TSX borked in Haswell and Broadwell

[Fjodor2001]

Anandtech article up on the subject now:

http://www.anandtech.com/show/8376/...rratum-found-in-haswell-haswelleep-broadwelly

 

[Enigmoid]

All CPUs have a ton or errata, many of them known. Its just that the cause is not fully understood or the error in not reproducible, cannot be fixed with time and budget constraints. Often they are discovered after the CPU is in production and are too small to consider changing.

There are a ton of HW errata. List here

http://www.intel.com/content/dam/ww...-core-family-desktop-specification-update.pdf

TSX was known to be problematic well before today. But the true scope likely wasn't known until recently.

HSD114- Added November 2013.

Intel® TSX Instructions May Cause Unpredictable System behavior

 

[SOFTengCOMPelec]

All CPUs have a ton or errata, many of them known. Its just that the cause is not fully understood or the error in not reproducible, cannot be fixed with time and budget constraints. Often they are discovered after the CPU is in production and are too small to consider changing.

There are a ton of HW errata. List here

http://www.intel.com/content/dam/ww...-core-family-desktop-specification-update.pdf

TSX was known to be problematic well before today. But the true scope likely wasn't known until recently.

HSD114- Added November 2013.

But for HD114, it says:

This erratum may result in unpredictable system behavior. Intel has not observed this
erratum with any commercially available system.

Which sounds like it has never been reported and/or detected in "real life", outside of Intels testing/validation and maybe other areas.

There are other TSX erratums in the pdf you linked to, but a quick skim of them, seems to indicate that they are rare/obscure events, rather than major problems.

The pdf does partly explain one of the TSX issues, which is something to do with the RDRAND instruction, being put inside TSX stuff. Which sounds to me like a cpu working ok test, rather than something that useful commercial software would want to do. But there could be reasons why software might do it, maybe ?
I don't know if Intel will ever publicly explain exactly what is wrong with the TSX, and how to examine (recreate) the problem, outside of Intel.

I get the impression, that the whole TSX mechanism, is a big commercial secret. So that could be why they are being very tight lipped about the TSX issues, and how to recreate them.

tl;dr
Although there was a TSX errata before recent developments, it was NOT considered a serious bug, and there were no reports of it in the wild and/or it was not detected in commercially available software.

 

[Enigmoid]

Very likely intel realized that TSX had a few flaws. Those flaws were believed to be minor until recently.

TSX looks like an extremely difficult to implement mechanism.

 

[Burpo]

Intel, AMD & Microsoft all do it (always have).
Release it now.. fix it later.. get the money..

Nothing new.. It will get fixed..
"You can't make an omelette without breaking eggs"

 

[SOFTengCOMPelec]

Very likely intel realized that TSX had a few flaws. Those flaws were believed to be minor until recently.

TSX looks like an extremely difficult to implement mechanism.

I agree.
What little has come out about how it works (that I have seen), makes it look extremely complicated. Caches are already difficult enough to properly understand (for me, at least), without adding TSX to it.

The problem with stuff like that, is that there can be very hard to reproduce/trigger fault/bug mechanisms, which don't necessarily show up, even after thousands of hours of testing. The relative lack of commercial software which uses TSX, also makes it harder to test (I don't know to what extent, Intel would use other peoples software to test stuff, but I presume they include such tests, as well).

 

[escrow4]

There should be a recall/refund. I didn't pay ~$340 twice over here for a pair of 4770s. Doesn't matter if I'll ever use TSX or not, I didn't pay for a defective product.

 

[ShintaiDK]

There should be a recall/refund. I didn't pay ~$340 twice over here for a pair of 4770s. Doesn't matter if I'll ever use TSX or not, I didn't pay for a defective product.

I assume you demanded this to all your hardware and software suppliers.

Any of those given you a recall or refund yet?

 

[Fjodor2001]

I assume you demanded this to all your hardware and software suppliers.

Any of those given you a recall or refund yet?

Yes, if a product does not perform as advertised, you normally have the option to get a refund or get it fixed.

For HW and SW that is normally handled though driver/firmware/software updates, and HW repairs or exchanges.

The problem in this case is that the TSX bug is apparently not possible for Intel to fix in easy way. So Intel has to make a decision whether they think the cost of offering a refund or exchange program is larger than the cost of pissing off the customers that are affected by the problem, and the bad PR that this may result in.

 

[SOFTengCOMPelec]

Redacted - to minimise risk of pulling thread off track

In partial defence of Intel. I think (it is planned for the future that) some bios's will let you re-enable TSX, if you want. Since no one (publicly) noticed this issue (until about 1 year after general Haswell availability), it can't be that bad.

E.g. If you mainly only play games on the PC, does a tiny risk of TSX issues really matter that much.

I think it may be the server community, which is hardest hit by this. If/when TSX is used by significant quantities of server software.

 

[Galatian]

I think Intel should really compensate buyer. They could give us an DC P3700 for free for example ;-)

 

[ShintaiDK]

Yes, if a product does not perform as advertised, you normally have the option to get a refund or get it fixed.

For HW and SW that is normally handled though driver/firmware/software updates, and HW repairs or exchanges.

The problem in this case is that the TSX bug is apparently not possible for Intel to fix in easy way. So Intel has to make a decision whether they think the cost of offering a refund or exchange program is larger than the cost of pissing off the customers that are affected by the problem, and the bad PR that this may result in.

MS even gave up on fixing XP. I guess we can all demand a refund?

AMDs TLB bug didnt give a refund either. Neither did their out of spec CPUs.

Apple didnt give a refund for their broken iPhone antennas either.

Nor does the 100s and 10000s of other unfixed hardware and software bugs that we just have to live with.

It basicly needs to be on the scale like the 6 series SATA bug for something like that to even happen.

 

[cytg111]

The level of intel propaganda here is mind boggling, - it rivals international russian propaganda on other social media. Whatever, this place is bunk.

 

[jones377]

The level of intel propaganda here is mind boggling, - it rivals international russian propaganda on other social media. Whatever, this place is bunk.

You knew this would happen tho, just like in every other Intel/AMD thread. Just use the ignore feature.

 

[Zor Prime]

The level of intel propaganda here is mind boggling, - it rivals international russian propaganda on other social media. Whatever, this place is bunk.

When you have Intel marketing representatives such as ShintaiDK what else would you expect?

 

[TerryMathews]

MS even gave up on fixing XP. I guess we can all demand a refund?

AMDs TLB bug didnt give a refund either. Neither did their out of spec CPUs.

Apple didnt give a refund for their broken iPhone antennas either.

Nor does the 100s and 10000s of other unfixed hardware and software bugs that we just have to live with.

It basicly needs to be on the scale like the 6 series SATA bug for something like that to even happen.

10 years later, MS gave up. Not quite the same as a flaw in a processor that is still under warranty for most users now is it?

 

[mosco]

In partial defence of Intel. I think (it is planned for the future that) some bios's will let you re-enable TSX, if you want. Since no one (publicly) noticed this issue (until about 1 year after general Haswell availability), it can't be that bad.

It was found by someone outside of intel according to this article here.

 

[coercitiv]

Apple didnt give a refund for their broken iPhone antennas either.

Steve Jobs: free iPhone 4 bumpers for everyone

Jobs announced during the iPhone 4 press conference Friday that every iPhone 4 user will get a free Apple Bumper to help mitigate antenna problems. This will apply to every user worldwide who buys an iPhone through September 30. iPhone 4 owners who have already purchased one of the $30 bumpers will get a full refund. Users who don't like the bumper solution and are still unhappy with the iPhone can bring it back for a full refund instead.

 

[TerryMathews]

Steve Jobs: free iPhone 4 bumpers for everyone

#rekt

 

[meloz]

Skylake cannot come to market soon enough.

 

[Arachnotronic]

Fail.

 

[GLeeM]

Also, if TSX is that pointless, why include it in the first place in those CPUs? And why use it as a differentiating factor between SKUs?

Mostly marketing. Make people think they have to have it, even if they will never use it. The reason it was developed was to speed up transactions in massive databases with hundreds of users at the same time. One big reason hyper-threading (HT) has a bad name is because it does not speed up DB transactions (one of the few places it slows it down). TSX will improve that.

If I was to talk to someone in the street, and mention "TSX" (Intel), they probably would not have a clue what I was talking about, in 99.999% of the population.

From all the flap here, the same applies to enthusiasts!
TSX is not something that your CPU would use directly. It would be used by large database servers. An example of the advantage to you would be a fraction of a second saved when buying something online if it is during a high use period. The server is what uses it not your computer.

Ignore the above if your computer has a many terabytes size database that serves hundreds of users every minute!

 

[Madpacket]

This seemingly small issue will blow up on Intel. The right thing to do is offer refunds / exchange on the affected CPU's else they will suffer greater loss from bad publicity.

 

[witeken]

else they will suffer greater loss from bad publicity.

For a feature no one uses and only few people know of its existence and even less know what it is?

 

[TerryMathews]

For a feature no one uses and only few people know of its existence and even less know what it is?

As an advertised feature of the processor, it is a portion of what was purchased is it not?