AMD Piledriver rumours ... and expert conjecture

Page 24 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Status
Not open for further replies.
We have had several requests for a sticky on AMD's yet to be released Piledriver architecture ... so here it is.

I want to make a few things clear though.

Post a question relevant to the topic, or information about the topic, or it will be deleted.

Post any negative personal comments about another user ... and they will be deleted.

Post flame baiting comments about the blue, red and green team and they will be deleted.

Enjoy ...
 
Based on what I've read, the MS hotfix makes Windows BD as 4 cores with HT.

This is basically what I predicted: Because the shared resources in CMT kill performance, the "fix" is to basically avoid using the second core of a BD module as much as possible. Same logic that exists with Intel HT.

You'll get better performance using 4 or less cores, but you also lose the ability to turboboost, as you will use more BD modules at one time. I can actually imagine some two-three thread tasks where performance could theoretically suffer with this patch as a result, due to the loss of turboboost...
 
i could hear all the collective sigh of relief among amd-hopefulls and the sudden but subsequent gasp of dissapointment.
still, if even microsoft and amd both say that the win 7 patch and win 8 don't increase performance subsantially, may be it is time to focus on perfecting the new piledriver and trinity.
 
^^ I told you so? Using more modules reduces the ability to turbo, which could hurt performance. Thats why I fully expect happened. As thus, you improve performance in one area [3-4 cores used], but hurt it in another [1-2 cores used].

A busted design is busted, no matter how many fixes you add in to it.

EDIT

Huh, Anand is saying the reverse is true; that BD by default treats the core as HTT, and avoids using both cores of a BD module, preventing turbo from kicking in. [I thought the opposite is true? Or are we totally re-writing reality as it suits us?]

My basic premise is still the same though: You have a choice between maximizing turbo [using both cores of a BD module first], or avoiding the ~20% performance hit of CMT. You can't have both at the same time.
 
I don't think not getting max turbo is going to hurt 1-2 thread performance as the 20% hit using the 2nd half of a module would require a 20% boost in ghz, BD turbo isn't that high. The only thing you will lose by not sharing a module is the possible advantage of shared l2 cache between the 2 cores of a module, but that would require the software to properly load the cache.

People got fired for a reason, Read likely saw wich way gave better performance and told the software team to fix it properly.
 
^^ But in the case where one thread is doing more work then the second, that 20% hit could apply only to 5% of the total work done on the module, which would then make any increase in speed on 95% of the total work more significant then the 20% performance hit on 5% of the work.
 
one could consider this concept to be "lightly threaded apps". The thing is how much core usage does it take to lower the turbo a notch, 5%, 10%, 25% ... Its much easier to estimate 100% assumption as anything else can be 100% unpredictable with only 50% accuracy and 25% ...
 
Read the update:
http://www.anandtech.com/show/5251/microsoft-releases-hotfix-to-improve-bulldozer-performance

Update: VR-Zone reports (and we can confirm) that the download link for the hotfix is no longer functional. There were apparently unexpected performance drops in some cases after applying the hotfix and Microsoft is investigating the issues. Modifying the scheduler in Windows is not something to be done lightly, as it changes a core element of the OS, so more testing and validation for such updates is always a good idea.

Credit goes to reynod (posted message on BD Reviews and Opinions thread).
 
Well if you had popped in to Perth for a stopover I could have met you for a Redback beer down in Freo.

http://matildabay.com.au/

They don't sell CPU's ... though if they did I fear they would overclock well ... and cause dizziness.

Well maybe next trip, in another 4 years which seems to be the max amount of time my wife will go without visiting her family.. This time we rented a whole bus for the family here in Saigon, and drove up to her hometown Binh Dinh (small village near Quy Nhon) to visit about 150 of her relatives 😛. We stopped over in Nha Trang but the weather was nasty & huge surf so no boat rides out to the islands.. Lots of Aussies in Nha Trang.

My nephew has been over in SE Asia since the beginning of September, backpacking throughout Thailand, Cambodia, Laos and now Vietnam. Having the time of his life apparently: http://phud418.posterous.com/. In fact, he decided to hang with his Aussie girlfriend today instead of coming over to meet us at my brother-in-law's house here in Saigon. Wife's family wants to throw him a party (lots of food and cases of Heiniken or Tiger beer), but guess that'll have to wait until Monday 😛.
 
Fazers thats pretty nice.

We have relatives in Ubon (Thai) ... thats a bit over.

I just used the AC to clean the dust out of my trusty PC today as the abmient is rising fast down here (Summer) and I also declocked the beast back to std and unvervolted it a bit.

Its too hard to seriously game when it gets hotter ...

"Be quiet, I'm doing the fishstick! It's a very delicate state of mind."

I also reopened the BD thread for you guys ... since the good doctor changed my meds.

Please keep the BD stuff in the BD thread and the PD stuff in the PD thread.

Thanks.

:)
 
Fazers thats pretty nice.

We have relatives in Ubon (Thai) ... thats a bit over.

I just used the AC to clean the dust out of my trusty PC today as the abmient is rising fast down here (Summer) and I also declocked the beast back to std and unvervolted it a bit.

Its too hard to seriously game when it gets hotter ...

"Be quiet, I'm doing the fishstick! It's a very delicate state of mind."

I also reopened the BD thread for you guys ... since the good doctor changed my meds.

Please keep the BD stuff in the BD thread and the PD stuff in the PD thread.

Thanks.

:)

I assume you mean Air Compressor and not Air Conditioner, when you mentioned cleaning out your PC 😀. Did the same a few months back - impressive dust cloud and now the system runs maybe 5 degrees cooler. I wonder if it is feasible to have a completely closed case if you use water cooling for the CPU and GPU with a radiator/fan heat exchanger outside the case?

Hopefully GF will be able to fix their PD (partially-depleted) SOI process in time for PD (Piledriver) so as to enable lower power usage and maybe higher stock frequencies, which is what BD needed to get to where AMD said it would be at the beginning of this year. That Inq rumor you linked in the IB thread, about Intel delaying IB until May, shows that we need competition if true.
 
Us Yankees will kindly take some Aussie heat...if you want to give for Christmas! :sol:

No thanks. When you live in Arizona, the cold winters here are a blessing from the 110F+ heat.

still not quite cold enough for a white Christmas here. We just had a couple inches of 35 degree rain a few days ago.

Back to PD: What changes is AMD making to it from BD? or is that not quite known yet?

For PD, as far as we know higher clocks. But thats about it. Unless they did BD to see how it works and are cutting the fat off of PD. Whos to say. AMD has been ever so secretive these days, worried that they wont be able to sell. Thing is they will sell. I still sell plenty of 500/2600Ks even with Ivy Bridge being all over the news. People sometimes don't want to wait.
 
For PD, as far as we know higher clocks. But thats about it. Unless they did BD to see how it works and are cutting the fat off of PD. Whos to say. AMD has been ever so secretive these days, worried that they wont be able to sell. Thing is they will sell. I still sell plenty of 500/2600Ks even with Ivy Bridge being all over the news. People sometimes don't want to wait.
I think it is pretty clear now that all the talk about AMD keeping Bulldozer under wraps for fear of the Osborne Effect, was just another manipulation in a long line of bulldust that AMD fed their gullible fans and the general public.

If they do the same again in the lead up to PileDriver, expect another Turkey.
 
I think it is pretty clear now that all the talk about AMD keeping Bulldozer under wraps for fear of the Osborne Effect, was just another manipulation in a long line of bulldust that AMD fed their gullible fans and the general public.

If they do the same again in the lead up to PileDriver, expect another Turkey.

I don't think AMD could do all the design changes that BD really needed to perform better - cache latency for one. However I suspect they will fix some smaller problems, plus GF might tweak their 32nm so that PD can clock higher at the same TDP, and perhaps MS will come out with some sort of scheduler fix that actually works.

Or perhaps AMD will shed another 800M transistors from PD and get it really slimmed down 😀..
 
I don't think AMD could do all the design changes that BD really needed to perform better - cache latency for one. However I suspect they will fix some smaller problems, plus GF might tweak their 32nm so that PD can clock higher at the same TDP, and perhaps MS will come out with some sort of scheduler fix that actually works.

Or perhaps AMD will shed another 800M transistors from PD and get it really slimmed down 😀..
I still think they should just go back to the drawing back and just "simplify" their thinking/not think too far out of the box with the arch ad everything.
I heard they want 10 cores for PD...


And that's a cruel joke :lol:
 
Based on what I've read, the MS hotfix makes Windows BD as 4 cores with HT.

This is basically what I predicted: Because the shared resources in CMT kill performance, the "fix" is to basically avoid using the second core of a BD module as much as possible. Same logic that exists with Intel HT.

You'll get better performance using 4 or less cores, but you also lose the ability to turboboost, as you will use more BD modules at one time. I can actually imagine some two-three thread tasks where performance could theoretically suffer with this patch as a result, due to the loss of turboboost...

Except the only part being shared is the front end prefetch decoder and branch prediction unit. Actual instruction execution units are separate unless your doing 256 FMAC instructions (very rare right now).

More like this update was supposed to allow the instruction schedule to schedule thread from the same program on the same modules so that the branch prediction unit could better utilize cache. I've said it all along, the performance issues aren't in the execution units but in the caching system.

Anyone saying BD modules are "like HT" has absolutely no idea how HT and CMT work. HT is virtualizing your CPU register stacks, BD doesn't do that.
 
Status
Not open for further replies.