Core i7-4770K is performance crippled

[Edrick]

That is the crux of the matter, a lot of PCs in the near future will be TSX enabled, if you plan on keeping your current CPU for at least 2-3 years I personally believe that it will come into play in a noticeable manner.

Agreed 100%.

So if you are planning to upgrade to Broadwell, Skylake, or Skymont, then not having TSX in your Haswell system is not a big deal. However, if you plan to keep your Haswell system and upgrade after Skymont, then it should be a deciding factor. Thats just my opinion.

 

[Revolution 11]

The ONLY CPUs without TSX are the K models. They only sell to overclockers. All systems Dell, HP, etc build will be with the non-K models. They vastly outnumber people using the K models. You can not use the small sample of people on this forum as a measurement in what the "top" mainsteam CPUs are.

I guarentee you all the Haswell Xeons will have TSX and all the Haswell-E when they come out.

Edrick, stop convincing me that the non-K chips are the better choice. I am already confused as it is with hyperthreading or no, I don't need to worry about TSX.

 

[Edrick]

Intel's IPP and MKL libraries, available with Intel Parallel Studio XE, for example were released 1 year ahead of the Sandy launch with AVX code paths, the same apply now with the AVX2 path, released at least a full year ago for core IPP imaging functions + more

I know, I read the whitepapers on both and adventured into the AVX world the since day SB came out. For my applications, FMA was the one I have been waiting for. And FMA has been around for decades (see other CPUs outside of Intel).

 

[R0H1T]

That is the crux of the matter, a lot of PCs in the near future will be TSX enabled, if you plan on keeping your current CPU for at least 2-3 years I personally believe that it will come into play in a noticeable manner.

Nope, outside of a few uber exotic business applications & perhaps top of the line AAA games no one & I mean the vast majority of developers or end users will even notice the lack of TSX, let alone it being more mainstream than say dGPU's, as in more people with have a dGPU than users actually employing TSX in the real world !

 

[Exophase]

Which ones? I was not aware of any. I may be mistaken however, but I really do not know of any.

All Celeron and Pentium branded processors are missing AVX. I haven't yet heard what the level of support will be with the Haswell variants. Maybe they'll have AVX2, maybe it'll have just AVX, or maybe it'll have neither. I'm hoping for the former but doubting it.

The ONLY CPUs without TSX are the K models. They only sell to overclockers. All systems Dell, HP, etc build will be with the non-K models. They vastly outnumber people using the K models. You can not use the small sample of people on this forum as a measurement in what the "top" mainsteam CPUs are.

I guarentee you all the Haswell Xeons will have TSX and all the Haswell-E when they come out.

The R series processors also don't have it:

http://ark.intel.com/products/76641/Intel-Core-i5-4670R-Processor-4M-Cache-up-to-3_70-GHz
http://ark.intel.com/products/76642/Intel-Core-i7-4770R-Processor-6M-Cache-up-to-3_90-GHz
http://ark.intel.com/products/76640/Intel-Core-i5-4570R-Processor-4M-Cache-up-to-3_20-GHz

And both i5-4430S and i5-4430 are missing it, showing it isn't just delineated by letter class:

http://ark.intel.com/products/75037/Intel-Core-i5-4430S-Processor-6M-Cache-up-to-3_20-GHz
http://ark.intel.com/products/75036/Intel-Core-i5-4430-Processor-6M-Cache-up-to-3_20-GHz

Who knows how Intel will play feature roulette with i3, Pentium, and Celeron models not to mention mobile variants.

 

[Edrick]

Edrick, stop convincing me that the non-K chips are the better choice. I am already confused as it is with hyperthreading or no, I don't need to worry about TSX.

I am not trying to convince anyone. I am just giving my personal opinions. I am a developer and I drool over new instructions. I will gladly give up a few 100mhz in order to get them. FMA and TSX are huge, just ask any other developer on these boards.

But since they are new to Intel, it will be sometime before we see a high adoption rate. And all legacy benchmarks do not account for them. So all the people asking "What do I gain today?", the answer is not much. So they generally go for the K models to try and get more mhz.

Like I said before, there is only so much more juice Intel can squeeze from legacy x86 code. New instructions are the future for tangible performance gains.

 

[Edrick]

All Celeron and Pentium branded processors are missing AVX. I haven't yet heard what the level of support will be with the Haswell variants. Maybe they'll have AVX2, maybe it'll have just AVX, or maybe it'll have neither. I'm hoping for the former but doubting it.

Thank you. I was not aware. As you can tell, I do not pay attention to those low level CPU lines. Seems stupid that Intel would omit AVX2 from any Haswell model, considering thats one of the main selling points.

 

[Exophase]

FMA and TSX are huge, just ask any other developer on these boards.

But there are plenty of developers (on these boards and elsewhere) that aren't so sure TSX will be that huge for desktop software..

 

[R0H1T]

All Celeron and Pentium branded processors are missing AVX. I haven't yet heard what the level of support will be with the Haswell variants. Maybe they'll have AVX2, maybe it'll have just AVX, or maybe it'll have neither. I'm hoping for the former but doubting it.



The R series processors also don't have it:

http://ark.intel.com/products/76641/Intel-Core-i5-4670R-Processor-4M-Cache-up-to-3_70-GHz
http://ark.intel.com/products/76642/Intel-Core-i7-4770R-Processor-6M-Cache-up-to-3_90-GHz
http://ark.intel.com/products/76640/Intel-Core-i5-4570R-Processor-4M-Cache-up-to-3_20-GHz

And both i5-4430S and i5-4430 are missing it, showing it isn't just delineated by letter class:

http://ark.intel.com/products/75037/Intel-Core-i5-4430S-Processor-6M-Cache-up-to-3_20-GHz
http://ark.intel.com/products/75036/Intel-Core-i5-4430-Processor-6M-Cache-up-to-3_20-GHz

Who knows how Intel will play feature roulette with i3, Pentium, and Celeron models not to mention mobile variants.

Great charging an arm & a leg for what is basically an L4 cache & then "crippling" it further, only yours truly "Intel" can be this wicked D:

 

[Edrick]

Great charging an arm & a leg for what is basically an L4 cache & then "crippling" it further, only yours truly "Intel" can be this wicked D:

All I am reading about the new L4 cache is that it really has a performance gain for the CPU and IGP. I really hope Intel does come out with a LGA1150 version of Crystalwell (L4 cache), I will be all over it.

 

[Exophase]

I swear, it's like Intel management dictate that a processor can only have so many total features checked, and if you add one another has to go.

 

[Edrick]

But there are plenty of developers (on these boards and elsewhere) that aren't so sure TSX will be that huge for desktop software..

Of course it all depends on the application itself. Naturally there will be some which can not befefit from it. From what I read, I believe that a lot will (including games).

Only time will tell if it lives up to what Intel says it can do. We have to give it time. I personally will be finding out for myself.

 

[beginner99]

I am not trying to convince anyone. I am just giving my personal opinions. I am a developer and I drool over new instructions. I will gladly give up a few 100mhz in order to get them. FMA and TSX are huge, just ask any other developer on these boards.

But since they are new to Intel, it will be sometime before we see a high adoption rate. And all legacy benchmarks do not account for them. So all the people asking "What do I gain today?", the answer is not much. So they generally go for the K models to try and get more mhz.

Like I said before, there is only so much more juice Intel can squeeze from legacy x86 code. New instructions are the future for tangible performance gains.

I disagree. Most software is nowadays developed in high level languages with a VM or interpreter (C#, Java, Python, PHP) and/or barley has any form of real multi-threading like JavaScript, Python, PHP.

So for any developer of those languages TSX is irrelevant until the VM / interpreter makes use of it and that will be years. So this basically already excludes most consumer applications.

What else remains? Games? Nope. As long as most users can't use TSX the whole game must be developed with that in mind and must run correctly and fast enough without it.

Only place were it will have impact is on server-side were you can more strictly control hardware and in general have newer hardware. Also concurrency of 100th to 1000th of threads is mainly an issue on servers and not consumer PCs. So yeah it might be relevant and interestign for developers of such software but for a gamer that wants to buy a gaming system now, forget it. He will be able to game on it in 5 years time after having changed the GPU once or twice already...

 

[Exophase]

HLE can be implemented transparently by the OS/thread libraries, so I guess we'll be able to find out pretty soon if games actually benefit from it. I think we will see very little use of restricted transactional memory in games.

I'm not a game developer so I don't really know what they're coded like these days, but I always thought they'd be of the nature where the threads are all chunked up to do one frame's worth of work then they merge then wait for the frame to end. And during their work they don't have to modify that many shared resources. Is this far off the mark?

 

[beginner99]

TSX makes it easier to write parallel code. But it has to be included in the overall design of the software. From the start of the design process.
...

Therefor I find it unbelievable that Intel did not include TSX in all their CPUs ? TSX will only be useful if programmers use it in the skeletal of their design. And you can't use it in the overall design, unless you know all hardware will have support for TSX. Otherwise you have to make 2 separate designs for your software.

Imho, big-time TSX acceptance/usage was going to be delayed until AMD would also support it. Now it's gonna be delayed another extra 1-2 years.

Weird.

Agree! Makes sense. And hence if i wanted to over clock i would easily pick the 4770k.

 

[BallaTheFeared]

how can you know if a low level library isn't using AVX, how do you monitor it?

IPC/Performance and heat

Can I say for sure? No, but if it's working in the things I used it for then it sucks anyways.

 

[Exophase]

I disagree. Most software is nowadays developed in high level languages with a VM or interpreter (C#, Java, Python, PHP) and/or barley has any form of real multi-threading like JavaScript, Python, PHP.

Generally speaking, that statement isn't true:

http://www.langpop.com/

You're possibly thinking of some particular domain. Total results are swayed by embedded. But even then, it's not like Java and C# are popular on say iOS, they're not even real options.

But instead of looking at statistics by usage I find it more relevant to look at statistics by how much it's being executed, because that'll be the most important. Of all the code running on your computer where do you think most of the runtime is going? Even some of the very high level languages you refer to still heavily utilize runtimes and libraries written in native compiled languages, not to mention the VMs/interpreters/JITs that execute the code.

It's this stuff that matters the most when we're talking about performance enhancements, even if only a small percentage of developers are responsible for it.

 

[BenchPress]

Your link to a benchmark isn't quantifiable proof, it's a joke. Where are the real world benchmarks, not theories and nonsense? How the hell do database transactions have any correlation to real world performance? How does that affect real world applications? Oh wait, it probably doesn't.

I've already explained why a database transactions benchmark is also relevant to other multi-threaded software. What's your explanation for saying that it "probably doesn't"? Please be specific.

None of this correlates into real world use or real world performance.

Again, why not? You haven't presented any arguments to back that up.

 

[iaco]

I disagree. Most software is nowadays developed in high level languages with a VM or interpreter (C#, Java, Python, PHP) and/or barley has any form of real multi-threading like JavaScript, Python, PHP.

So for any developer of those languages TSX is irrelevant until the VM / interpreter makes use of it and that will be years. So this basically already excludes most consumer applications.

What else remains? Games? Nope. As long as most users can't use TSX the whole game must be developed with that in mind and must run correctly and fast enough without it.

Only place were it will have impact is on server-side were you can more strictly control hardware and in general have newer hardware. Also concurrency of 100th to 1000th of threads is mainly an issue on servers and not consumer PCs. So yeah it might be relevant and interestign for developers of such software but for a gamer that wants to buy a gaming system now, forget it. He will be able to game on it in 5 years time after having changed the GPU once or twice already...

This. Games will NOT be using TSX for a very long time, probably when the next-next-gen consoles are out. That may be as long as 7-10 years away. Console ports will not be using TSX because resources won't be spent to optimize a game with a feature used by 1% of the installed base.

If you want higher frame rates then you a buy a better GPU, not a processor with TSX or AVX2.

If you want these instructions to be adopted by the rest of the market you should ask Intel not to cripple Celerons and Pentiums.

 

[Edrick]

What else remains? Games? Nope. As long as most users can't use TSX the whole game must be developed with that in mind and must run correctly and fast enough without it.

The TSX spec clearly states that any code that utilizes these new instruction will also work on CPUs without TSX enabled.....it will just not benefit from it hence not run as fast. So there is no reason why developers can not start using it today. It will work on ANY CPU, but only faster on TSX enabled CPUs.

 

[Edrick]

This. Games will NOT be using TSX for a very long time, probably when the next-next-gen consoles are out. That may be as long as 7-10 years away. Console ports will not be using TSX because resources won't be spent to optimize a game with a feature used by 1% of the installed base.

If you want higher frame rates then you a buy a better GPU, not a processor with TSX or AVX2.

If you want these instructions to be adopted by the rest of the market you should ask Intel not to cripple Celerons and Pentiums.

You clearly do not understand exactly what TSX brings to the table then. For starters it will make it easier to develop games with more threads, and hence able to utilize more cores. If you want more games to start using 8/10/12 threads, TSX can make it easier and cheaper to develop.

 

[Exophase]

The TSX spec clearly states that any code that utilizes these new instruction will also work on CPUs without TSX enabled.....it will just not benefit from it hence not run as fast. So there is no reason why developers can not start using it today. It will work on ANY CPU, but only faster on TSX enabled CPUs.

I mentioned this here http://forums.anandtech.com/showpost.php?p=35107534&postcount=163

Are you thinking of some other kind of solution? Because otherwise you won't want to try to transparently work around a lack of restricted transactional memory support, even if it's technically possible. But you're probably just referring to HLE.

 

[iaco]

You clearly do not understand exactly what TSX brings to the table then. For starters it will make it easier to develop games with more threads, and hence able to utilize more cores. If you want more games to start using 8/10/12 threads, TSX can make it easier and cheaper to develop.

It's clear you don't know the reality of the market. Most PC games are console ports and developers won't waste time developing with TSX in mind.

Like I said, if you want TSX to be adopted quickly by developers ask Intel to include it in their low end parts.

 

[colonelciller]

What business with the slighest respect for itself would use overclocked CPUs with the reliability issues and data corruption that follows?

All places I know, such an action gets you fired. And its considered one of the dumbest things you can possible do.

 

[Aikouka]

IMO Intel is at fault for marketing the 4770K as being a multiplier unlocked version of the 4770 (non-K) because that really isn't the case.

I don't exactly comb through marketing material as my bedtime reading or anything ( ), but I don't recall seeing anything that states that the only difference between K and non-K is that the prior has an unlocked multiplier. That may be the main difference that most sites will push, because most people probably don't care about the virtualization features that are also removed from the K series.

People are looking at TSX the wrong way around.
You're looking at how TSX can speed up existing applications that are multi-threaded. I don't think that is where the benefit is at all.

It can affect existing multi-threaded programs that are using locks to ensure data synchronization.

TSX makes it easier to write parallel code. But it has to be included in the overall design of the software. From the start of the design process.

Keep in mind that parallel code includes code that doesn't work on the same data sets. Parallelism is also common on large data sets where you simply break the data apart into chunks for each worker to process.

"Most users" don't care about overclocking either. So let's be clear here, people who are choosing between an i7-4770K or a i7-4770, care a lot about performance and use cutting-edge software that uses/will use the latest CPU features. Deliberately disabling a performance feature from an enthusiast CPU and asking more money for it is a major issue within this context.

You're still completely ignoring my main complaint. You're pushing the removal of TSX from the K-series processors as some great travesty, you have absolutely no idea of how this will affect the average user or even 99.9% of the users on this message board. You posted some silly SiSoft benchmark that uses database transactions as some be-all and end-all to indicate the performance problems with the K-series. Also, are you seriously comparing TSK to the clock multiplier unlock? Raising the frequency of the processor benefits anything computational as long as the processor remains saturated.

Honestly, I see no problem trying to make people aware that the K-series processors don't have TSX because they may want that if they do that sort of work. However, I do not think the way you went about it was correct for this setting (a technical sub-forum).