970 they likely disabled more working SMMs than they really needed to in order to cut it down.
But this is just speculation on my part. Not saying I don't agree that the ratio is worse, just some thoughts as to why.
 


I was updating my comment as you were writing. I didn't see it there. 😀 Makes sense.
 
"I think both 2K and 25 x 14 are stupid abbreviations. What is wrong with 2560 x 1440 or 1440p?"

1920x1080 = HD
2560x1440 = WQHD
3840x2160 = UHD
4096x2160 = 4K

Just use the proper acronyms for the respective resolutions and there will be no confusion. =)
 


evga classified has 14+3 phases and there are a number of 10+2 designs out there as well. hopefully they can up the voltage so it can actually be used. so far looks like voltage limits by nvidia are gonna keep the 1080's from being all they could be.

not seen custom 1070's yet but they may be just as limited...
 
I would really love to see a return to the more comprehensive comparisons against older cards as well. It would be nice to have dedicated benchmarks that you run on every card that overlap by a few generations. Perhaps a revamp of the Chart system is in order and the products are added immediately after the review. Seeing how my cards personally compare to the current gen would be ideal.
 
Status
Not open for further replies.

TRENDING THREADS