Bulldozer!

[red1776]

I am starting to read stuff like this as well. One guy is allegedly releasing a reg patch later tonight. On one hand it seems inconceivable that AMD would release this, even as opposed to just delaying or starting over. This has to be more damaging in the long run??

I sincerely hope you get an answer and fix ASAP Waco My Bulldozer will be here on Thursday unless the UPS truck breaks down. Gigabyte pulled in F6 BIOS revision for my GA-FXA-990-UD7 for a couple days and then re-released it, and there was not any additional CPU's added to the support list. So I don't know if that means anything or not.

At any rate , if I get mine in before Asus gets you a fix, I will post my efforts in another thread...but I am not selling my 1100T just yet.

Edited October 19, 2011 by red1776

[Guest_Jim_*]

Nor is the OS maker liable for a hardware mistake that fails to work correctly with the current or backwards compatible operating systems that make use of multiple cores currently.

 

Who is at fault for the performance at this point? Not the OS maker for sure. If your processor design does not work with the operating system properly then who is to blame. As a hardware vendor you need to make sure your stuff works as well as it can and work with the software vendors.

But can one really say it is not working properly? It may be less performance than with Win 8, but it's not throwing errors because of the system. I'm not interested in anyone getting blamed because there really isn't any point to it, in my mind.

To defend AMD, they've been working on it for a long time (remember the 213 million transistor count comes from February) so it is not like Microsoft can claim they did not have time to prepare something to improve performance. Maybe Win 7 cannot be patched as BD would like, but everyone was aware of what was coming and what the software is capable of. Further, do we know if a redesign to BD for full performance with Win 7 would have been possible or a good thing? Perhaps letting Win 7 take a hit will allow a greater boost for Win 8 than if Win 7 was great out of the box. I don't know, I'm only speculating.

There still is no point in blaming anyone or anything, which is not what I'm attempting to do here. Based on reviews it looks like AMD needs to pick themselves up and figure out how to fix things, whether the fix be in-house or having to go to software manufacturers to work on better support. For those who have bought a Bulldozer then, if they are able to use it, they should just be patient (at stock speeds) for fixes to come, and for those who aren't able to use them, return the chip or wait it out. Sure, you can be ticked off, but that won't help the situation.

Full disclosure: I like AMD and have no intention of getting BD for awhile (mostly because I haven't the money).

[red1776]

And now something for the conspiracy theorist's out there. There is a guy out there that has has the Bulldozer as his main story. He was claiming that BD was underperforming well below caabilities and claimed that he was working on a Patch for the problem. A couple hours ago he posted that he was releasing the patch in a couple hours to the extremely angry post's of supposed Intel fans. now have a look at the link

 

http://quinetiam.com/

 

and no...I do not think he had the magic cure for BD

[ccokeman]

But can one really say it is not working properly? It may be less performance than with Win 8, but it's not throwing errors because of the system. I'm not interested in anyone getting blamed because there really isn't any point to it, in my mind.

To defend AMD, they've been working on it for a long time (remember the 213 million transistor count comes from February) so it is not like Microsoft can claim they did not have time to prepare something to improve performance. Maybe Win 7 cannot be patched as BD would like, but everyone was aware of what was coming and what the software is capable of. Further, do we know if a redesign to BD for full performance with Win 7 would have been possible or a good thing? Perhaps letting Win 7 take a hit will allow a greater boost for Win 8 than if Win 7 was great out of the box. I don't know, I'm only speculating.

There still is no point in blaming anyone or anything, which is not what I'm attempting to do here. Based on reviews it looks like AMD needs to pick themselves up and figure out how to fix things, whether the fix be in-house or having to go to software manufacturers to work on better support. For those who have bought a Bulldozer then, if they are able to use it, they should just be patient (at stock speeds) for fixes to come, and for those who aren't able to use them, return the chip or wait it out. Sure, you can be ticked off, but that won't help the situation.

Full disclosure: I like AMD and have no intention of getting BD for awhile (mostly because I haven't the money).

 

It may be working as intended. If the Turbo mode is the root cause of the bouncing of threads between cores in a module clocked dis-similarly then that comes down to arch design. Can Win 7 be patched ? I would think so but thats beyond the scope of my knowledge base. The bottom line is whatever the issue is the processor has poor single thread performance with the current iteration of Win 7. Multi thread performance is still lower than PII in many cases with a higher clock speed. Sure it overclocks like stink and a patch may be the magic elixir to get it to perform as well as intended. I Have another board to throw mine in and see if some of the bugged board stories are true for myself without resorting to speculation. I can only work with what I have and make an assessment based on that. I was expecting more but this is what AMD delivered out of the box. Long term does it show promise? Who knows. Many were ticked off due to the wait and many more over the performance and you can't please everyone. I do hope they get whatever is amiss sorted out so that the next iteration of the architecture can be successful. But currently it is not what was expected from them. At this point until it is sorted out I can get more performance from an Intel chip that costs less than $40 more.

[DanTheGamer11]

<br>Nor is the OS maker liable for a hardware mistake that fails to work correctly with the current or backwards compatible operating systems that make use of multiple cores currently.<br><br> Who is at fault for the performance at this point?  Not the OS maker for sure. If your processor design does not work with the operating system properly then who is to blame. As a hardware vendor you need to make sure your stuff works as well as it can and work with the software vendors.<br>

<br><br>So it was

 

Intels fault when they released hyperthreading? MS still fixed it?

 

http://hexus.net/tech/news/cpu/32243-rumour-amd-ready-favour-tsmc-globalfoundries/ ???

Edited October 19, 2011 by Dan The Gamer

[VaporX]

It may be working as intended. If the Turbo mode is the root cause of the bouncing of threads between cores in a module clocked dis-similarly then that comes down to arch design. Can Win 7 be patched ? I would think so but thats beyond the scope of my knowledge base. The bottom line is whatever the issue is the processor has poor single thread performance with the current iteration of Win 7. Multi thread performance is still lower than PII in many cases with a higher clock speed. Sure it overclocks like stink and a patch may be the magic elixir to get it to perform as well as intended.

 

I do not think there is any kind of legitimate argument that can be made for any kind of "patch" to to the OS fixing performance in any meaningful way. The performance issue at the end of the day boils down to the individual cores being under powered. I mean in testing against a Phenom II core at the same clock speed, single core testing with affinity set to core zero only the FX chip is about 9% or so slower across the board. This is not an OS usage glitch this is the design causing this.

 

AMD is taking a gamble and moving in a new design direction based on the way they see the future of software development unfolding. This means a ground up redesign and if the focus is the future then they might have had to give up some of the past to achieve it. The good news is that CPUs are already more powerful than most people have use for, so this chip, while nothing to get excited about does deliver good enough performance. That however is also it's flaw. After all the hype and wait, good enough, is NOT enough.

 

If this design was a good move is something we cannot know for a while. We need the software industry to give a more clear picture of the future development paths. This is a pure roll of the dice by AMD and while today it is only a lateral move, in the future this could either be a huge move or a huge bust. There just is no way we can truly judge it today beyond saying, okay it works, yawn.

[Waco]

The only thing I can see that would "fix" performance is if the scheduler knows to place processes on half of the cores when load is lower. This would allow 4 cores to drop to sleep and enable max turbo on the first 4.

 

I can't see it being a great patch for all CPUs, which is probably why it's slow coming.

 

 

I'm hoping for a microcode (BIOS) update to fix the L3 cache and memory write speeds...

Edited October 19, 2011 by Waco

[DanTheGamer11]

I do not think there is any kind of legitimate argument that can be made for any kind of "patch" to to the OS fixing performance in any meaningful way. The performance issue at the end of the day boils down to the individual cores being under powered. I mean in testing against a Phenom II core at the same clock speed, single core testing with affinity set to core zero only the FX chip is about 9% or so slower across the board. This is not an OS usage glitch this is the design causing this.

 

AMD is taking a gamble and moving in a new design direction based on the way they see the future of software development unfolding. This means a ground up redesign and if the focus is the future then they might have had to give up some of the past to achieve it. The good news is that CPUs are already more powerful than most people have use for, so this chip, while nothing to get excited about does deliver good enough performance. That however is also it's flaw. After all the hype and wait, good enough, is NOT enough.

 

If this design was a good move is something we cannot know for a while. We need the software industry to give a more clear picture of the future development paths. This is a pure roll of the dice by AMD and while today it is only a lateral move, in the future this could either be a huge move or a huge bust. There just is no way we can truly judge it today beyond saying, okay it works, yawn.

 

The module was designed to have 90% of the power of a true dual core... this is pretty much a server chip as it's main point is multitasking...

 

http://en.wikipedia.org/wiki/Hyperthreading Sounds slightly familiar at the start? OS wasn't made for it?

[red1776]

The performance issue at the end of the day boils down to the individual cores being under powered. I mean in testing against a Phenom II core at the same clock speed, single core testing with affinity set to core zero only the FX chip is about 9% or so slower across the board.

Hi Ed,

I would submit to you that this is a meaningless statistic, or at least an unknown. The stated design goal of this chip was to increase throughput by two intertwined cores working together. An optimum formula of shared and exclusive resources. It reminds me of a small download program years ago called 'Star downloader' i think it was called that divided a download up into two parts, each about 65% of the speed of a regular download and re-assembled them upon completion.

weather BD is being effected by a bad design or faulty implementation we don't know. It could be a brilliant design that is plagued with fixable latencies all over in the caches,branch predictors, scheduler, etc. It may be that they didn't get their arms around the manufacturing process and have problems with gate length for example. The power usage of this thing would seem to indicate that it is a leaky design/manufacture.

It could also be a case of Phenom I all over again. remember the TLB bug? the TLB cache had to be disabled and performance was dropped by 15% to be fixed in the XX50 revision. I personally suspect something like this as it would explain the poor single thread performance, and its very hard to believe that they would attempt Williamette Netburst done right without a better idea for execution. It seems obvious that AMD engineers took a "the roller-coaster doesn't leave until all the seats are full" approach. Unfortunately the person that will know the definitive answer won't be able to talk about it as he/she will have to reverse engineer it to get the answer. I agree with your conclusion that it will a very long time until this becomes obvious. I wonder if it's a sign that AMD thinks it's fixable that they are working on a B3 revision according to Lens2fire?

I think the bigger question is, where is John Fruehe hiding?

Edited October 19, 2011 by red1776

[DanTheGamer11]

Hi Ed,

I would submit to you that this is a meaningless statistic, or at least an unknown. The stated design goal of this chip was to increase throughput by two intertwined cores working together. An optimum formula of shared and exclusive resources. It reminds me of a small download program years ago called 'Star downloader' i think it was called that divided a download up into two parts, each about 65% of the speed of a regular download and re-assembled them upon completion.

weather BD is being effected by a bad design or faulty implementation we don't know. It could be a brilliant design that is plagued with fixable latencies all over in the caches,branch predictors, scheduler, etc. It may be that they didn't get their arms around the manufacturing process and have problems with gate length for example. The power usage of this thing would seem to indicate that it is a leaky design/manufacture.

It could also be a case of Phenom I all over again. remember the TLB bug? the TLB cache had to be disabled and performance was dropped by 15% to be fixed in the XX50 revision. I personally suspect something like this as it would explain the poor single thread performance, and its very hard to believe that they would attempt Williamette Netburst done right without a better idea for execution. It seems obvious that AMD engineers took a "the roller-coaster doesn't leave until all the seats are full" approach. Unfortunately the person that will know the definitive answer won't be able to talk about it as he/she will have to reverse engineer it to get the answer. I agree with your conclusion that it will a very long time until this becomes obvious. I wonder if it's a sign that AMD thinks it's fixable that they are working on a B3 revision according to Lens2fire?

I think the bigger question is, where is John Fruehe hiding?

 

I posted links to both of BD's problems!

 

"Tom's Hardware commented that the lower than expected performance in multi-threaded workloads may be because of the way Windows 7 currently schedules threads to the cores. They point out that "if Windows were able to utilize an FX-8150’s four modules first, and then backfill each module’s second core, it’d maximize performance with up to four threads running concurrently". This is similar to what happens on Intel's Hyperthreaded CPUs - Windows 7 "schedules to physical cores before utilizing logical (Hyper-Threaded) cores"."

http://www.amdzone.com/phpbb3/viewtopic.php?f=52&t=137255 so 2 baby cores is 180% faster than one Phenom II core so 2 baby modules are 20% slower than 2 Phenom II cores?

Edited October 19, 2011 by Dan The Gamer

[red1776]

I posted links to both of BD's problems!

 

Thats my point though. they are pointing out what and how it performs, not what the cause is. They don't know if for example if it has 100's of very small latencies that add up to its poor showing or a major problem or two. You said that you posted to BOTH of BD's problems. It may have hundreds of small problems that need to be fixed/ addressed. I do not profess to know mind you, I am just saying that they don't know exactly what is wrong either and or if it can be fixed.

Some of the most architecturally informed people around are scratching their heads because they can't figure why it does somethings well that they didn't think it would, and falls on its face at some things it appears that it should do well.

[HarryTaco]

What I dont understand is just "who" AMD thought they were going to sell this chip to? They absolutely know what performance they had comming out at what pricepoint. They have to know how it would be recieved in the PC community. They have been at this for too long not to know what to expect.

Basically they tried to dump a PC version of their server chip on us. From what I understand, we are not their prime market focus anymore anyway. So they get rid of the engineers that can hand craft chips and go all outsourced automated developement at an efficiency handicap and expect us to take the PC version. sheesh sorry AMD but when you actualy release a worthy PC chip at a worthy price, you wont have my money or that of like minded millions.