AMD CPU speculation... and expert conjecture

[8350rocks]

A10-6700 3dmark vs. i5-3570k and A10-5800k:

http://wccftech.com/amd-a10-6700-gpu-performance-exposed-80-faster-core-i53570k-20-a105800k-3dmark-11-fire-strike/

30% improvement over A10-5800k is very accurate...and that's not the flagship part...though it's about 90% or so of the performance from the 6800k

EDIT: It blows the doors off HD4K from intel...in everything...even the top of the line i5 CPU doesn't help their iGPU enough to make up the difference. It's 80% faster @ Firestrike!

 

[hcl123]

Web-browsing can be done in parallel

Scales naturally per connection.

If you mean one process/thread per tab , then yes. What chrome does, mostly.
If you mean more than 1 thread per tab, then no. A modern, standards compliant browser cant do this for most of the things. (Though they use the GPU for some things like composting, SVG shaders, and maybe text shaping too.)

More or less correct. What in terms of browsing within the confines of a single tab can you thread? Not much.

Some things can be parallelized, specially CSS selector matching and JS parsing, of the top of my head. The reason it hasnt been implemented much is because if how these interact with all of the other components.
So the summary is what we two have been saying all along :its difficult to parallelize complex code. Which the noncoders dont get.

Better tools, cooperative threads, a better model of threading which should be in next toolchains, as probably the HSA one. The hardware for the rescue would be transactional memory if it could be done simpler. Locks then would be much less of a headache... memory management, synchronization and race conditions issues could be gone or tremendously mitigated.

Yes, yet it wouldn't make it simple or easy to parallelize code, but the evolution of hardware demands this effort, or it would not make sense chips with more cores than they have now, being the worst picture that those cores can't have much more more IPC than they have now, a dead end... while with threads you can double or triple....

 

[noob2222]

Spoiler

Yes somehow i agree with that, but some how AMD like almost all others, have no say in this... its quite more complicated than that... it will be akin to force or require software developers to choose methods and tools for development, which logically isn't legal and can't happen.

If Maxon wants to use ICC for their Cinema 4D (from which Cinebench is derived), no one can't force them otherwise, or should, even if it means "obviously" a certain biasing towards Intel (ICC = Intel compiler collection).

The problem with Cinebench or more concretely with ICC, that already generated a lawsuit that intel lost, is that ICC forces different binary code paths that it generates attending to the CPUID, that is as example, if the target is genuine intel it will use AVX, if it is anyone else CPU, it forces SSE2.

VIA ( old Cyrix ) Nano found a solution by tricking CPUID, that is, it can give a false CPU identification when asked, but this is a solution that can gain them a lawsuit for false representation of product... the other solution imposed by court is to force ICC to be as fair as possible.

But with all this, i believe even with R11.5, when ppl look at those numbers and charts, it will be good to know that what they are seeing is a comparison of AVX crunching on Intel with SSE2 on anyone else.

Hardly oranges to oranges... if by chance ICC used SSE4 for "others", according to numbers floating around (SSE4 on BD uarch can be ~ 50% better than SSE2), then BD would be winning at Cinebench without much trouble... which wouldn't fair for intel either.

The best solution, since all comparisons are always *RELATIVE*, is to md5 the benchmark binaries ( a rock stable release) and use it along the time. At least that way we would know in all those comparisons, that what changes is the hardware not the software(the software will be frozen in time).

i've read about intel's compiler restricting non-intel cpus so many times... i proposed amd denouncing cinebench knowing that. i wasn't proposing that amd should ask maxon to change compilers. i meant that amd should just drop cinebench from their benchmarketing and use something else. something that scales well with more cores yet doesn't suffer from intel-bias. for example, softwares like 7zip can show how much advantage amd (fx) cpus have when all cores are loaded.

The problem is getting the money out of the way to allow Maxon to update their "Cinebench". Cinema 4d r14 from what I gathered has upped AMD Fx to sse 4.2. Cinebench is soo easy to use, you just click start. Where is the updated bench?

Cinebench 11.5 is just outdated software.

A CPU is not stronger than other just cause of his design but cause of the software and how much that software is optimized.

http://www.c4dcafe.com/ipb/topic/684...es-in-c4d-r13/

Cinema 4D R13 (RC 45040)
FX-8120 4.5Ghz 171sec
PII X6 3.8Ghz 240sec
FX 30% faster / X6 40% slower

Cinebench R11
FX-8120 4.5Ghz 7.25pts
PII X6 3.8Ghz 6.36pts
FX 14% faster

See the pattern?

http://www.xtremesystems.org/forums/showthread.php?276245-AMD-quot-Piledriver-quot-refresh-of-Zambezi-info-speculations-test-fans&p=5138329&viewfull=1#post5138329

Why is cinebench 11.5 so popular still? It shows Intel's superiority with sse2 vx avx without actually telling anyone thats whats happening. Find me a review that doesn't use cenebench ... What happens if you "denounce maxon". I wonder how fast maxon would drop support alltogether for AMD processors.

There was another aspect, but I can't find it now, on another site that showed the cinebench render in time along with the scores. Somehow the faster time scored lower on the 8120 (5.56) vs PII x6 (5.88) at stock speed.

 

[hcl123]

amd should prove and point out the fact that cinebench uses icc and is unfairly biased towards intel. then, as a gesture, they should drop publishing cinebench benchmark results from their promotions and ask reviewers too...

They already did most of that, but for AMD it doesn't matter since they can only "effectively" compare with their own previous solutions(its equally biased for all, making it non biased)... at least in my country marketing laws so demand.

In the true matter of fact neither intel or AMD do reviews, in the true sense of it. All its "free speech" of the internet, and none should be imposing anything whatsoever.

If you *believe* the speech or not is up to you.. and then you can add your own free speech to...

 

[Cazalan]

So the summary is what we two have been saying all along :its difficult to parallelize complex code. Which the noncoders dont get.

It's quite logical actually. Software and hardware has an inverse relationship. Writing parallel code is difficult where instantiating multiple CPU cores is easy. Increasing serial performance (IPC) is hard for ASIC engineers where the software engineer doesn't have to do much of anything.

There are only so many man hours between a TICK and a TOCK so you can only get so much IPC gain per 12mo cycle. On the software side you get a benefit of having over 1000 to 1 software engineers vs hardware engineers. This is an oversimplification but the onus is really on the software side due to sheer manpower.

 

[hcl123]

Actually Samsung is larger than Intel by a long shot, and Qualcomm is nearly the size of Intel.

ARM is formidable for many reasons but one key reason is ARM Holdings isn't greedy. They license the cores for pennies. They basically make it so cheap it's idiotic to even try designing a new CPU from scratch.

We don't know how well it scales into the 64bit world yet but we should have a better idea of that by the end of this year.

Yes i believe samsung is bigger, but is most consumer electronics not "IT". Qualcomm is not yet that big, and it has other businesses to, like energy.

ARM not being greedy is a reflexion of mentality and posture... if ARM would had wanted everything for itself, in the most aggressive competitive posture that many seem to like when comparing CPUs (beat, trash, wipe the floor with, that the winner does, if even is only by a couple %, to the loser that then sucks badly, is lame and incompetent ... sometimes i wonder if idiocy can be more abhorrent), it would had not licensed a single core.. but then probably it would be a tiny player even now ( there is a *moral* for every story).

 

[de5_Roy]

The problem is getting the money out of the way to allow Maxon to update their "Cinebench". Cinema 4d r14 from what I gathered has upped AMD Fx to sse 4.2. Cinebench is soo easy to use, you just click start. Where is the updated bench?

Spoiler

Cinebench 11.5 is just outdated software.

A CPU is not stronger than other just cause of his design but cause of the software and how much that software is optimized.

http://www.c4dcafe.com/ipb/topic/684...es-in-c4d-r13/

Cinema 4D R13 (RC 45040)
FX-8120 4.5Ghz 171sec
PII X6 3.8Ghz 240sec
FX 30% faster / X6 40% slower

Cinebench R11
FX-8120 4.5Ghz 7.25pts
PII X6 3.8Ghz 6.36pts
FX 14% faster

See the pattern?

http://www.xtremesystems.org/forums/showthread.php?276245-AMD-quot-Piledriver-quot-refresh-of-Zambezi-info-speculations-test-fans&p=5138329&viewfull=1#post5138329

Why is cinebench 11.5 so popular still? It shows Intel's superiority with sse2 vx avx without actually telling anyone thats whats happening. Find me a review that doesn't use cenebench ... What happens if you "denounce maxon". I wonder how fast maxon would drop support alltogether for AMD processors.

There was another aspect, but I can't find it now, on another site that showed the cinebench render in time along with the scores. Somehow the faster time scored lower on the 8120 (5.56) vs PII x6 (5.88) at stock speed.

is maxon so much bigger than amd that dropping support for amd cpus will not cause any backlash? i think a lot of people use opterons and pro/workstation cards made by amd.
i guess you're hinting that it's intel who gets reviewers to use older cinebench to make their own cpus look better. that's quite possible and knowing intel's 'tactics', may even be true. all the more reason for amd to campaign for compiler-neutral benchmarking, mainly the multicore-friendly ones. btw, i never said amd should denounce maxon (only cinebench, especially the versions that bias towards intel, if what you said is correct).

 

[hcl123]

Naaa!.. AMD should campaign for very compiler un-neutral, use HSA toolchain and get an order of magnitude better performance than even the top "competitor" CPUs lol

Can you imagine how much faster Cinema4D can be if coded with OpenCL/C++11 ?

Then reach a very solid stable version of cinebench out of it, and md5 it frozen in time. Cinebench then will be standard for measuring rendering with compute power, because even if it is frozen in time, the hardware will not, any advantage or bias this year can be reversed the next, and we would know that is because of the hardware... and reviews of hardware would have a truly coherent meaning.

That is the most important, above all.. meaning. All bias are only temporary, cause hardware changes a lot, and everybody has bias. Fraude is intentionally maintaining a bias with the intention of deceiving... which is not a far fetched hypotheses in the current status quo... an hypotheses which md5 and time freeze can single handily resolve.

 

[8350rocks]

Nvidia's top marketing guru leaves to join AMD:

http://semiaccurate.com/2013/05/20/sean-pelletier-leaves-nvidia-for-amd/

'Lightning Bolt' renamed "dock port" now available for AMD based laptops...want to power your laptop with a USB 3.0?

http://semiaccurate.com/2013/05/24/amd-finally-puts-dock-port-on-a-device/

 

[esrever]

Hmmmm... I wonder how AMD is gonna respond to a 400$ 770 (680)

http://www.anandtech.com/show/6994/nvidia-geforce-gtx-770-review/1
http://www.tomshardware.com/reviews/geforce-gtx-770-gk104-review,3519.html

over $150 in bundled games for the 7970GE + MIR on the current cards already compete pretty well. I do expect an official $50 price drop eventually.

 

[truegenius]

linux everytime !

ok, which linux or ubunto (if they are seperate things) will you recommend for a beginer (with no experience of any other os except windows) ?????
i will give it a try

 

[de5_Roy]

i liked the last ubuntu i tried. imo it's good for newbies. but i liked puppy linux more even though i am a linux newbie myself. it helped me to resuscitate a couple of dead pcs. and often helps me with pc maintenance. :lol: i think the latest ubuntu puts a bit too much pressure on old gfx.

 

[esrever]

ubunto works fine and gives pretty much a windows like experience.

 

[gamerk316]

Better tools, cooperative threads, a better model of threading which should be in next toolchains, as probably the HSA one.

No one is willing to go back to cooperative threading. No chance in hell. Anyone who proposes it to a SW engineer gets laughed out of the room.

The hardware for the rescue would be transactional memory if it could be done simpler. Locks then would be much less of a headache... memory management, synchronization and race conditions issues could be gone or tremendously mitigated.

Understand how transaction works: Essentially, do the processing without a lock, and if the memory contents have not changed from what they were when you started, then you can safely save your data. If they have, then you have to put a traditional lock in place, and do the processing again.

Now, in a latency sensitive program, why in the hell would you take the chance your processing time could DOUBLE in the worst case, for a very minimal potential speedup?

 

[8350rocks]

linux everytime !

ok, which linux or ubunto (if they are seperate things) will you recommend for a beginer (with no experience of any other os except windows) ?????
i will give it a try

Ubuntu 13.04 is the most stable, user friendly release available. You can get it from Canonical for something like $5 (If you want a disk, if you have the freespace, it's a free download and you can make your own OS disk or even flash drive). It's a very user friendly design with an easy to navigate UI, you'll need to get familiar with some of the commands in the OS, but they're not terribly complicated.

 

[The Q6660 Inside]

linux everytime !

ok, which linux or ubunto (if they are seperate things) will you recommend for a beginer (with no experience of any other os except windows) ?????
i will give it a try

Ubuntu is a distribution of Debian Linux, one of the first of its kind. I would suggest trying out Linux Mint, a distribution of Ubuntu with amazing visuals and said to be one of the best distributions of the former, you can install Ubuntu and Mint as a Windows Application as well for east deletion.

Someone with no knowledge of the OS trying Mint 14 out. http://www.youtube.com/watch?v=I-PTQVeCF9k

http://www.linuxmint.com/

 

[juanrga]

if cinebench is so biased, amd should really stop using cinebench. they should officially and explicitly denounce cinebench for cpu benchmarks instead of putting it in their own promotions. they should also ask hardware reviewers and put it in the review guides to not use cinebench. iirc they denounced some other benchmark claming they were partial towards intel or something, that was good. they should do the same with cinebench and start using softwares like handbrake and 7zip for cpu benching (i use both so they'd be far more useful for me ).

Cinebench is explicitly named in the FCT settlement as one of the biased benchmark that use the Cripple_AMD function

http://sharikou.blogspot.fr/2009/12/ftc-accuses-intel-of-rigging-benchmarks.html

Sites such as Anadtech continue using biased benchmarks like sysmark and Cinebench in their AMD vs Intel comparisons. They used biased benchmarks in their recent 'review' of jaguar. People at forums complained about their use of Cinebench and their hiding of information to general users. Moderators such as idontcare tried to justify their biased reviews by accusing users of not excercising due diligence when purchasing!!!

http://forums.anandtech.com/showpost.php?p=35070529&postcount=142
http://forums.anandtech.com/showpost.php?p=35070983&postcount=145
http://forums.anandtech.com/showpost.php?p=35077229&postcount=216

The whole thread is full of interesting info, including the dark history of Sysmark and detailed analysis of how non-Intel chips can get almost double performance when are faked as Intel chips

http://forums.anandtech.com/showpost.php?p=35064550&postcount=104

When more people started calling Anandtech biased. Moderators closed the thread.

I put this here because I see people often cite their 'reviews' here at Toms forums when attacking AMD.

 

[juanrga]

It is not confirmed but trying to press details out of a collegue that has a 6800K to test the verdict is impressive, trying to butter him up nicely I kinda got a 30% figure. I guess I will be selling my 5800K very soon.

30% on games or synthetic? And what memory? 2133?

 

[palladin9479]

linux everytime !

ok, which linux or ubunto (if they are seperate things) will you recommend for a beginer (with no experience of any other os except windows) ?????
i will give it a try

Ubunto is pretty much the "go to" starter linux. After your comfortable working inside the environment and know how to make a proper build chain you can get a bit more creative. Personally I prefer CentOS though I tend to heavily customize it after install. Debian has by far the largest software base but you'll run into problems with drivers and such due to their licensing model. Debian requires all code to be opensource, hardware venders don't always like to release their IP (they spend a lot of money developing optimized drivers) as OS for their competitors to copy.

 

[amdfangirl]

linux everytime !

ok, which linux or ubunto (if they are seperate things) will you recommend for a beginer (with no experience of any other os except windows) ?????
i will give it a try

Personally I use Scientific Linux. That's more or less the same as CentOS except it sounds cooler and has a better logo.

If you're just starting out I suppose you would do well to try Mageia.

http://www.mageia.org/en/

Go for the KDE version.

 

[palladin9479]

Regarding office apps, lets talk about basic things like search and replace. That is still single threaded. Saving a Docx to a Pdf, that is still single threaded.

Actually this provides a good opportunity to demonstrate what I've been talking about.

Now the common mentality is to see it as a serial process, and the act of searching and replacing ~is~ a serial process. What isn't serial is the data your working on. Assuming you have a 100,000 word document, how would we go about multi-threading that process (purely academic)? First you break the work load into chunks, either by having a pre-set "core count" variable, or by running a quick function at program start to determine maximum "threads" or what not. You can then dynamically break that 100,000 word data set into X numbered sets of 100,000/X words each with the breaking function set to ensure the data point start / ends are aligned to words. You can then run the "search and replace" function on all those sets simultaneously.

Of course this would have to be automated (search and replace all) as ~you~ the human are not capable of clicking multiple "accepts" at once. That is not a limitation of code but of the human operator.

Now this presents another problem, assuming "1" is your maximum divisor, doing all that work to attempt a breakup is wasted cycles vs just doing it in one swoop because there is no performance gained via multi-threading on a singe core CPU. And in all honestly something like "search & replace" is very rarely performance intensive enough to justify the extra coding work required to make it parallel, especially as there would be next to no benefit due to human interface limitations.

You can not parallize code that was designed to be serial, it's just not going to work. Instead the problem itself needs to be redefined such that the data and operations can happen simultaneously. Take "rendering" for example, and I'm talking about the CPU part of setup and data analysis. That's typically one giant thread on a system, you can break it into two threads with each having a different data scope and when finished it deposits it's results into a separate data segment then the others. Now in order to do this you need to consume significantly more memory as you don't want each thread messing with the others data sets, also the process needs to be VERY carefully designed and documented with respects to writing values out. DX11 does help this immensely (API limitations really do limit how parallel you can make things).

 

[hcl123]

Better tools, cooperative threads, a better model of threading which should be in next toolchains, as probably the HSA one.

No one is willing to go back to cooperative threading. No chance in hell. Anyone who proposes it to a SW engineer gets laughed out of the room.

I don't think so... But with so many things it depends on the targets the tools and the developers. Speculative threading (or speculative multithreading spMT) is a good target for a cooperative threading model. Matter of fact i think it was intel that originated the "Mitosis" compiler that centered about making a good job of (semi)automatically parallelize code speculatively.

http://pdf.aminer.org/000/542/803/mitosis_compiler_an_infrastructure_for_speculative_threading_based_on_pre.pdf

The hardware for the rescue would be transactional memory if it could be done simpler. Locks then would be much less of a headache... memory management, synchronization and race conditions issues could be gone or tremendously mitigated.

Understand how transaction works: Essentially, do the processing without a lock, and if the memory contents have not changed from what they were when you started, then you can safely save your data. If they have, then you have to put a traditional lock in place, and do the processing again.

Now, in a latency sensitive program, why in the hell would you take the chance your processing time could DOUBLE in the worst case, for a very minimal potential speedup?

You are describing *only* "hardware lock elision", in what seems a *software* transactional approach, which depending on implementation cannot be that uber advantageous.

A true Hardware Transactional Memory model, as i like to see it (lol)... is(can be) a speculative threading mechanism above all, or better said a thread oriented "data" speculation mechanism, it goes much behind locks, though those are essential... their threads by inherent propriety are cooperative and speculative (abort or commit).

Is it not possible to build good " hardware" support for speculative multi threading ? ... i think it is... and nothing can augment more the IPC of a sequential piece of code, than dynamically on the fly, break that code in pieces that execute in parallel, no matter if its in a speculative way. ILP (instruction level parallelism) pretty much is a dead end, doesn't hurt to try to speculate a little lol

HTM AMD way (over 100% speedup possible on that "test")
http://llvm.org/pubs/2010-04-EUROSYS-DresdenTM.pdf

The beast ( can't remember if this was posted here already)
http://translate.google.com/translate?langpair=auto|en&u=http%3A%2F%2Fdiybbs.zol.com.cn%2F11%2F11_106489.html
(EDIT: Original link http://diybbs.zol.com.cn/11/11_106489.html , the google translalate link seems to break on tis site)

4x speedup of the performance of a FX8150 on integer... and GCN compute cores based FlexFPU on those modules for a new kind of APU... AVX kind of code can be almost 9x speedup lol... base on a what can be an HTM based speculative multithreading uarch (spMT).
(probably based on simulations, but was not 1th April and the piece says AMD presentation at Beijin )

fake ? .. pipe dreams ?... then academia is full of those since looong, meaning none of those ideas is far fetched, or never thoughted before.

[EDIT 2: if HTM based and spMT, this could have advantages on a ARM 64bit uarch)

 

[de5_Roy]

Cinebench is explicitly named in the FCT settlement as one of the biased benchmark that use the Cripple_AMD function

http://sharikou.blogspot.fr/2009/12/ftc-accuses-intel-of-rigging-benchmarks.html

Sites such as Anadtech continue using biased benchmarks like sysmark and Cinebench in their AMD vs Intel comparisons. They used biased benchmarks in their recent 'review' of jaguar. People at forums complained about their use of Cinebench and their hiding of information to general users. Moderators such as idontcare tried to justify their biased reviews by accusing users of not excercising due diligence when purchasing!!!

http://forums.anandtech.com/showpost.php?p=35070529&postcount=142
http://forums.anandtech.com/showpost.php?p=35070983&postcount=145
http://forums.anandtech.com/showpost.php?p=35077229&postcount=216

The whole thread is full of interesting info, including the dark history of Sysmark and detailed analysis of how non-Intel chips can get almost double performance when are faked as Intel chips

http://forums.anandtech.com/showpost.php?p=35064550&postcount=104

When more people started calling Anandtech biased. Moderators closed the thread.

I put this here because I see people often cite their 'reviews' here at Toms forums when attacking AMD.

sigh. this post looks much more like knocking anandtech than providing anything informative about cinebench biasing towards intel cpus (nothing new to add) or how amd should raise public awareness to drop cinebench(versions that bias for intel) from being widely used. the sharikou link uses data from 2003 and 2009, this is 2013. the agner site is p.i.t.a. to navigate. both sites are even harder to get to relevant information through anti-intel tirades and pro-amd rants. both, including pro-intel rant and anti-amd tirades, are useless to me. my point was not about what intel did. i know about those already. my point was about how amd can fight the intel bias and raise awareness. amd being the smaller company doesn't have the resources to fight intel. but with some creativity and strategy, they can raise enough public awareness and industrywide awareness against benchmarks biased for intel cpus. that's what i was pointing out. i simply don't care if you(or anyone else) got your feelings hurt by anandtech forum members and posted here to vent. :lol:

 

[palladin9479]

Spoiler

Cinebench is explicitly named in the FCT settlement as one of the biased benchmark that use the Cripple_AMD function

http://sharikou.blogspot.fr/2009/12/ftc-accuses-intel-of-rigging-benchmarks.html

Sites such as Anadtech continue using biased benchmarks like sysmark and Cinebench in their AMD vs Intel comparisons. They used biased benchmarks in their recent 'review' of jaguar. People at forums complained about their use of Cinebench and their hiding of information to general users. Moderators such as idontcare tried to justify their biased reviews by accusing users of not excercising due diligence when purchasing!!!

http://forums.anandtech.com/showpost.php?p=35070529&postcount=142
http://forums.anandtech.com/showpost.php?p=35070983&postcount=145
http://forums.anandtech.com/showpost.php?p=35077229&postcount=216

The whole thread is full of interesting info, including the dark history of Sysmark and detailed analysis of how non-Intel chips can get almost double performance when are faked as Intel chips

http://forums.anandtech.com/showpost.php?p=35064550&postcount=104

When more people started calling Anandtech biased. Moderators closed the thread.

I put this here because I see people often cite their 'reviews' here at Toms forums when attacking AMD.

sigh. this post looks much more like knocking anandtech than providing anything informative about cinebench biasing towards intel cpus (nothing new to add) or how amd should raise public awareness to drop cinebench(versions that bias for intel) from being widely used. the sharikou link uses data from 2003 and 2009, this is 2013. the agner site is p.i.t.a. to navigate. both sites are even harder to get to relevant information through anti-intel tirades and pro-amd rants. both, including pro-intel rant and anti-amd tirades, are useless to me. my point was not about what intel did. i know about those already. my point was about how amd can fight the intel bias and raise awareness. amd being the smaller company doesn't have the resources to fight intel. but with some creativity and strategy, they can raise enough public awareness and industrywide awareness against benchmarks biased for intel cpus. that's what i was pointing out. i simply don't care if you(or anyone else) got your feelings hurt by anandtech forum members and posted here to vent. :lol:

Agner's is not a public news site but his own personal blog that he lets others read / post on. He makes many of his own libraries and messages around lots in ASM. That's how he's able to pinpoint exactly what Intel is doing and the backdoor tricks their implementing to buff their benchmark numbers. There are some people on there trying to make vender-neutral libraries that replace things like the Intel math library or some of the vector libraries. He's about as far from a "fanboy" that you can get.

 

[sarinaide]

It is not confirmed but trying to press details out of a collegue that has a 6800K to test the verdict is impressive, trying to butter him up nicely I kinda got a 30% figure. I guess I will be selling my 5800K very soon.

30% on games or synthetic? And what memory? 2133?

DDR31600

30% general performance gains as I understand, games and apps vary though, some yield marginal results while others are exponential hence Susan Xu stated that you will get a 20-40% performance boost. I don't have numbers at this point in time but I did get told that my beloved F1 2012 which I can get 39-42FPS @ 1080 High/Ultra settings sees a very nice gain so I am pleased and by the fact I am rocking a DDR3 2800 kit.