Core i7-4770K is performance crippled

[BenchPress]

Beware! All of the K-model Haswell CPUs deliberately lack an important performance feature for multi-threading: TSX. Therefore the 'flagship' i7-4770K may actually be slower than the i7-4770, even when overclocked. D:

 

[Mopetar]

Do you have a link or anything else to verify all of that? I only skimmed through Anand's review the other day, but even linking to that would be helpful.

 

[Ayah]

http://ark.intel.com/products/75123/Intel-Core-i7-4770K-Processor-8M-Cache-up-to-3_90-GHz

No TSX-NI.

 

[MisterMac]

We kind of already knew this - but don't benchmarks show there isn't a massive performance difference?

 

[JimmiG]

It's no more "crippled" than Ivy or Sandy Bridge, though, since the feature debuted with Haswell.

Software has to be written to take advantage of the feature. It may become an issue in the long-term, though... but that depends on the rate of adoption and whether there's any benefit in desktop applications and games.

Generally, Intel’s TSX should be helpful for improving the programmability and scalability for concurrent workloads. Even with a modest number of threads, locks can easily limit the benefits from additional cores. While that is not a problem for 2-4 core processors, it is a much bigger factor going forward. Extremely popular applications such as MySQL have well-known locking issues that HLE or RTM could significantly alleviate.

http://www.realworldtech.com/haswell-tm/

It's strange how Intel chose to "cripple" the K-series by removing features that don't matter at all for 99.9% of the users, yet could potentially be very important for the remaining 0.1%. I guess they needed to find a way to penalize people for wanting to overclock.

 

[SOFTengCOMPelec]

Beware! All of the K-model Haswell CPUs deliberately lack an important performance feature for multi-threading: TSX. Therefore the 'flagship' i7-4770K may actually be slower than the i7-4770, even when overclocked. D:

What I would love to know is, did they (Intel) do this on purpose, purely for greedy marketing purposes, to stop people overclocking it, and using it as a cheaper version of their more expensive (sometimes) Xeon and/or 2011-socket chips OR ...

Are there genuine technical reasons for this, such as that the TSX instructions (and/or other disabled options, such as virtualization stuff) need more levels of gate delays (or similar), and so can't work at the 3.5 GHz (the fully enabled chips e.g. 4770 (nonK) are at 3.4 GHz I think) ?

 

[BenchPress]

Software has to be written to take advantage of the feature.

Not necessarily. TSX speeds up synchronization between threads, and this functionality can be part of a library, a framework, or even the operating system. So software can benefit from without even needing adjustment.

It may become an issue in the long-term, though... but that depends on the rate of adoption and whether there's any benefit in desktop applications and games.

It is comparable to Hyper-Threading in usefulness. Many of today's multi-threaded applications waste a lot of time trying to synchronize operations between threads. With 8+ threads, that's no easy task and can take up to 30% of all cycles. TSX eliminates the overhead of the cases where no synchronization was necessary. Games can definitely benefit from this too since they have to perform lots of different tasks within a matter of milliseconds.

It's strange how Intel chose to "cripple" the K-series by removing features that don't matter at all for 99.9% of the users, yet could potentially be very important for the remaining 0.1%. I guess they needed to find a way to penalize people for wanting to overclock.

It will matter to a lot more than 0.1%. What they've done is remove a performance feature from what's supposed to be their fastest CPU! :thumbsdown:

 

[BenchPress]

What I would love to know is, did they (Intel) do this on purpose, purely for greedy marketing purposes, to stop people overclocking it, and using it as a cheaper version of their more expensive (sometimes) Xeon and/or 2011-socket chips OR ...

Are there genuine technical reasons for this, such as that the TSX instructions (and/or other disabled options, such as virtualization stuff) need more levels of gate delays (or similar), and so can't work at the 3.5 GHz (the fully enabled chips e.g. 4770 (nonK) are at 3.4 GHz I think) ?

David Kanter from RWT, who has expert knowledge in TSX, says it's marketing.

 

[R0H1T]

What I would love to know is, did they (Intel) do this on purpose, purely for greedy marketing purposes, to stop people overclocking it, and using it as a cheaper version of their more expensive (sometimes) Xeon and/or 2011-socket chips OR ...

Yup !

Are there genuine technical reasons for this, such as that the TSX instructions (and/or other disabled options, such as virtualization stuff) need more levels of gate delays (or similar), and so can't work at the 3.5 GHz (the fully enabled chips e.g. 4770 (nonK) are at 3.4 GHz I think) ?

Nope !

 

[SOFTengCOMPelec]

David Kanter from RWT, who has expert knowledge in TSX, says it's marketing.

Yup !

Nope !

Obviously, you both are already using 4770K chips, if you had TSX it would have stopped you simultaneously posting at the some time (JOKE).

Thanks, it is so annoying that they have disabled the functionality, just for marketing reasons (assuming that is the reason why).

It's so contradictory, paying for the most expensive (domestic rather than extreme) chip, the FLAGSHIP model, which has full hyperthreading and yet no TSX, to help exploit it in the future.

It's a bit like paying top whack for an eight cylinder engined car, which has only got the first four cylinders enabled, and is speed limited, to just below the top speed of the four cylinder model.

 

[MisterMac]

So where's the benchies ?


Get some multipurpose lib or video something that's fully threaded to support TSX, and test clock for clock.


If when the first TSX enabled applications in daily use arrive - and the use is not more than 10% - i call red herring.


It'll be useful once we go +10threads, mainstream then and that's still sadly far out :/

 

[Ancalagon44]

I've never understood why the highest end Intel CPUs lack the virtualization extensions.

 

[BenchPress]

It's no more "crippled" than Ivy or Sandy Bridge, though, since the feature debuted with Haswell.

Exactly. So what's the point in getting a Haswell when you're not getting all of Haswell's features?

AVX2 and TSX are its main innovations relevant for a performance desktop system. So the i7-4770K takes away half the reason for upgrading to Haswell. Sure, AVX2 is a major feature on its own but when you don't have efficient multi-threading it can't be put to good use.

Note that both AVX2 and TSX have to be used by the software to benefit from them. So it will take some time for Haswell to show its true potential. An i7-4770 is a good investment. But with the i7-4770K you're paying a lot extra for only half the features.

 

[Atreidin]

So where's the benchies ?


Get some multipurpose lib or video something that's fully threaded to support TSX, and test clock for clock.


If when the first TSX enabled applications in daily use arrive - and the use is not more than 10% - i call red herring.


It'll be useful once we go +10threads, mainstream then and that's still sadly far out :/

In order to be a red herring it has to distract from an issue. What issue is being distracted from by pointing out that Intel intentionally removed a performance feature from a performance CPU? Who cares if nothing uses it yet. They made the feature so that people would use it. They need CPUs with that feature out there for people to use it. By lowering the number of CPUs with that feature out there, they are making it more difficult to make that feature become common. It seems pretty stupid to me.

 

[MisterMac]

In order to be a red herring it has to distract from an issue. What issue is being distracted from by pointing out that Intel intentionally removed a performance feature from a performance CPU? Who cares if nothing uses it yet. They made the feature so that people would use it. They need CPUs with that feature out there for people to use it. By lowering the number of CPUs with that feature out there, they are making it more difficult to make that feature become common. It seems pretty stupid to me.

Yes it's stupid.

Yes the basic feature set - in it self is pretty darn kickass.


...But if the actual practical real life CURRENT improvement - if one had a application rewritten for TSX - and then tested before\after and gains aren't great then what?


Then it's marketing hype is a red herring imho - AVX2 has proven it will speed certain workloads by crazy fast speed notch 11.

I don't see TSX doing that until we go into mainstream parallel crazy world (+10 threads for phones, laptops, desktops etc).

Which is a money grab from intel yes (Knowing that by the time TSX really gets around - there will be more threads, and haswell will be ancient and gone from enthusiasts).

But the feature in itself scales well with more parallelism - and we just don't have enough parallelism atm.
(Atleast that's how i see it - and why it's a requirement going forward, but not in this current instant).

 

[R0H1T]

Obviously, you both are already using 4770K chips, if you had TSX it would have stopped you simultaneously posting at the some time (JOKE).

Nice catch :biggrin:

Thanks, it is so annoying that they have disabled the functionality, just for marketing reasons (assuming that is the reason why).

Its all about moar money with Intel, if they could they'd suck you dry instantly without remorse

 

[JimmiG]

So where's the benchies ?


Get some multipurpose lib or video something that's fully threaded to support TSX, and test clock for clock.


If when the first TSX enabled applications in daily use arrive - and the use is not more than 10% - i call red herring.


It'll be useful once we go +10threads, mainstream then and that's still sadly far out :/

The application also needs to have well-known locking issues for the feature to become relevant. Just because a game spawns multiple threads doesn't mean it has locking issues. It seems it may mostly affect database applications and such.

We'll just have to wait and see on this one. It's a brand new feature. It will probably be time to upgrade again by the time it becomes an issue (if it does).

Exactly. So what's the point in getting a Haswell when you're not getting all of Haswell's features?

There's no point in getting a HW if you already have a fairly recent Intel CPU. If you have an older CPU (Nehalem, Core2, Phenom) and you're upgrading, you're probably not doing it because of TSX. If you are, just buy a non-K version.

I do agree that Intel's crippling of K processors is puzzling and annoying (AMD doesn't do this), but it's not a deal breaker IMO.

 

[MisterMac]

The application also needs to have well-known locking issues for the feature to become relevant. Just because a game spawns multiple threads doesn't mean it has locking issues. It seems it may mostly affect database applications and such.

We'll just have to wait and see on this one. It's a brand new feature. It will probably be time to upgrade again by the time it becomes an issue (if it does).

That's my point.

And the feature will at base only really shine when being applied to 16 20 thread EP\EX Monster CPUs.


Until the pragmatical way of programming for ... "client\desktop" changes.

 

[SOFTengCOMPelec]

That's my point.

And the feature will at base only really shine when being applied to 16 20 thread EP\EX Monster CPUs.


Until the pragmatical way of programming for ... "client\desktop" changes.

But there is another potentially serious consequence of the disabled TSX functionality.

Although software is suppose to be written to check that TSX is available on the CPU before using it, in practice, some software might not adhere to this policy.

E.g. Small one developer open source software, where they have written it specifically for "THEIR" TSX capable computer, and they have made it publicly available on their website, for others to use if they want.

Some pre-made virtual machine downloads, may have been created with TSX already enabled (I have already encountered a similar problem, when I tried to use pre-made virtual machines on an older computer, which unfortunately "crashed" the virtual machine instance, before I had time to adjust its features in the VM software, effectively making the pre-made VM's unusable on older computers, the missing feature was a virtual machine extensioned CPU, rather than TSX, obviously).

 

[ShintaiDK]

You need new code to use TSX.

The code can be using HLE or RTM. HLE is backwards compatible for CPUs without TSX. And most likely what will mainly be used for the next few years. Only the RTM mode requires TSX.

And games dont seem to be the prime candidate for TSX either.

Intel TSX targets a certain class of shared-memory multi-threaded applications; specifically multi-threaded applications that actively share data. Intel TSX is about allowing programs to achieve fine-grain lock performance without requiring the complexity of reasoning about fine-grain locking.

However, if there is high data contention the algorithm would need to change in order to have an opportunity for high scalability. There are no magic bullets that can solve the problem, since true high data contention implies that the algorithm is effectively serialized.

 

[SPBHM]

that's the price you pay for overclocking, you are not paying just $30 more,
it's a shame really, but I guess you can upgrade from your 4770K next year to a TSX enabled K CPU!!! maybe.

 

[SOFTengCOMPelec]

but I guess you can upgrade

I keep on reading rumours about even that 'OPTION' (to upgrade) being deleted every other generation (Tick/Tock), due to soldered down cpu motherboard *ONLY* intel generations in the future.

Back on topic, I agree, it is more annoying (Missing TSX) than end of the world stuff.

 

[cytg111]

Some more info on TSX

http://www.sisoftware.co.uk/?d=qa&f=ben_mem_hle

 

[SOFTengCOMPelec]

Some more info on TSX

http://www.sisoftware.co.uk/?d=qa&f=ben_mem_hle

Thanks for the link, it's interesting.

Quoting from it :-

HLE allows the rate of modify-only transactions increase by a massive 5x - blowing both classic and even R/W locks out of the water! Applications that use many threads and locks will see a huge increase in performance when changing their locking to HLE.

Five times faster, sounds significant, even if it only translated to a 50% or whatever overall speed up in practice.

 

[ShintaiDK]

Its worth noticing they used a database for the test as application. And that will be one of the best cases.