K404 Posted November 15, 2012 Posted November 15, 2012 Obviously, it is too late for this cycle of 3DMark, but if FM could make a benchmark that is truly 100% GPU, even 5 years after release*, that would be awesome Based on how GPUs advance and CPUs advance, CPUs are "always" the bottleneck. To get a better 3DMark score, I need a better CPU much more than I need a better GPU. IMO, that's a bit wrong multi-GPU scaling is way off as well, especially at the top-end *I appreciate it is very difficult to guess how to code to make something future-proof Quote
FM_Jarnis Posted November 15, 2012 Author Posted November 15, 2012 Obviously, it is too late for this cycle of 3DMark, but if FM could make a benchmark that is truly 100% GPU, even 5 years after release*, that would be awesome Based on how GPUs advance and CPUs advance, CPUs are "always" the bottleneck. To get a better 3DMark score, I need a better CPU much more than I need a better GPU. IMO, that's a bit wrong multi-GPU scaling is way off as well, especially at the top-end *I appreciate it is very difficult to guess how to code to make something future-proof I want to see the system where 3DMark: Fire Strike Extreme is CPU limited... Even five years from now. Even with Quad GPUs. Sorry that I can't give more details yet but I'll just drop a little bit here as a starter; 2560x1440. The only obviously broken multi-GPU scaling bit with 3DMark 11 is the case of Quad Crossfire (driver issue). Granted, you need to use Extreme preset (it is there for that exact reason). Quote
Christian Ney Posted November 15, 2012 Posted November 15, 2012 Well not sure a lot of people will be able to run the extreme preset Quote
IanCutress Posted November 15, 2012 Posted November 15, 2012 A 2D calculation based GPU benchmark would easily scale over generations. But then no pretty pictures Quote
K404 Posted November 15, 2012 Posted November 15, 2012 I'll be honest.... it's not the answer I was hoping for Quote
FM_Jarnis Posted November 15, 2012 Author Posted November 15, 2012 Well not sure a lot of people will be able to run the extreme preset Note that the resolution is not a problem if you have smaller monitor. It runs fine no matter your actual monitor native resolution. I'm only talking about internal rendering resolution before scaling to your monitor native res... Quote
Christian Ney Posted November 15, 2012 Posted November 15, 2012 Note that the resolution is not a problem if you have smaller monitor. It runs fine no matter your actual monitor native resolution. I'm only talking about internal rendering resolution before scaling to your monitor native res... okok Quote
hokiealumnus Posted November 15, 2012 Posted November 15, 2012 Note that the resolution is not a problem if you have smaller monitor. It runs fine no matter your actual monitor native resolution. I'm only talking about internal rendering resolution before scaling to your monitor native res... Well that's a neat trick. Wonder if having a 1440p monitor will give a (minor) score boost since there isn't any processing power lost to scaling it down? ...and so it was, benchmarkers everywhere shed a tear that they'd have to upgrade their benching monitors too. Quote
FM_Jarnis Posted November 15, 2012 Author Posted November 15, 2012 (edited) Well that's a neat trick. Wonder if having a 1440p monitor will give a (minor) score boost since there isn't any processing power lost to scaling it down? ...and so it was, benchmarkers everywhere shed a tear that they'd have to upgrade their benching monitors too. Same "trick" is being used (in a slightly different way, but in general same idea) on 3DMark Vantage and 3DMark 11 already. Vantage extreme preset works just fine on a smaller monitor, for example. I guess in theory if you are squeezing out those last points for WR, it might matter. In "normal" benchmarking the difference is not meaningful. Edited November 15, 2012 by FM_Jarnis Quote
FM_Jarnis Posted November 15, 2012 Author Posted November 15, 2012 Like 1 % or more ? Less. Far less than normal variance between runs. But again, if you are trying to get 5 points more or something, it might matter. Quote
Christian Ney Posted November 15, 2012 Posted November 15, 2012 if it's far less than 1 % that's fine. Actually a good idea to bypass hardware limitation. Quote
IanCutress Posted November 15, 2012 Posted November 15, 2012 Certain mobile benchmarks do an 'offscreen 1080p' test to be consistent across screen sizes rather than run at native. Also stops scaling up/down as required Quote
FM_Jarnis Posted November 15, 2012 Author Posted November 15, 2012 Certain mobile benchmarks do an 'offscreen 1080p' test to be consistent across screen sizes rather than run at native. Also stops scaling up/down as required Excatly same thing. Next 3DMark uses only "offscreen", so to speak. No matter what resolution is used to render a particular test, it always uses your monitor native resolution to display it. Also allows circumventing all kinds of GPU/monitor scaling issues (I'm sure you've seen 3DMark 11 running in "centered" mode with large black borders around a small (720p) rendering area on some monitor/GPU configs...) Quote
Lucky_n00b Posted November 15, 2012 Posted November 15, 2012 Excatly same thing. Next 3DMark uses only "offscreen", so to speak. No matter what resolution is used to render a particular test, it always uses your monitor native resolution to display it. Also allows circumventing all kinds of GPU/monitor scaling issues (I'm sure you've seen 3DMark 11 running in "centered" mode with large black borders around a small (720p) rendering area on some monitor/GPU configs...) Whoa that's great to know :D Btw a Question : Does the "3DMark : Fire Strike" have an option to detect if the tesselation settings have been modified from the driver level during the runs/before we run it? (something like : preventing the 3dmark to actually run if it detected that tesselation level has been changed) Thanks Man Quote
FM_Jarnis Posted November 16, 2012 Author Posted November 16, 2012 Whoa that's great to know :D Btw a Question : Does the "3DMark : Fire Strike" have an option to detect if the tesselation settings have been modified from the driver level during the runs/before we run it? (something like : preventing the 3dmark to actually run if it detected that tesselation level has been changed) Thanks Man I can't say yet what we're doing about this but I guess I can say that yes, this issue is known and 3DMark will work differently from 3DMark 11. (and for those who have already seen some early build through their contacts with BDP members (mind the NDAs!), what I'm talking about is not present yet in the build you may have seen) Quote
FM_Jarnis Posted December 3, 2012 Author Posted December 3, 2012 First teaser on the Feature Level 11 demo (Fire Strike) is now up on Youtube The original res is 2560x1440 and this is from Fire Strike Extreme. Note that this scene is harsh and you probably need quad-GPUs to get anywhere near the framerate of this video when running as Extreme Related PR blurb here; http://www.futuremark.com/pressreleases/new-3dmark-fire-strike-trailer ...and yes, this means launch is getting close. No date yet, but I guess instead of Soon, it would be Very Soon. Quote
Christian Ney Posted December 3, 2012 Posted December 3, 2012 Thanks Janis, played the video in Original Quality made my computer crash, instant BSOD Quote
FM_Jarnis Posted December 3, 2012 Author Posted December 3, 2012 Thanks Janis, played the video in Original Quality made my computer crash, instant BSOD 3DMark - so extreme that puny computers can't even play the video without failing. (we also saw here on couple of crappy office computers that the 2560x1440 YouTube playback isn't working smoothly) Quote
Christian Ney Posted December 3, 2012 Posted December 3, 2012 Ain't that crappy pc though: i7 990X + HD 7970 and 2560x1440 monitor ... Quote
mavihs Posted December 3, 2012 Posted December 3, 2012 Ain't that crappy pc though: i7 990X + HD 7970 and 2560x1440 monitor ... that is punny!!! the minimum requirement to play that video are: an (AMD Opteron 16 core @ 4GHz processor)*2 with a 7970 in a Quad Crossfire & 192GB RAM !!! :banana::banana: then you'll be able to play without any strutter or lag!!! plus a 1Gbps net connection if your streaming it!!! :woot: Quote
Eeky NoX Posted December 3, 2012 Posted December 3, 2012 Yeah! Nice Some real action scene now, finally!!! Quote
ƒ0rteOC Posted December 11, 2012 Posted December 11, 2012 Yeah! Nice Some real action scene now, finally!!! Agreed! Better than trolls playing 2x2 checkers. Quote
FM_Jarnis Posted January 31, 2013 Author Posted January 31, 2013 FYI: Monday, February 4th. 18:00 UTC Better stock up on needed liquids for some benching (both for cooling use and for personal use). Quote
Recommended Posts
Join the conversation
You can post now and register later. If you have an account, sign in now to post with your account.