Nvidia Pascal Announcement

I'm not sure whether drivers or Windows 10 update or whatever else solved the Dx12 performance "issue" but it seems like I'm getting more consistent performance and higher fps with smoother frame times with Dx12 over Dx11:

Dx11
tomclancysthedivisionj9ll4.jpg


Dx12
tomclancysthedivisiona4aox.jpg
 
I'm not sure whether drivers or Windows 10 update or whatever else solved the Dx12 performance "issue" but it seems like I'm getting more consistent performance and higher fps with smoother frame times with Dx12 over Dx11:

Dx11
tomclancysthedivisionj9ll4.jpg


Dx12
tomclancysthedivisiona4aox.jpg

Excuse me but, what is your monitoring tool ? Doesn't look loke RTSS.

Thx.
 
It is Afterburner + RTSS but it's the latest beta which includes live graphs and other goodies :)

Thx, I'll look at that. I use RTSS 7 beta 19 +HWinfo for my stats. I've all I need, but the presentation is not that good. And I had some problem with afterburner in the past :/
 
Last edited:
Is the GTX 1070 Ti based on GP104? When I read the name I naturally assumed it would be a further cut down GP102.

I see where you're coming from. Nvidia's lineup probably needs something GP102-based between 1080 Ti and 1080 more than it needs something between 1080 and 1070.

However, based on what we know, it's strongly suggested that this would be GP104-based.
  • Purely based on historic Nvidia naming conventions, a 1070 Ti will perform between a 1070 and 1080.
    • The 660 Ti performed between a 660 and 670.
    • The 1050 Ti performs between a 1050 and 1060.
    • And so on.
  • The "ASUS Strix O8G" in that image is an ASUS model number used for "Overclocked 8 GB" cards such as the 1070 and 1080, while ASUS Strix 1080 Ti use "O11G" (or "11G" for stock cards). I outlined several real-world examples in the post below.
I get where you're coming from. The thing that convinced me was the ASUS model number. For the Strix line, "O8G" means "Overclocked 8 GB".

This shows up on model numbers for the 1080 and 1070.

  • ASUS GTX1080-8G
  • ASUS ROG STRIX-GTX1080-8G-GAMING
  • ASUS ROG-STRIX-GTX1080-O8G-11GBPS
  • ASUS ROG STRIX-GTX1080-A8G-GAMING
  • ASUS ROG STRIX-GTX1070-O8G-GAMING
  • ASUS ROG STRIX-GTX1070-8G-GAMING

The 1080 Ti would've used an "O11G" like other ASUS 1080 Ti cards, e.g.

  • ASUS STRIX-GTX1080TI-O11G-GAMING
  • ASUS STRIX-GTX1080TI-11G-GAMING

PCPartPicker is a nice source for these kinds of model numbers: https://pcpartpicker.com/products/video-card/#m=8&sort=-price&c=380,373,369,367,390,221

3VqSaqW.jpg
 
Is the GTX 1070 Ti based on GP104? When I read the name I naturally assumed it would be a further cut down GP102.

This reminds me of the flipside of this conclusion where many people think that the 1080Ti is just a souped up 1080 and isn't as big of a difference as 1070 to 1080. I've read many posts of people astonished that the difference is instead twice of the latter and it's a whole different chip just shy or beating the Titan. For all the talk about bad AMD marketing, this is quite the blind spot nvidia have. The titan branding is not enough when calling it, say, 1090Ti might attract more sales.
 
What's wrong with covering all your bases?

If being threatened by the competition were the only reason to release GPUs, Nvidia would have had to wait for another year before releasing a Titan Xp or a 1080Ti.

AMD just semi-forced them into adding more pro features to the Titan. This seems like an appropriate response. :)
It's not though. To extract revenue they need ways for even a 1080 owner to upgrade without going all the way to 1080Ti/Titan. A 1070Ti would be a play for 1060 owners and an improved 1080 to close the gap with To makes sense. Pick up the sales before repeating the process with Volta. There may be a faster 1080 variant coming though.

Nvidia's lineup probably needs something GP102-based between 1080 Ti and 1080 more than it needs something between 1080 and 1070.
1080TI over 1080Ti? Naming is already reversed though and To is relatively new in comparison.
 
I'm not sure whether drivers or Windows 10 update or whatever else solved the Dx12 performance "issue"
I haven't gotten the win10 update on my main rig yet. :( On my old backup PC, MS even popped up an offer to cut to the head of the line, but since it's just my backup I declined, assuming I'd get the same offer on the PC I use every goddamn day. Nope! *sigh*
 
It's not though. To extract revenue they need ways for even a 1080 owner to upgrade without going all the way to 1080Ti/Titan. A 1070Ti would be a play for 1060 owners and an improved 1080 to close the gap with To makes sense. Pick up the sales before repeating the process with Volta. There may be a faster 1080 variant coming though.

You surely are aware that the grand majority of people upgrade every 2, 3 or even 4 years and not every year right? Majority of potential 1070 Ti customers are people with GTX 770 or at most 970/980.
 
You surely are aware that the grand majority of people upgrade every 2, 3 or even 4 years and not every year right? Majority of potential 1070 Ti customers are people with GTX 770 or at most 970/980.
Tell that to Apple with their 6 month iPhone cycle. I realize most users won't cycle products that quickly, but there is a segment of consumers with lots of cash that will. Plus Nvidia can just send marketers on to forums to suggest selling old cards and upgrading to the newest regularly to drive sales of higher margin parts.
 
Plus Nvidia can just send marketers on to forums to suggest selling old cards and upgrading to the newest regularly to drive sales of higher margin parts.
Funny, I was just thinking of a guy who proposed this strategy in a german forum for a long time now. So he's just a marketing shill? Sad times we live in.
 
Funny, I was just thinking of a guy who proposed this strategy in a german forum for a long time now. So he's just a marketing shill? Sad times we live in.
Possibly, but wouldn't need to be a shill to promote the process. Takes actual buyers to make it work and ultimately turns into a subscription model. That and frequent refreshes of the top product line.

Well, Apple only releases new iphones once a year. Not that most people upgrade nearly that often, especially these days when phones are already as mature and capable as they are.
Around here they still have lines to buy the latest model. They have has faster refreshes with the various S models, but we're getting off topic here.
 
Stumbled on this interesting comparison between the Quadro P6000 (GP102) and the Quadro GP100 in the V-Ray benchmark. GP100 has 10 TFLOPS of compute (3,584 cores @1430MHz) and 720GB of HBM2 bandwidth, while GP102 has 12 TFLOPs (3,840 cores @1530MHz) and 480GB of GDDR5X bandwidth. So the GP102 is 20% faster on compute, while the GP100 has 50% more bandwidth.

The results are the GP100 is often anywhere between 30% to 40% faster than GP102! despite being slower with compute, is V-Ray that much bandwidth limited, or there is something else at play here?

https://labs.chaosgroup.com/index.p...pu-benchmarks-on-top-of-the-line-nvidia-gpus/
 
Last edited:
Back
Top