3DMark 11 out on 30th November

[FM_Jarnis]

yes i have my pre-order for like a month

thanks Jarnis, i think i'll send an email, my junk folder gets deleted automatically, but all the other email went through in my inbox

Send an email to pasi [at] futuremark.com and he'll dig up the key for you. Include details of your purchase.

[FM_Jarnis]

Any news about release date?
1-2 days or 1-2 weeks?

Officially: No
My educated guess: Most likely early next week (with the note that Monday is Independence Day in Finland, so it can't be Monday), pending on a bunch of things - nobody knows for sure yet and it won't be announced until it is sure.

[dinos22]

That won't happen - the combined 3DMark score is given only when the benchmark is completed using a specific preset - it is the only way to ensure that the score is comparable. Older benchmarks gave a number that was effectively meaningless unless you confirmed it with a qualifier "with default settings".

You do get GPU and CPU score. For game tests, the GPU score can be used for comparing different tweaks and individual tests (or CPU score for 5th test and both for 6th).

who says it doesnt have to use a specific preset but without needing to run a complete run. surely the program can be designed to detect change of preset if a bencher decided to get smart............

anyways since you guys are interested in overclockers all of a sudden and what they want perhaps you should start entertaining this idea again because i am confident you will see an incredible show of hands in you put up a poll here asking if we'd be interested in a bench like that.

[FM_Jarnis]

who says it doesnt have to use a specific preset but without needing to run a complete run. surely the program can be designed to detect change of preset if a bencher decided to get smart............

One thing that has been corrected is the length of the run.

In 3DMark 11 you have two demo parts, four graphics test (GPU benchmark) parts, one CPU test and one CPU-GPU combined test. You can choose to run the whole thing (default), just the demo or just the score-producing parts and the score-producing bits are fairly short - we know people may end up running the benchmark 20 or 30 times over the course of the day and the goal was to keep the actual score-producing part short and to-the-point. Note that even the free Basic version offers the choice between "whole thing", "demo only" and "benchmark only".

Unfortunately the 3DMark score itself is calculated from all six tests, so you have to run them all to produce a 3DMark score (and you have to use one of the default presets). However, you can run any demo or test part independently (advanced version) - just without a composite 3DMark score since there isn't required data to calculate it.

And yes, there are a ton of things you can tweak to drill down to differences between hardware setups with individual tests and custom settings. There just isn't a composite 3DMark Score in such cases because, again, it requires data from all six tests and default preset settings.

[Manicdan]

is there a way to do a batch of tests so that i can have it run the 4 benchmarks, but the first time do no AA/AF, then the second time a little AF, then a little AA then both, etc...

sounds like that would be with the super expensive version?

[Borden]

Any news about release date?
1-2 days or 1-2 weeks?

I believe it will be released after December 7th.

[FM_Jarnis]

is there a way to do a batch of tests so that i can have it run the 4 benchmarks, but the first time do no AA/AF, then the second time a little AF, then a little AA then both, etc...

sounds like that would be with the super expensive version?

That would be "Command Line Automation", allowing you to create batch files that run the benchmark from the command line at various settings. And yes, it is part of the Professional Edition ($995).

Advanced gives you all the options but you have to run the benches manually from GUI.

[Manicdan]

That would be "Command Line Automation", allowing you to create batch files that run the benchmark from the command line at various settings. And yes, it is part of the Professional Edition ($995).

Advanced gives you all the options but you have to run the benches manually from GUI.

good to know

i think for a future release it would be nice to have a step between advanced and pro where it will run a prebuilt list of settings and output a very nice datasheet so a reviewer can quickly see the AA, AF, tessellation, LoD, etc, performance hits and determine which ones are practically free and which are bottle necks. the time it takes to test each setting and then organize and compare the data is quite exhausting, but required for optimal gameplay settings. sure each game acts differently, but it would be great to see in a review the full comparison of the architecture for a starting point of how you should set up the gpu in games.

[FM_Jarnis]

good to know

i think for a future release it would be nice to have a step between advanced and pro where it will run a prebuilt list of settings and output a very nice datasheet so a reviewer can quickly see the AA, AF, tessellation, LoD, etc, performance hits and determine which ones are practically free and which are bottle necks. the time it takes to test each setting and then organize and compare the data is quite exhausting, but required for optimal gameplay settings. sure each game acts differently, but it would be great to see in a review the full comparison of the architecture for a starting point of how you should set up the gpu in games.

This kind of features are already there for the reviewers - they are expected to be using the Professional Edition. Note that FM has repeatedly suggested members of the (hardware) press to contact us.

Quoting from FM.com

"On request, we freely furnish versions of our products and offer both support and technical review assistance to qualified media organizations and professionals. Please contact press@futuremark.com for more information. "

[T_M]

One thing that has been corrected is the length of the run.

In 3DMark 11 you have two demo parts, four graphics test (GPU benchmark) parts, one CPU test and one CPU-GPU combined test. You can choose to run the whole thing (default), just the demo or just the score-producing parts and the score-producing bits are fairly short - we know people may end up running the benchmark 20 or 30 times over the course of the day and the goal was to keep the actual score-producing part short and to-the-point. Note that even the free Basic version offers the choice between "whole thing", "demo only" and "benchmark only".

Unfortunately the 3DMark score itself is calculated from all six tests, so you have to run them all to produce a 3DMark score (and you have to use one of the default presets). However, you can run any demo or test part independently (advanced version) - just without a composite 3DMark score since there isn't required data to calculate it.

And yes, there are a ton of things you can tweak to drill down to differences between hardware setups with individual tests and custom settings. There just isn't a composite 3DMark Score in such cases because, again, it requires data from all six tests and default preset settings.

After all of that you still managed to avoid the point that overclockers want to be able to run individual tests, and when all are finished being run you can get the score. YOu can program complex 3D software, but not do something simple like that?

[Loque]

will it detect the correct cpu speed this time ? :p

[FM_Jarnis]

After all of that you still managed to avoid the point that overclockers want to be able to run individual tests, and when all are finished being run you can get the score. YOu can program complex 3D software, but not do something simple like that?

I'm not sure what you mean.

In order to get a 3DMark score, all tests must be run with one of the predefined presets. This is done to ensure that the scores produced are comparable (apples-to-apples) to any other run with the same preset.

If you could run each individual test separately and then "glue them up" for a 3DMark score, that would obviously change the test parameters - the system doesn't have to be stable through the whole benchmark - just an individual test. Useful? I would imagine so. Fair for comparison vs. a system that had to stay stable throughout the test? No.

So yes, the requirement that all tests are run, in sequence, in order is by design to get results that are useful for comparing across different hardware.

If all you want to do is to compare the results of individual tests, you can do that as well (with sub-scores) - just not with a 3DMark score.

[dinos22]

you are not getting it. this is what makes 3DMARK01 special and it also a new aspect to tweaking, working out the run order etc also

[FM_Jarnis]

will it detect the correct cpu speed this time ? :p

As of latest SystemInfo (3.51), that problem should be sorted. The issue was simply due to all this new-fangled turbo boost stuff truly blurring the whole issue. I mean, without constant software monitoring, do you know what is your CPU frequency? I know mine is "it depends".

There are still issues with graphics card core & memory clocks but we're working on that.

[dinos22]

we do because we lock the speed in and disable turbo to have better control over the OC

[FM_Jarnis]

you are not getting it. this is what makes 3DMARK01 special and it also a new aspect to tweaking, working out the run order etc also

I think I am.

What you are describing is a way to game the system with 3DMark 2001 as things like this were not taken into account during the development.

It is not possible to game that way on our modern benchmarks. You may disagree with that decision, but we think it is important to do it like this to get results that are useful for something other than just going for a record score.

It ultimately boils down to the question - do you want benchmark to be;

- A program that spits out a number that depends on your hardware and on how you tweak and poke the settings to the maximum in order game the number to be bigger.

OR

- A program that spits out a number that gives an indication of the performance of your hardware. Period. Want a bigger number? Tweak the hardware.

[massman]

I think I am.

What you are describing is a way to game the system with 3DMark 2001 as things like this were not taken into account during the development.

It is not possible to game that way on our modern benchmarks. You may disagree with that decision, but we think it is important to do it like this to get results that are useful for something other than just going for a record score.

It ultimately boils down to the question - do you want benchmark to be;

- A program that spits out a number that depends on your hardware and on how you tweak and poke the settings to the maximum in order game the number to be bigger.

OR

- A program that spits out a number that gives an indication of the performance of your hardware. Period. Want a bigger number? Tweak the hardware.

You could just add a Freestyle preset that allows per-stage scores to be saved (~ 01). Freestyle has the same default settings as another preset (performance), but allows 'gaming' the system like 01.

[NaMcO]

- A program that spits out a number that gives an indication of the performance of your hardware. Period. Want a bigger number? Tweak the hardware.

Anyone, even i (a noob regarding benchmarks), would want it to be like this. Right? Right.

[dinos22]

You could just add a Freestyle preset that allows per-stage scores to be saved (~ 01). Freestyle has the same default settings as another preset (performance), but allows 'gaming' the system like 01.

this is a good compromise because extreme OC guys will all want to run separate tests on their own and work out how to run it best

good suggestion

[MadDias]

this is a good compromise because extreme OC guys will all want to run separate tests on their own and work out how to run it best

good suggestion

afaik you can run each test individually in the advanced version. you even get a score for the test, let it be gpu score or cpu score. you just don´t get a validated 3dmark score that u can publish in the orb.
you can run each test individually and see what effects ur tweaks have taken. but to get an overall 3dmark score u have to run the full benchmark without changes during the test. changing hardware settings during tests is considered cheating i guess.

[dinos22]

^ lol ok man

[massman]

you can run each test individually and see what effects ur tweaks have taken. but to get an overall 3dmark score u have to run the full benchmark without changes during the test. changing hardware settings during tests is considered cheating i guess.

Manufacturers have been releasing cheat-tools for quite some time now.

[flopper]

the overclockers need to write thier own program they can compete with.
instead of complaining and whining.

go ahead, there is enough of people running ln2 and people writing code, write your own benchmark in your own way.

[Katzenschleuder]

I think I am.

What you are describing is a way to game the system with 3DMark 2001 as things like this were not taken into account during the development.

It is not possible to game that way on our modern benchmarks. You may disagree with that decision, but we think it is important to do it like this to get results that are useful for something other than just going for a record score.

It ultimately boils down to the question - do you want benchmark to be;

- A program that spits out a number that depends on your hardware and on how you tweak and poke the settings to the maximum in order game the number to be bigger.

OR

- A program that spits out a number that gives an indication of the performance of your hardware. Period. Want a bigger number? Tweak the hardware.

I agree with your intention. The irony however is that you do not see that nearly all people "using" 3DMark don't care in the least about the relevance of what they are measuring, simply because that is what you have delivered for so far.

The reality is that you could play back a fancy looking pre-rendered movie for ten minutes and just run very trivial fillrate/shader tests in the background for some pseudo performance measurement. You would deliver exactly the same quality of service with a fraction of effort.

There are people who have the noble idea to destroy hardware which is worth thousands of Euro just to run an absolutely useless calculation a few miliseconds faster. And your product attracts people who support this mentality.


PS.: Yes people, I know that you despise me for having the brassiness to citisize your glorious hobby.

[saaya]

hey jarnis, you guys should add a line to the contracts with big prtners that they can change the launch dates BUT have to announce it a certain time before the launch, to prevent last minute delays like this... cause announcing a launch and then delaying it last minute makes you look unprofessional imo... no offense, im just saying...
ati and nvidia etc need to know more than a month ahead whether the next 4 weeks are enough for them to tweak drivers or not, and if they first think it is and then its not, well too bad for them...

and yes, bringing back the invidiual test run like in 2k1 would be awesome!