New Instruction Set for C2D increasing performance by another 15%?

[Eastcoasthandle]

Two or three things emerged from the event that are worth reporting. The most interesting was told to me by one of the strategists at the company. These new chips have new instructions. This hasn’t been widely reported but the instructions will be rolled out probably in the next month under what I suspect will be a copyrighted moniker, “MEDIA BOOST.” And yes they are aimed at the consumer electronics side of things. I’m told that as fast as these new chips work with codecs, using the new instructions can increase the effectiveness by at least another 15-percent. I’m sure these new instructions will be documented by the hackers sooner than later.

Link (http://www.dvorak.org/blog/?p=6336)
If my understand is clear so far, before C2D it was called TNI, is it now called something else? And it will add 15% more performance then what we have seen so far?

[charlie]

mmmmmmm....

[Mr.X]

So, does this mean we should wait a couple of months to buy C2D so that we can take advantage of this?

[Tranzmit]

TNI is Tejas New Instructions which was going to come out with Tejas but as Tejas was killed Intel moved them into Conroe. Whereupon they were officially named given the more marketable name SSE4.

Same with SSE3 which were internally called Prescott New Instructions.

http://www.intel.com/products/processor/coreduo/digitalmediaboost.htm
http://www.cs.umd.edu/class/fall2001/cmsc411/projects/SIMDproj/project.html

Basically all these are are ways of getting more done for less work so the idea is SIMD or Single Instruction, Multiple Data so say getting a processor to do video encoding is something where they can save clock cycles (which is the nsame of the game for designers) by getting the chip to do more work per clock cycle

[Mikesnav]

Eh.

If these require a core revision (one where new instructions/engines are needed), wouldn't it show up on Intel's roadmaps?

[Ibaun]

Guess that by saying "instructions will be rolled out" they mean that they will release the design docs for C2D with the new bytecodes, and not some core revision.

[Tranzmit]

No no I'm guessing thats an old article. All Conroes have SSE4/TNI

Go have a look at some of the screenshots of people with Conroes who've been overclocking them and you'll see the Instruction sets supported in the CPUID/CPUZ screens or Crystal as well.

Basically intels big hooah about being able to do the 128bit SSE instructions in a single cycle IS basically it's MediaBoost marketing name anyways.

Reread my earlier response as i edited it with some links

[TyroPyro]

... Overclocking ... They will not condone it but will have zero roadblocks or “locks” to prevent or hamper the process.

But the retail multis are still locked as far as anyone has seen, right? What other roadblocks would they mean by this

[jbizzler]

"Media Boost". Maybe it'll just be an update in the OS? I don't know much about this, but are these the same kind of instrcustion that you refer to when talking about x86 and PPC and stuff?

[Eastcoasthandle]

Does anyone with a X6800 (for example) have the retail box with instructions thtat discuss Media Boost? I suppose this is the real question...

[horrorwood]

Does anything use SSE3?

Nevermind SSE4...

[Eastcoasthandle]

Game devs don't use SSE3 often enough if you ask me. SSE3 does offer a boost for Netburst P4's through XP compiler. It's been proven if a game (like BF2) was coded using SSE3 for P4 CPUs; it would see a huge performance gain. However, it wouldn't work for AMD processors (except Athlon and later AMD CPUs that use SSE3). It would all boiled down to 2 different BF2.exe executable files that the devs didn't want to make (one for Intel the other for AMD). It was a heated argument that was discussed in many forums.

[freecableguy]

Conroe has SSE4 and it is kicking (128-bit data decode, ie. "Media Boost") Check SANDRA Multimedia scores to see this in action...

-FCG

[Eastcoasthandle]

Correction it was ExtremeTech (http://www.extremetech.com/article2/0,1558,1895906,00.asp). Using the BF2 as an example here is why AMD out performs P4

# It was compiled using Visual Studio 2003 C++ compiler.
# It was compiled in blended mode almost without any optimizations..

It has been argued that this blended mode is the the mode used most in PC games with the exception to Q4, etc. This mode is optimized for earlier CPUs which can become a bottleneck for more advanced CPUs like P4 and Athlon. However, P4 benefits more from /G7, SSE3 method then Athlon from what I read.

Blend mode (http://msdn.microsoft.com/library/default.asp?url=/library/en-us/dv_vstechart/html/vctchoptimizingyourcodewithvisualc.asp):
When no /Gx option is specified, the compiler defaults to /GB, "blended" optimization mode. In both the 2002 and 2003 releases of Visual C++ .NET, /GB is equivalent to /G6, which is said to optimize code for the Intel Pentium Pro, Pentium II, and Pentium III.

With this new TNI instruction set it looks like we will need MS to come up with a new complier for game developers to take advantage of or else as it stands now only good for certain applications and benchmarking. But this also promotes the same problem with SSE3. If the game developers don't implement the new complier then you will never see the benefit in games (continued use of the "blend" method). But from the looks of this...it doesn't appear to be something for games to begin with.

But I still wonder if this new compiler (if it actually exists on this plain) will be call /G8?

[iterations]

Yeah SSE4 is activated in all Conroe chips. It is just a matter of using the appropriate bytecodes in software. As FCG said above, you can see this in Sandra Benches (INT and FP multimedia scores). As codec software writers become aware of these new SSE instructions you can expect an improvement in encoding and compression performance.

So for example, if you do alot of h264 encoding, a simple software update will get you an additional 15% boost in encoding speed. You don't need a new chip, just new software that uses the new instructions, which up until very soon, Intel has been very hush hush.

It will probably be a nice marketing move (given their VIIV stuff and all) to say hey, not only did we reengineer our SSE engine so it completes all SSE operations with a 1 op-per-cycle throughput, but we have ALSO added 8 new instructions that when used appropriately, can add ANOTHER 15% boost in performance.

[Eastcoasthandle]

Yeah SSE4 is activated in all Conroe chips. It is just a matter of using the appropriate bytecodes in software. As FCG said above, you can see this in Sandra Benches (INT and FP multimedia scores). As codec software writers become aware of these new SSE instructions you can expect an improvement in encoding and compression performance.

So for example, if you do alot of h264 encoding, a simple software update will get you an additional 15% boost in encoding speed. You don't need a new chip, just new software that uses the new instructions, which up until very soon, Intel has been very hush hush.

It will probably be a nice marketing move (given their VIIV stuff and all) to say hey, not only did we reengineer our SSE engine so it completes all SSE operations with a 1 op-per-cycle throughput, but we have ALSO added 8 new instructions that when used appropriately, can add ANOTHER 15% boost in performance.

Agreed, this looks like something we may see in Vista (I can only speculate it will be in Vienna). Although I do hope to see it in XP. This will only add futher boost to Vista 64 bit operating system truly blowing AM2 away by a long shot. However, this is all talk if developers don't use it. But developers can't use it if MS doesn't incorporate a new compiler for it. Man, if this compiler was added to all 64 bit applications alone the gains from P4 to C2D would be a wide as we see it now in 32 bit applications.:eek: :banana:

[Eastcoasthandle]

But the retail multis are still locked as far as anyone has seen, right? What other roadblocks would they mean by this
Upping the multi, not sure really

[andyisc00l]

Sorry I don't understand, there will be new revisions of the chip or is this something that can be downloaded:

[iterations]

It is there in all of the chips, even the ones you buy today. It is just something software developers can take advantage of in the future, no further core revisions neccessary for this feature.

[Eastcoasthandle]

Correct me if I am wrong, but isn't it Intel's compilers that make the x86 world go round (for the most part)? I was under the impression that as Intel releases new compilers (to support new options built into their processors) that microsoft licenses many of those compiler options/pieces (if not the whole compiler) to 1) use in their compilers 2) compile their software. Biggest issue seems that if we wanted Vista to support these new instructions, the compiler would have to have been sent to MS LONG ago (i.e. early in Vista's development).

I've been wrong before, but that was my understanding.

Part of that argument was the fact that developers were very hesitant to use /G7. Also, they don't want to limit compatibility for older Intel CPU's and other AMD CPUs that are still floating that simply cannot use it. As far as Intel’s relationship with MS is concern we do have /G7 complier which uses SSE3 instruction set for netburst CPUs. However, the problem is it’s use in games vs regular applications.

Now, this is not my argument...it's what I read between arguments in other forums sometime ago. Now it gets ugly and there is a lot of technical jargon thrown out there. . You know how arguments go towards the end...total flame bait. The argument is that using /G7 exclusively limits compatibility to SSE2 compatible CPUs only. Therefore, a certain level of blending is use to meet compatibility for most CPUs. The counter to this argument is the use of 2 executables that would either use blending (which AMD shines most in) and for SSE3 which Intel shines most in. The counter to this argument was time constraints and lack of resources to include such vast increase in coding which they simply do not have time for.

Right now we have /G7 compiler which is ready for use by devs however, the major of PC gaming devs don't use it. They claim compatibility issues with their blend method. But, we already know that you can't use /G7 with blending its either one or the other (thus the reasoning behind 2 different executable files. However, they seem to have the knack to be able to use some optimization and still keep the code blended. But the performance gain is nothing compared to using /G7 exclusively.

But what about SSE4 then?
Well, that depends on who you ask, but I am sure MS will impliment the compiler for it very soon. The question is will we see it for XP? And, will we see updates for home application and pc gaming applications that will use this new technology? As much as I like to see this in XP and not just a Vista exclusive (to force you to buy that bloat) and, as much as I like to see future patch updates across the board, it's slim to none IMO. Unless Intel sends their teams to help dev's introduce this in their applications. Just as SSE4 is new to the industry, it's also new for the devs and it's all trial and error unless Intel?mS work together with the devs and helps them use it properly.

[Eastcoasthandle]

Again, unless the compiler is released to support the Media Boost correctly (AKA implementation of SSE4, etc) through MS's compiler which is given by Intel and used by the developers for business applications, home use applications and pc games through some level of patching it benefits no one.

[nn_step]

All it is is SSE4, which AMD is going answer back with SSE4a

[Carfax]

I wonder if video card drivers could be optimized with these instructions?

SSE2 optimized drivers gave quite a healthy boost in performance back in the day.

[ahmad]

Well this performance increase will only be for a few things at best. Don't expect games to benefit from it...

[[XC] Lead Head]

Well this performance increase will only be for a few things at best. Don't expect games to benefit from it...

Yup, I was messing around with a game, entering the commands to enable 3D Now, MMX, SSE, SSE2, etc.. gave me about 2 extra FPS tops

[RangerXLT8]

Yup, I was messing around with a game, entering the commands to enable 3D Now, MMX, SSE, SSE2, etc.. gave me about 2 extra FPS tops

3DNow, MMX, SSE, SSE2 are kind of irrelevant in this thread don't ya think lol. We're talking SSE4(which as of now is exclusive to Intel Core 2 Duo).

[DilTech]

3DNow, MMX, SSE, SSE2 are kind of irrelevant in this thread don't ya think lol. We're talking SSE4(which as of now is exclusive to Intel Core 2 Duo).

Exactly the problem. A lot of developers won't use something if it won't benefit a majority of it's users.

Most developers wait until something is wide spread before implementing it... In the case of processors, until both companies are using it, and even then it's iffy.

[Eastcoasthandle]

Exactly the problem. A lot of developers won't use something if it won't benefit a majority of it's users.

Most developers wait until something is wide spread before implementing it... In the case of processors, until both companies are using it, and even then it's iffy.
There is no proof they won't. The main factor being this time around is Vista and Vienna. Before there was no new OS around the corner where you can start fresh from. This is why I believe if they do impliment the compiler it may not be for XP.

[ahmad]

There is no proof they won't. The main factor being this time around is Vista and Vienna. Before there was no new OS around the corner where you can start fresh from. This is why I believe if they do impliment the compiler it may not be for XP.

It's enough burden for devs to optimize using SSE/SSE2/SSE3 (well SSE/SSE2 is fairly common now, so its familiar to most), and now you want SSE4? And now you want some new instructions as well?

It will take at least a few months to half a year to get compilers optimized for these new instructions, and thats if they feel like it. Besides, a lot of the Linux/UNIX boys prefer AMD, so I wouldn't imagine they are in any hurry.

[Eastcoasthandle]

It's enough burden for devs to optimize using SSE/SSE2/SSE3 (well SSE/SSE2 is fairly common now, so its familiar to most), and now you want SSE4? And now you want some new instructions as well?

It will take at least a few months to half a year to get compilers optimized for these new instructions, and thats if they feel like it. Besides, a lot of the Linux/UNIX boys prefer AMD, so I wouldn't imagine they are in any hurry.
That's why I said what I said. If Vista/Vienna wasn't around the corner I would agree with you. But the OS is about to take a serious change. That alone will force dev's to change. :D

[Eastcoasthandle]

It's enough burden for devs to optimize using SSE/SSE2/SSE3 (well SSE/SSE2 is fairly common now, so its familiar to most), and now you want SSE4? And now you want some new instructions as well?

It will take at least a few months to half a year to get compilers optimized for these new instructions, and thats if they feel like it. Besides, a lot of the Linux/UNIX boys prefer AMD, so I wouldn't imagine they are in any hurry.

Your comrade states:
All it is is SSE4, which AMD is going answer back with SSE4a

If this is true then this is your indicator that dev's are going to impliment this compiler for SSE4 :rolleyes:
And according to you this would define as "common" among CPU :cool:

[nn_step]

Your comrade states:


If this is true then this is your indicator that dev's are going to impliment this compiler for SSE4 :rolleyes:
And according to you this would define as "common" among CPU :cool:
If that is the case why isn't 64Bit common, since ALL current AMD and Intel processors are 100% 64 bit compatable :stick: they both use the x86-64 instruction set

[Eastcoasthandle]

Because re-coding for a bit transition is a HUGE undertaking. It's not like adding a few instructions here and there to help out w/ a few specific tasks (like SSE/2/3/4 would do). Things like Kernels have to be re-written FROM THE GROUND up to take advantage of 64-bit. Heck, it took 5-7 years just to get OSes from 16 to 32-bit (too long ago for most of you young ones to remember).

Processors have always implemented changes YEARS and sometimes DECADES in advance of the software taking advantage of them.
So, is the rumor true that Vista will lack this and we will see this exclusively in Vienna? Not sure how true it is but it's been said that Vista doesn't really take advantage of it. Not sure how or why though...

[nn_step]

Vista will not take advantage of this.

[nn_step]

This is correct. But it doesn't rule out a service pack or something adding the feature for a limited number of aps that might benefit from it (like say, Windows Media Player and Movie Maker).
The only part of Vista that "MAY" take advantage of this is Aero. however applications such as Windows Media Player are not actually part of the Windows Operating system.

[nn_step]

umm then why don't they just remove them and make them available for download?

[nn_step]

strange it isn't even a problem to remove them preinstall or post install for me :confused:

[shimq1]

Remove WMP, why? It won't do any harm to your computer.... Performance increase? Not unless you're on a computer from 1995.

[Eastcoasthandle]

Man, this will make Vienna awesome then. I just hope the rumors are true for a release date of 2007-2008 time frame...