TSX borked in Haswell and Broadwell

[meloz]

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.

Exchange CPUs with what? A new CPU which has the very same bug?

People should at least read up on the issue before commenting. This problem is deep and not easily remedied with Broadwell. Skylake is when we might get first working TSX implementation, we are more than a year away from this problem having a final solution.

As for cash refunds, will people/business do without their computers? Or will they use the cash to buy another PC that has Intel CPU and same bug? I don't think even this TSX bug is big enough to make the atrocious single-threaded AMD performance seem tolerable.

So Intel's solution of disabling TSX altogether seems reasonable (from their PoV), they are in a monopoly position that they can get away with these kind of 'oops'.

At best Intel can create some sort of program to give a minor cash compensation but these type of schemes involve a lot of lawyers and accountants and are hard to implement, for all we know they might still be working out the details.

One good thing is Intel have not sold Haswell-E (including Xeons), so potential consumers of those products will know beforehand that they are buying a defective CPU. Thus informed, they can make a better decision whether to buy Haswell/Broadwell server chips or sit out this cycle of CPU upgrades. They will have no real reason to complain.

I just wish this issue forces Intel to bring forward Skylake for both small and large socket, wishful thinking I know....

 

[*NixUser]

Skylake cannot come to market soon enough.

1H 2016 can't come soon enough, then.

I think i5-4670 and i7-4770 owners should get a free replacement 4670K and 4770K, respectively, just so they could at least overclock. Or maybe a full refund.

 

[jhu]

1H 2016 can't come soon enough, then.

I think i5-4670 and i7-4770 owners should get a free replacement 4670K and 4770K, respectively, just so they could at least overclock. Or maybe a full refund.

What? Get K version to overclock? That means those people weren't even using or cared about TSX in the first place.

 

[meloz]

What? Get K version to overclock? That means those people weren't even using or cared about TSX in the first place.

When they bought the CPU, they (theoretically) could have cared enough about TSX to forgo the option of 'k' SKUs.

Now, of course, everyone is in the same boat. But you cannot judge people's action and motivation retrospectively like this, that is illogical.

 

[jhu]

When they bought the CPU, they (theoretically) could have cared enough about TSX to forgo the option of 'k' SKUs.

Now, of course, everyone is in the same boat. But you cannot judge people's action and motivation retrospectively like this, that is illogical.

I can and did. TSX mainly speeds up database transactions. It's pretty much useless for home users. It is the most useful for banks and such. The ones using this won't be overclocking. If they are, they weren't serious enough about their data in the first place to matter.

 

[witeken]

1H 2016 can't come soon enough, then.

Skylake is planned for 2015.

 

[WhoBeDaPlaya]

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

There are certain members who are consistently acerbic. Just toss them on ignore and carry on.
If no one responds, they'll just get bored and piss off.

 

[DrMrLordX]

1H 2016 can't come soon enough, then.

I think i5-4670 and i7-4770 owners should get a free replacement 4670K and 4770K, respectively, just so they could at least overclock. Or maybe a full refund.

Unfortunately, the 4770k (among others) has still more instruction sets disabled, such as VT-d. Switching to that processor could cause more problems than would be solved by moving away from the 4770 (and you still wouldn't have TSX, either).

 

[*NixUser]

Skylake is planned for 2015.

I have seen some slides that claim that. However, I'm skeptical to say the least. Indeed it would be great if it was true, but I'm not so sure that I could go to a retailer and buy SKL-S at any point in 2015. It is much more likely we are getting Broadwell refresh after Q3-2015 than Skylake.

Also if you consider this TSX fallout and that Intel is required to rework the ISA it is very likely there will be a delay.

 

[ShintaiDK]

I have seen some slides that claim that. However, I'm skeptical to say the least. Indeed it would be great if it was true, but I'm not so sure that I could go to a retailer and buy SKL-S at any point in 2015. It is much more likely we are getting Broadwell refresh after Q3-2015 than Skylake.

Also if you consider this TSX fallout and that Intel is required to rework the ISA it is very likely there will be a delay.

Broadwell only comes in K model for desktop. Else its Skylake in Q2 2015.

For laptops the fix is most likely the F stepping Broadwell. Broadwell-Y aka Core M however will not be updated.

 

[Fjodor2001]

Broadwell only comes in K model for desktop.

A source posted earlier by witeken says differently, it said:

"Desktop Broadwell is coming in Q2 2015 and it should come in the form of Core i7 and Core i5 branded processors. Core i3 and Pentium based Broadwell parts might ship at later date, but there is a chance that Intel will skip Broadwell based LGA processors and move directly to Skylake-S, the Tock of the 14nm that also comes in the first half of 2015."

See this and then this.

 

[ShintaiDK]

Yes I know and its Fudzilla. About equal to asking the fortune teller

Still in doubt?

 

[witeken]

I have seen some slides that claim that. However, I'm skeptical to say the least. Indeed it would be great if it was true, but I'm not so sure that I could go to a retailer and buy SKL-S at any point in 2015. It is much more likely we are getting Broadwell refresh after Q3-2015 than Skylake.

Also if you consider this TSX fallout and that Intel is required to rework the ISA it is very likely there will be a delay.

You can go to the Q2 earnings call transcript and CTRL + F SkyLake (or is it SkyLink?) and you'll see Brian Krzanich make 2 unambiguous statements about Skylake being released in 2015.

Also, note that Skylake is just an architecture, so not dependent on manufacturing schedules of new fabs.

 

[StrangerGuy]

Intel QA has done a terrible lately with P67 SATA degradation, Haswell chipset USB3 bug, and this on the very CPU itself. Compared to the virtually perfect track record for at least a decade before 2011.

 

[ShintaiDK]

Intel QA has done a terrible lately with P67 SATA degradation, Haswell chipset USB3 bug, and this on the very CPU itself. Compared to the virtually perfect track record for at least a decade before 2011.

It doesnt mean that the same kind of bugs didnt exist before. The errata lists isnt much different in bug count from generation to generation.

 

[Fjodor2001]

Yes I know and its Fudzilla. About equal to asking the fortune teller
[...]
Still in doubt?

Some of those slides are quite old. The roadmap has been revised numerous times according to leaks along the way.

Also, I'm not sure what you're trying to say with the slides anyway. Only the second slide indicate that Broadwell-K will be the only Broadwell desktop CPU. The rest of the slides say differently, or nothing about it at all.

 

[Madpacket]

Exchange CPUs with what? A new CPU which has the very same bug?

People should at least read up on the issue before commenting. This problem is deep and not easily remedied with Broadwell. Skylake is when we might get first working TSX implementation, we are more than a year away from this problem having a final solution.

I did read up before commenting on the issue. Intel is "fixing" (removing a major performance feature) this "errata" (bug) via firmware updates. Nothing deep about this, it's pretty simple to understand.

Some people purposely bought a CPU with TSX over a K series one. If I were one of them I would be pretty upset and they should probably be compensated.

Intel should:

A) Give owners a credit or refund a specified amount of money (depending on model of Haswell CPU purchased) provided proof of purchase.
B) Give people the option to exchange to a newer CPU (of the same family and when available) or revision without the TSX bug.

For really good PR

C) Give people the ability to return the process for a full refund (as no Intel CPU currently exists with proper TSX implementation).

The argument of "TSX was only good for DB's anyway" is a little short sighted. First of all that's not true, there are many areas where TSX can improve performance and second, some of us do more than just play games on our PC's, some people actually develop HPC software and were planning on taking advantage of TSX to dramatically improve performance!

From one of David Kanter's articles on TSX explaining where TSX may be beneficial beyond improving SQL locks.

"There are immediate applications in a few areas particularly focused on low level software systems. First, as Azul and Sun already demonstrated, lock elision and TM are powerful tools for scalable garbage collection in Java, and other dynamic languages. Additionally, all the existing research on software TM systems can take advantage of Intel’s RTM to reduce overhead. Research from the University of Toronto has shown that software TM can provide a mild performance gain for a Quake game server (compared to lock-based versions), and Intel’s RTM should have far better results. It will be quite fascinating to see what various software vendors can do with the building blocks that Intel has provided."

http://www.realworldtech.com/haswell-tm/4/

Intel's white paper on TSX performance shows some real world performance gains (up to 1.41x) in multiple areas, not just DB's. Look at section 5 - Evaluating real-world workloads. TSX can even improve bandwidth in things like TCP/IP stacks.

http://pages.cs.wisc.edu/~rajwar/papers/SC13_TSX.pdf

 

[frozentundra123456]

I think the fairest thing to do would be to give a full refund for anyone who wants to turn in their cpu that was supposed to have TSX. This would definitely be good PR, and I seriously doubt many would turn in their cpus.

If they really bought for TSX, they have no other choice now anyway, so why bother. And if they didnt buy for TSX they dont really care and still would probably not turn it in.

Surprisingly Intel stock price is up today. I actually thought it would take a big hit, not so much for consumer cpus, but what negative effect it might have on delaying Haswell server purchases.

 

[ShintaiDK]

Some of those slides are quite old. The roadmap has been revised numerous times according to leaks along the way.

Also, I'm not sure what you're trying to say with the slides anyway. Only the second slide indicate that Broadwell-K will be the only Broadwell desktop CPU. The rest of the slides say differently, or nothing about it at all.

The oldest slide says nothing about Broadwell-K. The 2 others do. And the last one is quite recently. Both having the exact same information. And all 3 puts Skylake in Q2 2015.

 

[Sweepr]

Some of those slides are quite old. The roadmap has been revised numerous times according to leaks along the way.

Also, I'm not sure what you're trying to say with the slides anyway. Only the second slide indicate that Broadwell-K will be the only Broadwell desktop CPU. The rest of the slides say differently, or nothing about it at all.

Here's a few more stating 4C+3e Broadwell will be the only LGA part. It seems like you have no problem believing in the existance of 20-core FX chips without iGPUs based on 1 rumour yet you keep questioning everything related about Broadwell, Skylake, Intel's 14nm. I wonder why.

 

[Fjodor2001]

Here's a few more stating 4C+3e Broadwell will be the only LGA part. It seems like you have no problem believing in the existance of 20-core FX chips without iGPUs based on 1 rumour yet you keep questioning everything related about Broadwell, Skylake, Intel's 14nm. I wonder why.

I just mentioned that witeken posted a source saying Broadwell-K will not be the only Broadwell desktop LGA, that's all. It's not me questioning it, I just notice that there is contradictory information about this out there.

You don't have to go all hostile and bite my head of for that, do you? Go attack witeken instead if you absolutely must have someone to blame for posting the article about this.

 

[Gikaseixas]

It seems like you have no problem believing in the existance of 20-core FX chips without iGPUs based on 1 rumour yet you keep questioning everything related about Broadwell, Skylake, Intel's 14nm. I wonder why.

To be fair, the intel squad does the exact same thing when it's anything related to AMD or Nvidia ARM division and i never saw you complaining. Wonder why...

 

[Sweepr]

I just mentioned that witeken posted a source saying Broadwell-K will not be the only Broadwell desktop LGA, that's all. It's not me questioning it, I just notice that there is contradictory information about this out there.

All I see is a guess from Fudzilla (at best), nothing concrete about non-K LGA Broadwell. We already know you don't believe in Broadwell-K and Skylake launching at the same time but given the number of rumours / leaks pointing at that direction I honestly think you are beating a dead horse when you bring that up. Just my 2 cents.

 

[Fjodor2001]

We already know you don't believe in Broadwell-K and Skylake launching at the same time

Initially I didn't believe it based on the info available at that time - like a lot of other people. Because it's hard to see how it makes sense. Given the latest leaks, I agree things are pointing in that direction though. But following all these leaks, the Broadwell and Skylake roadmaps have changed back and forth quite a bit over the past year. So I don't take anything for granted and things may very well change going forward. There could be a Skylake-S delay for example. It's happened to Intel CPU releases before you know.

Regardless, I think Intel should explain their intentions with the Broadwell and Skylake-S desktop release plan. Because I don't think it's totally obvious why they would release the tick (Broadwell) and tock (Skylake) on desktop at the same time. They've never done that before.

Anyway all this is kind of OT. The thread is about the Intel TSX bug after all. Which may cause CPU releases to be delayed...

 

[jpiniero]

IIRC, there will be 4+3e Broadwell Xeons.

Regardless, I think Intel should explain their intentions with the Broadwell and Skylake-S desktop release plan. Because I don't think it's totally obvious why they would release the tick and tock on desktop at the same time. They've never done that before.

It is though - DDR4 is expensive... but something has to be the guinea pig. OEMs will sell both 4+2 Haswell Refresh and 4+2 Skylake in an attempt to get volume going for DDR4. It's not a high volume product but needed to get DDR4 out there.

Also remember the markets don't overlap.