Core i7-4770K is performance crippled

[BenchPress]

....and you still have not produced any quantifiable benchmarks showing any such benefits of TSX.

Yes I have. It's a benchmark, and it's quantified.

Still fear mongering about a feature that no software uses, and one which is supposedly only benefit programmers with minimal performance benefit.

You have nothing to support that theory. Again, absence of proof is not proof of absence. Instead we know for a fact that TSX speeds up synchronizing between threads, which all multi-threaded software has a need for. So the fear that a 4770K will run future multi-threaded applications slower than a 4770, is justified and very real.

And again, "only" benefiting programmers is a misnomer. Anything developers do or don't affects the performance of the applications you run. If it's too hard to optimize multi-threaded software without the use of TSX, it affects not just the developers but also you, depending on whether you have a CPU with TSX support or not.

Lots of people in this forum have blamed developers for not creating software which optimally uses their 4+ cores. But that's just because it was too hard for the developers to get things right and still achieve good performance, with the primitive tools they had before. TSX goes a long way to fix the fundamental issue. But now that revolutionary feature is fused off in the flagship 4770K!

So yes, I'm calling it crippled.

 

[allenk09]

I have a fear my 4770k wont make it above 4.2Ghz

 

[wasabiman123]

Well, it's looks like I'd go with the non K version, will produce less heat and power, and has more features! That would beneficial to a ITX build. Do we know if the CPUs in the new consoles have TSX?

 

[mrmt]

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.

This criticism is spot on.

+1

 

[Face2Face]

I have a fear my 4770k wont make it above 4.2Ghz

Even @ 4.2Ghz it's still pretty dang quick. Should be close to a 3770K @ 4.5Ghz

 

[blackened23]

absence of proof is not proof of absence.

Again the burden of proof is on US since you're the one making outrageous claims. That isn't backwards thinking, definitely not. 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. Are our homes data centers? Give me a break, man. None of this correlates into real world use or real world performance. Again, you don't have any quantifiable, meaningful, real world benchmarks - thus making your claim of "performance crippled" laughable.

 

[Phynaz]

And that's probably very straightforward compared to NVIDIA and AMD's drivers.

So you don't know, therefore don't have any proof.

 

[Hixbot]

Intel has decided to cut TSX from K CPUs so they have something special to offer socket 2011 and Xeon. It's not lost on Intel that the "mainstream" K parts get much attention from enthusiasts as they are a generation ahead of the "enthusiast" socket 2011.
Some people will go to socket 2011 for the extra cores or the PCIE lanes, but many enthusiasts see the increased IPC and sooner availability of the mainstream parts and go that route.
Intel want enthusiasts to have a good reason to buy the socket 2011 parts, so they save what they can for those. They also deliberately design the mainstream K parts to be somewhat poor overclockers. Lets face it, if Haswell could overclock to 5ghz+ on air, would anyone wait for the next 2011 chips?

Updates to code to take advantage of TSX will only come if CPUs supported it. This is why it's available in some of the mainstream SKUs instead of none. It's simply to get the tech available to developers. TSX will make it's real debut with 2011 chips and Xeon, Intel is hoping there will be sofware and OS improvement by then to show it off. Only after that will it trickle down to the mainstream K parts.

 

[Idontcare]

True, but I do think it's too early to proclaim that the 4770k is "performance crippled" due to the lack of TSX - the fact of the matter is, it's a mostly unknown variable since nothing uses it as of yet. I'm just taking a wild guess here, but since we know that AVX instructions increase the variable voltage of Haswell, perhaps TSX is similar in that respect? Adding a feature that can exacerbate any potential thermal issues wouldn't be desirable - That's the only possible reason I can think of for the removal, just a wild guess. In any case, TSX is an unknown variable so to claim that the 4770k is "crippled" is a complete exaggeration.

I think we can agree that it is "feature crippled"...as was the case of the K-versions of Sandy Bridge (no VT-D) and Ivy Bridge. The difference here is that Intel has expanded the list of features which are now absent in their "mainstream enthusiast unlocked processors".

Whether or not the crippling of the TSX feature results in a performance penalty is the question - until we know the answer to that question I also agree we shouldn't characterize the situation as being a "performance crippling" situation.

 

[SOFTengCOMPelec]

we shouldn't characterize the situation as being a "performance crippling" situation.

The very fact that Intel have gone to all this trouble of disabling this feature (TSX), gives me a stomach feeling that it is either rather significant and/or especially problematic.
If it makes hardly any difference (performance wise), then why go to all the trouble of disabling it on the flagship overclockable version (unless it causes technical difficulties, such as excessive heat generation, or something).

EDIT: I think there is also a psychological reasoning behind wanting it. Now I know that I CAN'T have it on the K series, I WANT IT!!!. I want the triple meat version with all the trimmings and extra double toppings with quadruple sized extras!.

 

[BallaTheFeared]

I think we can agree that it is "feature crippled"...as was the case of the K-versions of Sandy Bridge (no VT-D) and Ivy Bridge. The difference here is that Intel has expanded the list of features which are now absent in their "mainstream enthusiast unlocked processors".

Whether or not the crippling of the TSX feature results in a performance penalty is the question - until we know the answer to that question I also agree we shouldn't characterize the situation as being a "performance crippling" situation.

In the same vein this thread reminds me of the "DRM in Sandy Bridge" that was all the rave at it's release.

That turned out to be nothing, but during that period it encompassed everything from youtube content to bittorrents.

I think the point I'm so poorly attempting to make can be summed up better with a picture...

 

[SOFTengCOMPelec]

In the same vein this thread reminds me of the "DRM in Sandy Bridge" that was all the rave at it's release.

In practice, for most K series customers, it is UNLIKELY to matter much (disabled TSX).

If TSX gives hardly any performance boost, then it does not really matter.

If TSX can give a performance boost, it will probably be a number of years, before lots of software filters through which utilises the TSX efficiently. By which time, the typical overclocking K series owner will have upgraded their computer.

 

[Exophase]

It's not about not running "properly". TSX is roughly comparable to Hyper-Threading in usefullness. Applications do run properly on a CPU without HT, but not optimally. Likewise without TSX applications won't run any worse than they did before, but they won't run any faster.

For HLE the TSX instruction prefixes will be ignored on a non-HLE processor, and they should be treated as normal locks the same way as if the transaction failed. But what about the more general forms of TSX - how will that be handled? Wouldn't the kernel need to trap the xbegin/xend instructions and emulate it them setting eax with with a fake status bit then calling the fallback after xend is reached? There isn't actually a status bit specifically for indicating the processor doesn't support TSX so it'd have to claim one of the other failure modes..

That would add a ton of overhead. More likely that only HLE will used transparently to non-TSX processors.

True, but I do think it's too early to proclaim that the 4770k is "performance crippled" due to the lack of TSX - the fact of the matter is, it's a mostly unknown variable since nothing uses it as of yet. I'm just taking a wild guess here, but since we know that AVX instructions increase the variable voltage of Haswell, perhaps TSX is similar in that respect? Adding a feature that can exacerbate any potential thermal issues wouldn't be desirable - That's the only possible reason I can think of for the removal, just a wild guess. In any case, TSX is an unknown variable so to claim that the 4770k is "crippled" is a complete exaggeration.

I doubt TSX adds much of anything to the power budget because it's mostly leveraging work that's already being done on the processor in order to maintain cache coherency.

TSX makes it (a lot) easier to do fine-grained locks.
Therefor, it should encourage programmers to make more use of fine-grained locks.
This will cause multi-threaded applications to make better use of more cores on average.

The way I see it, HLE encourages people to use coarser locking, not finer locking. Since the processor turns the coarse locking into nearly perfect fine locking. The only thing you have to watch out for is that you don't make your critical sections so heavy that they risk overflowing the transactional capabilities of the processor, or make the rollbacks prohibitively expensive. Although that could be a subtle problem in some obscure cases, AFAIK the transaction will fail if the processor attempts an L2 eviction of a transactional store. That could happen accidentally more quickly than you'd expect if you cause an associativity collision, which could happen repeatably for some code. There is of course a good chance this will never become an observable problem in practice..

The big problem is that if you let HLE influence how you write your code it'll come at the expense of everyone who doesn't have it. And it may not be a majority feature you count on for several years. Especially if Intel is very selective in where they enable it.

 

[bronxzv]

Whether or not the crippling of the TSX feature results in a performance penalty is the question

there is a whole chapter (chapter 12 in [1]) devoted to TSX optimizations, that's 28 pages of performance tuning advices, it's all about performance, what else can it be? I'm sure the guy who wrote this chapter and a lot of technical people at Intel are pissed off to see that it's not included in the flagship 4th generation product

it looks like most people here talk from the viewpoint of software consumers (using legacy software) when the OP talk from the viewpoint of a software producer (envisioning future software)

[1] Intel® 64 and IA-32 Architectures Optimization Reference Manual, Order Number: 248966-027, June 2013
https://www-ssl.intel.com/content/d...4-ia-32-architectures-optimization-manual.pdf

 

[Edrick]

I posted some TSX info in the other threads on this topic, so I wont post it again here. My opinion is that for the near future, it will be far more important to developers than it will be to gamers. 2-3 years from now, it will be important to everyone.

 

[BallaTheFeared]

there is a whole chapter (chapter 12 in [1]) devoted to TSX optimizations, that's 28 pages of performance tuning advices, it's all about performance, what else can it be? I'm sure the guy who wrote this chapter and a lot of technical people at Intel are pissed off to see that it's not included in the flagship 4th generation product

it looks like most people here talk from the viewpoint of software consumers (using legacy software) when the OP talk from the viewpoint of a software producer (envisioning future software)

[1] Intel® 64 and IA-32 Architectures Optimization Reference Manual, Order Number: 248966-027, June 2013
https://www-ssl.intel.com/content/d...4-ia-32-architectures-optimization-manual.pdf

How many pages are dedicated to AVX(1), how much did that matter to you, or most everyone here?

I can't think a of single thing I did with my Sandy Bridge CPU that used AVX, you know except stress tests that made it 20C hotter due to combined pipelines.


Who knows what TSX will bring, and who it will benefit, where is that person? That's who I want to talk to!

 

[Exophase]

I posted some TSX info in the other threads on this topic, so I wont post it again here. My opinion is that for the near future, it will be far more important to developers than it will be to gamers. 2-3 years from now, it will be important to everyone.

Steam survey doesn't have stats for AVX yet but it says only 56.13% have processors supporting SSE4.2 (and I'd guess Steam users to be using slightly better than average CPUs). Seeing how AVX isn't even enabled on all SB and IB based processors I expect the number to be much, much lower. It is now over 2 years old. TSX is being segmented even more aggressively. Intel processors are improving more and more slowly and giving people less incentive to upgrade. The PC market is slowly getting smaller and you're not going to find TSX on Atom any time soon. So how will it be important to everyone in 2-3 years when likely only a pretty small percentage of x86 users will even have hardware supporting it? Or by everyone did you just mean all types of users?

 

[Revolution 11]

I posted some TSX info in the other threads on this topic, so I wont post it again here. My opinion is that for the near future, it will be far more important to developers than it will be to gamers. 2-3 years from now, it will be important to everyone.

To play devil's advocate, it would take 2-3 years if every CPU released from this day forward had TSX. The fact that the top mainstream Intel CPUs (and most AMD CPUs) do not have TSX means even a bigger delay. I can see TSX not being widespread until 5 years from now. What is that, Skymont CPUs?

How many pages are dedicated to AVX(1), how much did that matter to you, or most everyone here?

I can't think a of single thing I did with my Sandy Bridge CPU that used AVX, you know except stress tests that made it 20C hotter due to combined pipelines.


Who knows what TSX will bring, and who it will benefit, where is that person? That's who I want to talk to!

To play devil's advocate again, you already have a i5-4670K. Confirmation bias?

I do agree with your conclusions.

 

[SammichPG]

Like with LGA2011 desktop chips?

You must have a lot of intel stocks and/or are too emotionally involved with the success of a corporation.

Just saying.

It's such an obvious marketing ploy in order to make more money that your unwillingness to accept the facts is quite amusing.

 

[Edrick]

To play devil's advocate, it would take 2-3 years if every CPU released from this day forward had TSX. The fact that the top mainstream Intel CPUs (and most AMD CPUs) do not have TSX means even a bigger delay. I can see TSX not being widespread until 5 years from now. What is that, Skymont CPUs?

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]

Seeing how AVX isn't even enabled on all SB and IB based processors

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

 

[bronxzv]

How many pages are dedicated to AVX

not much actually, most of the pages are the old SIMD ones with SSEx examples

(1), how much did that matter to you,

not much to me since all my AVX ports were finished before the 1st manual with AVX recommendations was released

or most everyone here?

can't speak for them

I can't think a of single thing I did with my Sandy Bridge CPU that used AVX,

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

 

[Edrick]

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

Perfect!!!

No one can talk about how AVX is not being used, until they can answer this question!

 

[wasabiman123]

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.

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.

 

[bronxzv]

Perfect!!!

No one can talk about how AVX is not being used, until they can answer this question!

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