FM_Jarnis
-
Posts
188 -
Joined
-
Last visited
Content Type
Profiles
Forums
Events
Blogs
Posts posted by FM_Jarnis
-
-
Assuming we want to disallow altering the Tesselation level, there are still the following problems:
- we can't see it in the benchmark GUI (so screenshots will be meaningless)
If this is, in your opinion, a feature that should be there, we can look into adding it. It won't be instant but we plan to release a patch to 3DMark 11 this year.
- we can't see it in the ORB submission page (at least not for a few weeks)
You will see it there before the drivers are approved. The whole point is that, from FM's point of view, only the result page on 3dmark.com is the "official" result and whatever is shown in benchmark UI is a preliminary score that still needs to be validated by the online service. You have to understand that there are several known ways to cheat with the score already that would alter the score in the UI but would be rejected on 3dmark.com as the benchmark does no analysis to the score data or the data gathered by SI - it just packages them, displays the score and then prepares for upload to 3dmark.com.
- FM's implementation of the Tesselation detection in the benchmark GUI is to remove E/P/X score indications or something similar that makes it impossible for us to verify the score
No. If we add something to the benchmark UI, it would be a separate additional note as to the tessellation settings used for the run. Most likely it would say "AMD default tessellation settings" vs. "x x x settings" depending if they were changed or not (and the whole line would not be there on NVIDIA cards, at least until they add something like this to their drivers). The actual score with E/P/X would still be same.
-
Well, speaking about tessellation, you made the point clear... if nvidia isn't going to release a driver version in which you can control the tessellation, then this feature must be ON for both of them.
Even if they would, we would probably do the same thing. What is the point of benchmarking DX11 cards for doing major DX11 features (Tessellation) if drivers can silently remove a major feature workload ("disable tessellation"). The score from such a run would be pointless for figuring out the DX11 performance of the hardware.
Unless you plan on playing every DX11 game with Tessellation Off as well?
-
Ok, so no tessellation, no lods.
What is overclocking a vga about, then? Pure frequency?
Oh, interesting, so the skills will be the amount of binning needed to find the right vga, the right cpu etc...
I strongly disagree. I can understand the decision behind prohibiting the mipmap tweaking: it was not a "tweak" it was a blank screen from the beginning to the end of the bench.
But the lods... don't change the rules even on that please -.-
I'm not saying we are doing anything with LOD tweaking yet. I'm just saying we are open to the idea if OC community thinks that detecting it would be useful.
Tessellation thing has to be done - without it, AMD runs could be done with major benchmark components completely disabled and with completely different looking image. That just isn't right and that is why no AMD driver since the addition of tessellation controls have been approved.
Our goal is always that all hardware run the exact same work, preferably rendering exactly identical images and then the benchmark tells you which piece of hardware is best at doing it. Changing the rules by changing the workload makes any comparisons flawed.
-
Oh, interesting. Did not know about OB's connection with Nvidia!
I don't think there's faul play here, though. If FM was biased like that, they would never have disallowed GPU physics.
Trust me, there is no bias. Both AMD and NVIDIA hate us equally
Edit: Well, one week one of them hates us more, the next week its the other. Depends a bit on what piece of hardware they are shipping at the time and how it performs. Meanwhile FM just measures the performance and reports the gory details, always striving for unbiased benchmarks.
-
That's what's being discussed here: http://hwbot.org/forum/showthread.php?t=17844
Tbh, since we can't spot the difference in the benchmark GUI, I'm not too eager to do anything. Actually, we should just figure out how the disable Tesselation on Nvidia cards and the playing field is even again
If this is something that you would want, it can be considered as something that could be added in the next patch for 3DMark 11. We are planning on releasing another patch for 3DMark 11 this year (since we have to bundle 4.2 SI with the benchmark, might as well do some other fixes).
So whatever SI reads from AMD settings could possibly be displayed in the benchmark UI. Would this be useful?
(In our view, the online service page is the "real" result, validated - whatever is shown on the benchmark UI could be fake/photoshop/whatever anyway, so it isn't something we thought would be a big deal)
-
Well... tessellation is low ( AMD optimized ) by default so what exactly is legit?
Are hwbot going along with this and removing all old dx11 amd records?
AMD Optimized, according to our tests, does application detection before changing anything. With 3DMark 11, "AMD Optimized" does not change the score or the workload in any way. AMD Optimized reduces tessellation only in specific games AMD thinks are "unfair" (and Green Team sponsored), and does it in ways that AMD thinks doesn't reduce image quality, just improve performance. If this mode would detect 3DMark 11 and reduce tessellation, it would immediately run afoul of the driver approval policy which says that drivers cannot specifically detect 3dmark executable and change anything based on that (with the exception of SLI/Crossfire mode selection ONLY) and AMD drivers would all be binned as "no good" until they'd change it.
But the situation is that, as run with default settings, AMD drivers produce fine scores. The only issue is that until SI 4.2, we could not tell if those driver settings were used to alter the workload or not, so the only solution was to reject all drivers until this could be detected.
The goal here is to ensure that the benchmark runs exact same workload on Green and Other Green team hardware.
LOD tweaking is different in that, as I understand it, it is possible on both AMD and Nvidia hardware. However, once this is in place, I guess adding a check for that too would not be too complex (would you want it to be there?)
-
IMPORTANT
Straight from FM:
Practically, that means more problems for us. With 3DMark Vantage's Physics, we could easily see if it was enabled or not by comparing the CPU scores. With the Tesselation level, this is much more difficult since there is scaling possible. Unless FM will show what level is used in the submission page, we won't be able to be 100% sure.
This also means that for official FM HOF results, you can now use AMD graphics as long as you don't finetune the tesselation level. It's now also possible to use the cards for competitions at Futuremark (which previously wasn't possible due to non-FM approved Catalyst drivers).
Let's look into this more closely.
Confirming this. SystemInfo 4.2 can tell if you are "cheating" with tessellation controls in 3DMark 11.
We are still working on the 3dmark.com result page side of this - it may be possible to include the details on the settings used on the result page (SystemInfo 4.2 reads and saves the settings that AMD drivers use and they are included in submits as of, well, now). Feedback on that is more than welcome as to what exactly you would want the result page to show. All that is certain at this point is that it will show that the result is not valid (similar to GPU PhysX + Vantage) if tessellation loads have been reduced with driver settings.
I don't have a date yet for when 3dmark.com results page will be updated and when AMD driver approval is completed but it is coming soon ("a few weeks" is a good educated guess). At the moment we are at the stage where we are collecting data and testing the backend code so we can reliably say which results conform to the driver approval policy and which don't.
Sorry that this has taken us so long to implement but it was a fairly large addition to SystemInfo capabilities and it kinda dropped on top of all other scheduled work (annoying the makers of those schedules and plans greatly). The wheels of software development sometimes turn slowly, but they do turn...
To emphasise: You cannot (yet) get to HoF / get approved driver results with AMD drivers later than 11.1 or thereabouts, but any results that you submit after installing 4.2 SystemInfo with the current AMD drivers should turn into valid, approved results when the work related to this is complete and pushed live (assumption: you do not touch the tessellation controls)
-
Has system Info 4.2 been included in the main Vantage download? Or... to re-phrase... if we download Vantage today, do we get the latest build with the latest system info already included?
Not yet.
It is a much larger operation to (re)build benchmark installers, including the relevant QA.
We will eventually release repackaged installers of all supported benchmarks but it is most likely that those will wait until a few fixes are ready & tested for the benchmarks themselves. Probable timeframe 1-3 months.
Until then, you have to download the installer and then separately install the latest SI. Or the other way around. The older SI that is in the benchmark installer won't overwrite 4.2 if you already have it installed.
Another way to install SI 4.2 is to just run Game-o-Meter once - it will update SystemInfo to 4.2 if it is not present or up-to-date.
-
Does this work with W7 yet? Last time I installed 1.10 I just got crash after crash and had to revert to 1.02
AFAIK that was caused by SystemInfo 4.0 which had some issues. 4.2 should fix those.
http://www.3dmark.com/support/systeminfo-updates/
Download latest Vantage, install SI 4.2 update and it should work.
-
SystemInfo changelog (and latest standalone update) is always here:
http://www.3dmark.com/support/systeminfo-updates/
(4.2 just went live, we'll post the changelog for it later today)
-
Scores from before patch and after patch comparable(ofcourse if Physx was disabled)???
Yes. There may be some very small variance due to the fact that the new build is recompiled with newer complier and libraries, but it was deemed to be within margin of error and there was no reason to invalidate the vast amount of existing scores. The difference should be around 1-2%.
-
3DMark Vantage version 1.1.0
- Trial Edition discontinued
- Basic Edition now free, was previously $6.95
- Added display scaling to all presets: benchmark now scales the output to best fit the display
- GPU physics acceleration is now disabled by default during the CPU tests
- Systeminfo component updated to version 4.0
Mirrors are getting files online over the next few hours. Official pages won't update until tomorrow, so consider this as an early heads-up. There will also be a discount for existing Basic Edition owners to upgrade to Advanced for $5.
-
Just an early heads up, the work is complete and 3DMark03 and 05 benchmarks are again fully supported by 3dmark.com (new submits)
Old results are in progress of being transferred and converted. This is expected to take until sometime next week. Don't panic if yours is not there yet - they will appear.
Related to this, there is also a process of converting old 06/Vantage/11 scores to a new (internal) format - existing old scores are still viewable, but may lack some detailed information until the process is complete. Again, this is expected to take until sometime next week (millions and millions of benchmark results take some time to process).
Let us know if you have any issues with 03 or 05. We'll make an official announcement of this next week but I thought you guys would like to know as soon as the code was live.
-
Is there a switch to disable this like we had in other FM 3DMarks?
Not that I know of. I think it wasn't seen as needed because no matter what the 3DMark 2001SE detection code gets, the benchmark runs ignoring all that completely. It would only affect submits (no longer supported) and viewing of detailed results.
Patch probability = 0%. Sorry.
-
My guess: "too new" hardware totally confusing the ancient hardware detection routines of 3DMark 2001...
-
Further SystemInfo stand-alone update is available
http://www.futuremark.com/download/files/Futuremark_SystemInfo_v36111_installer.exe
This will be rolled into benchmark patches soon but we made it available as stand-alone hotfix. Should fix any remaining hardware detection issues, including the infamous "broken water" in 3DMark Vantage with SLI.
-
Just FYI, 3DMark 11 patch v1.01 is now available.
http://www.futuremark.com/download/3dmark11/update/3DMark_11_v100_to_v101_updater.exe
3DMark 11 version 1.01 release notes
All Editions
* Fixed an issue with SystemInfo that prevented the Physics and Combined tests from running.
* Fixed an issue that could cause a crash if the DirectX 11 API or .NET was not available.
* Fixed an issue that could cause a crash if the number of system GPUs was unknown.
* Letterboxing and scaling now works with wide displays.
* The benchmark UI is now minimized during the benchmark run.
* Improved result security when submitting to 3dmark.com.
* Added automatic update notification.
* Updated SystemInfo component to version 3.54.1.1.
Advanced and Professional Edition only
* Added option for manual GPU Count selection to the Advanced tab.
* Added Scaling Mode selection to the Advanced tab, choose from centred or stretched.
* Results tab now shows the individual scores for each of the six tests. <- as requested by HWBot community and others
Professional Edition only
* Demo audio now works when running from the command line.
* Image Quality tool now handles unexpected parameters correctly.
* Image Quality tool now works correctly with the reference rasterizer option.
-
i think its fair enough that we as a community pay futuremark if the bench in question meets the requirements we need it to meet,
some of the key to me being,
* flexing up to date hardware, dx11 - TICK
* heavy on GPU when it first comes out - TICK
* good level of security built in - this bench is a step backwards from 03,05,06,vantage which is a step backwards again from 3d01
* subtests visually available - we really need to see some GT and CPU/combined FPS without going into the results.xml and digging them out..
I think futuremark have delivered the hard parts of the deal, but left a couple of the minor parts out..
We'll fix the minor parts. Give us a bit to sort it all out...
-
I think what gets me is that even if you pay for the software, they are forcing you to go to their website for revenue reasons. I can understand the free version, but give us a break if we have already paid for it.
As it stands, I say no to boints for 3d11
We'll fix the sub-score display offline on advanced edition. Just wait for the patch.
-
This announcement was just sent to everyone who preordered:
---
It took us a little longer than planned, but 3DMark 11 has now gone gold. Earlier today we announced that 3DMark 11 will be released at 14:00 GMT on Tuesday December 7, 2010.
From that time onwards you will be able to download 3DMark 11 from the Futuremark website and start submitting results to the new online service. You will need the upgrade code from your pre-order confirmation email to activate the Advanced Edition so make sure you have it to hand on Tuesday.
Till then, happy benchmarking!
-
I assume that the basic edition will launch with it then?
All versions launch at the same time - they are the same executable.
-
Over the weekend we made the difficult decision to postpone the launch of 3DMark 11 by a few days.
Our aim is that 3DMark 11 should provide accurate, reliable and consistent results from the start. With that goal in mind we are taking some extra time now to fix a couple of difficult bugs rather than patch the benchmark immediately after launch.
It might take a day, it might be a week, but either way we'll confirm the new launch date as soon as possible. Thank you for your patience and understanding.
-
Issues with card clocks are related to SystemInfo, not the benchmarks themselves or the ORB. I cant recall if the latest build improves on those or just fixes compatibility issues but if you want to try it, currently the way to install it is to run Game-o-Meter once (it will install the new 3.51 systeminfo) and then running the benchmark. I know it is bit of a kludge-workaround to get the update, but that was the quickest way we could do it.
Benchmark-specific SystemInfo patches will also come soon but they are unfortunately delayed due to 3DMark 11 launch taking up the developer resources. 3DMark 11 will also ship with the new SystemInfo so once you install that, it also gets updated (affecting also older benchmarks as far as the hardware detection goes).
-
So PCMark11 is coming as well
Ooops.
<jedi mind trick>You never saw that. </jedi mind trick>
3DMark Vantage v1.10 patch
in Benchmark software
Posted
It can be done in the registry. The control panel has some limitations but if you just change the relevant keys in the registry, poof, whole tessellation load gone.