PerformanceTest 8 when?

[rawr]

It's been over 2.5 years since the latest version of the program was released, and I noticed that anyone who purchased prior versions of the program more than 6 months before the release of V7 could purchase V7 at a $10 discount, with those in the 6 month gap able to request a new key for free.

V5 - 12/10/2003
V6 - 09/04/2005
V7 - 02/24/2009

~1.75 years before new version, ~3.5 years to the next. By that pattern you would assume it would be over 6 years to the next, but I was curious to know if V8 is on the horizon or if V7 is just going to be slowly updated to test DirectX 11 and add new features, and if I should buy a key now which will be able to be used for a few years or wait until V8 if it will be released soon.

[David (PassMark)]

Even though we haven't done a release with a major increment in the version number, we have be maintaining and upgrading V7. See,
http://www.passmark.com/products/pt_history.htm

The problem with frequent major releases is that we break comparability. Meaning that if we introduce new tests, new compiler versions, etc.. it is not longer possible to compare results across releases. This invalidates the millions of baseline results that people have collected.

Having said that we have started work on V8. We are about 50% done. My guess at the moment is a release ~5 - 9 months from now depending on what goes in. Might be an alpha / beta release in 4 - 6 months.

And yes, there will be a DirectX 11 test in PT8.

Update: Here is a link to the PT8 development thread.

[Chrysalis]

I would like to see this also as I think 7 is looking outdated now. Especially on the 3d tests.

[David (PassMark)]

We are hard at work on PT8.

This week we are looking at improvements in the Integer maths test and prime number test in the CPU Suite.

The 3D tests are never going to look as good as a current AAA game. The reason for this is that A) It is expensive to develop good looking games (or benchmarks that look like a game) and B) We want to keep the benchmark small. The last thing we want is a benchmark that is 500MB in size and takes 30min to download.

[Seronx]

We are hard at work on PT8.

Will you be using FMA?

[David (PassMark)]

I assume you are talking about the fused SIMD multiply–add CPU instruction from AMD?

PerformanceTest already uses SIMD instructions. We try and restrict low level instruction usage to just the (fairly large) set supported by both Intel and AMD. It would seem unfair to use instructions that are just supported by one manufacturer, especially when those instructions aren't used in any real software.

[Seronx]

PerformanceTest already uses SIMD instructions. We try and restrict low level instruction usage to just the (fairly large) set supported by both Intel and AMD. It would seem unfair to use instructions that are just supported by one manufacturer, especially when those instructions aren't used in any real software.

I don't get your point

It would seem absolutely fair to use instructions that are just supported by one manufacturer, especially when those instructions can speed up processing by 3x(up to 70x if the code is really ugly).

Sandy Bridge and Bulldozer now support AVX so atleast get AVX down FMA3/FMA4 will be for 2013 when Haswell gets FMA3

http://software.intel.com/sites/prod...bk_avx_fma.htm

Here is the intrinsics for FMA for Intel

[David (PassMark)]

If those instructions were in wide use in real world software applications, it would then make sense for a benchmark to reflect real world usage. But they aren't, so it doesn't.

We won't be using the Intel’s Quick Sync transcode function from Sandy Bridge either. For the same reason.

Might be a place for them in the advanced tests, but not in the standard CPU test which needs to reflect maintream usage of a CPU (not exotic usage). By 2014 the instructions might be in more common use, but not in 2011.

[Seronx]

If those instructions were in wide use in real world software applications, it would then make sense for a benchmark to reflect real world usage. But they aren't, so it doesn't.

What would be considered a real world software application?

AVX 128bit in some applications replacing SSE 128bit

Might be a place for them in the advanced tests, but not in the standard CPU test which needs to reflect maintream usage of a CPU (not exotic usage). By 2014 the instructions might be in more common use, but not in 2011.

That will work advanced test options are pretty cool wish I could control the thread amount in "Multitasking"

[David (PassMark)]

There is no hard definition for 'real world software applications'. But I think it would be reasonable to say that if certain CPU functions are not used by 99% of available applications, then that function isn't a good candidate for a general usage CPU benchmark.