AMD Piledriver rumours ... and expert conjecture

Page 56 - Seeking answers? Join the Tom's Hardware community: where nearly two million members share solutions and discuss the latest tech.
Toggle sidebar Toggle sidebar
Status
Not open for further replies.
Reynod

Reynod

Administrator
Mar 26, 2007
14,921
16
42,215
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 ...
 
Sort by date Sort by votes
AMD knew the disadvantages of going with a longer pipeline probably better than anybody but Im sure it had to be done to isolate parts of the chip like the fpu as well as cranking up the ghz also "buffers" the horrible cache latencies. The Pipeline is nowhere near as long as prescott, which was I think about 30 stages. Bulldozer has I think 20 or so stages. The sandy and phenom has about 16 or so. If the branch prediction worked out, the added ghz can easily overcome the longer pipeline.

I would guess they had to use a longer pipeline for the module because of the shared front end with the instruction decode and such. Having slightly longer pipeline isn't a problem in itself, intel's been adding a few stages since core 2 and nothing bad has happened, sandy has 2-3 more stages compared to core2 and yet its IPC is much higher.
 
Upvote 0 Downvote
nice assessment, good posting.
Click to expand...
Its kind of crazy. A few months ago i set out to learn what would be best to go into a PC if i were to build one. Now i could tell the average person what they should use without thinking twice about it. I learned what is faster, but i am loving these discussions about what components make up a cpu/gpu. Now i want to know why its faster.

What i find really sad is all the problems that PD has. The things i mentioned were 100% based off of logical thinking. I may be putting myself in a level i really couldn't compete in, but it seems like AMD's engineers were not thinking strait when they made BD as it is. Making a cpu must be extremely hard, but with the resources that AMD has, they should be able to do a far better job than what they did.
 
Upvote 0 Downvote
Its kind of crazy. A few months ago i set out to learn what would be best to go into a PC if i were to build one. Now i could tell the average person what they should use without thinking twice about it. I learned what is faster, but i am loving these discussions about what components make up a cpu/gpu. Now i want to know why its faster.

What i find really sad is all the problems that PD has. The things i mentioned were 100% based off of logical thinking. I may be putting myself in a level i really couldn't compete in, but it seems like AMD's engineers were not thinking strait when they made BD as it is. Making a cpu must be extremely hard, but with the resources that AMD has, they should be able to do a far better job than what they did.
Click to expand...


If BD in fact is no more than failed server CPUs I'd say that Amd did a great job,that they sold so well that AMD is about
to release more models is testament of how well they sell . Intel has a harder time unloading their best offerings.
I have to conclude that BD is a success .
No matter how it scores in benchmarks it will still do anything you throw at it,AMD did a great job selling it.
" unlock 8 cores " hooked a lot of people haha.
 
Upvote 0 Downvote
yes, correct marketing is the key to more profit.

Example1
i was planning to build a budget pc for someone with a 'pentium' g850 and a guy nearby me said that instead of this 'old' pentium the better option is core2duo,
i asked which model of c2d, he replyed 'core2duo' :lol:
means pepole know that 'c2d' is better than 'pentium' brand and thus makes their decisions based on them.

most people buy/compare graphics on the basis of graphics memory only. And they don't know about the the gpu.

Even some shopkeeper also don't know these things.
1 shopkeeper was bulding a pc for someone with c2d+board with upto ddr3 800mhz support + 4gb ddr3 1333 ram. Without knowing that that 1333mhz will work at 800mhz.
 
Upvote 0 Downvote
hope i got this right:
amd claims that multithreading is the future.
amd wants to introduce a cpu gpu combined ecosystem that will efficiently handle multithreaded and single threaded tasks.
amd aims to introduce a fully heterogenous platform within 2013-2014~.

until 2014 amd will slowly integrate more and more fusion/hsa-type hardware into their cpus. they've already started with the apus and bulldozer.
this makes me think of bd and the apus as intermediate products. problems with intermediate stuff is that they tend to come out partially successful - which is what happened with bd, imo. but amd has successfully introduced apus and gcn cards. they didn't fail (performance, efficiency) with them. this is why i wouldn't call bd a success. i wouldn't call it a failure, but i wouldn't call it a success either. it's an ongoing process that will probably complete in 2014.
as for bd being a sales success - i think it has more to do with low supply, urge to upgrade from older cpu, cult following than actual performance.
 
Upvote 0 Downvote
bear with me
If you have a longer pipeline doesnt cache "misses" become more of an issue?
that it takes longer for the CPU to "recover" in that situation?
still learning :)
Click to expand...

Yes. That was one of the points hat was made when discussing BD pre-release. Of course it was overlooked or stated as a non-issue by many, some who are not even here now (Again).

Longer pipelines allow for higher clocks. I am sure if Intel lengthened SBs pipelines they could easily push 10GHz with current tech. Probelm is that that 10Ghz CPU wouldn't be nearly as fast as current SB CPUs are.

And cache misses affect it even more since the CPU will have to flush the pieplines and start again. Albiet they have more advanced pipelines, and this is where branch prediction comes into play, but if branch prediction is also not so hot then that combined with long pipelines can mean a majorly bad performing part.

Its kind of crazy. A few months ago i set out to learn what would be best to go into a PC if i were to build one. Now i could tell the average person what they should use without thinking twice about it. I learned what is faster, but i am loving these discussions about what components make up a cpu/gpu. Now i want to know why its faster.

What i find really sad is all the problems that PD has. The things i mentioned were 100% based off of logical thinking. I may be putting myself in a level i really couldn't compete in, but it seems like AMD's engineers were not thinking strait when they made BD as it is. Making a cpu must be extremely hard, but with the resources that AMD has, they should be able to do a far better job than what they did.
Click to expand...

I agree they were not thinking straight. That or they just didn't have the right tools for the job. Or the major changes (going FABless) made it hard to control. Who knows.

With more and more load shifting from CPU to the GPU the CPU is becoming a fancy fetcher.Ultimately the workings of the CPU is becoming less important.
Click to expand...

Sorry but when it comes to multiple tasks, a CPU cannot be beaten. If its a single task (like F@H) then a GPU will be massivley better. But a GPU has its place as does a CPU. And considering Quick Sync, we can see that GPUs are not always better at everything.

If BD in fact is no more than failed server CPUs I'd say that Amd did a great job,that they sold so well that AMD is about
to release more models is testament of how well they sell . Intel has a harder time unloading their best offerings.
I have to conclude that BD is a success .
No matter how it scores in benchmarks it will still do anything you throw at it,AMD did a great job selling it.
" unlock 8 cores " hooked a lot of people haha.
Click to expand...

They sold "well" because the yields form GF were pretty bad to start. A new arch on a new process (one AMD has no control over) means pretty bad yields to start.

Intel on the other hand has a mature process, much better yields and can also make many more CPUs than GF can. Thats why Intel cannot "offload" all thir CPUs as fast. They have way more. Thats always been the case when it came between AMD and Intel. Intel can make more at a time.
 
Upvote 0 Downvote
TBH, I have also seen posts by some presumably experienced design engineers stating that often such hand-tuning results in worse performance, since there is no way a team of humans is going to be able to check all the permutations that a computer tool running on a supercomputer, and using the design experience of thousands of experts, can.
Click to expand...

It comes down to human error and software being buggy. But you don't get the nice square cores and ram blocks without a fair amount of hand setting. Tons of input goes into the place and route of the logic blocks or when you look at the pictures of die it would be a jumbled mess.

It's a juggling act for the management team to allocate engineers for design, verification and test. I hear numbers like 5 verification per designer, but that probably needs to double as the software gets more complex.


 
Upvote 0 Downvote
I would guess they had to use a longer pipeline for the module because of the shared front end with the instruction decode and such. Having slightly longer pipeline isn't a problem in itself, intel's been adding a few stages since core 2 and nothing bad has happened, sandy has 2-3 more stages compared to core2 and yet its IPC is much higher.
Click to expand...

It's a trade off they made for a server chip, which for that it was ok.
The 16 core Interlagos is hard to beat in a quad rack (64 cores). That's a lot of power in 1 box.

It just wasn't right for desktop performance. Which I'm sure someone has already been fired for. They killed off the Phenom II before a 32nm die shrink.
 
Upvote 0 Downvote
Once you hand tune to get the parameters, I don't see why you can't automate the process using the hand tuned parameters.
Click to expand...

Maybe you can, if you have enough computing power and expert software. I think Intel probably does, given the size of their R&D and process budgets. Probably IBM too. And at least up until recently, IBM shared their process research with their fab club which included AMD.

However as somebody mentioned, there seems to be a lot of wasted space in BD.

 
Upvote 0 Downvote
the common computer user was fooled by Intel back in the day because they would just see that Intel had 3ghz while AMD had 2.2 and just think that the Intel was better
kind of like HP in cars
that is why even though Pentium 4 was a failure overall as a design it sold extremely well
to this day in my computer shop I see so many Pentium 4 computers and very few Athlon XP towers come in for repair
Intel was almost a genius in marketing the Pentium 4 and was smart enough to switch gears with the Core2 design
this maybe a simplistic viewpoint on my part since I am nowhere near an expert in CPU design and just barely understand the basics of CPU architecture
but I can appreciate how Intel (BTW very happily own a PHII x 4 925 Deneb @ 3.4 :) )
was able to position itself the past 10 years or so
Got to give Intel credit their design team is matched by their marketing team LOL
Click to expand...

Netburst had a lot of innovative ideas incorporated into its design, but its branch prediction and OoO processing was not able to keep that long pipeline from having to flush incorrect results too often. However Intel also spent many millions of $$ on advertizing, which IIRC AMD did not do. In fact, I recall AMD Opterons being advertized on posters on the back of city buses and some subway posters 10 years ago, but no TV or radio ads. In contrast, Intel had all the media including the more expensive TV & radio outlets covered. Which is why we all remember the Intel jingle, but nada from AMD.

The old adage "you have to spend money to make money" holds particularly true in advertizing 😛..
 
Upvote 0 Downvote
Netburst had a lot of innovative ideas incorporated into its design, but its branch prediction and OoO processing was not able to keep that long pipeline from having to flush incorrect results too often. However Intel also spent many millions of $$ on advertizing, which IIRC AMD did not do. In fact, I recall AMD Opterons being advertized on posters on the back of city buses and some subway posters 10 years ago, but no TV or radio ads. In contrast, Intel had all the media including the more expensive TV & radio outlets covered. Which is why we all remember the Intel jingle, but nada from AMD.

The old adage "you have to spend money to make money" holds particularly true in advertizing 😛..
Click to expand...


I know that even Sandybridge has had Netburst influenced design incorporated into it.
What Intel was smart about at the time was marketing Pentium based on ghz
The average computer buyer could grasp that idea
If it was a higher ghz it must be better right?
Just like with a sports car they will advertise HP numbers
Doesnt matter if the HP number is at 6000 rpms and unusable for normal driving
and that torque also matters when it comes to cars
The average public just wants a simple number to compare
I worked for RadioShack from 1999-2007
We mostly sold HP and Compaqs with AMD cpus
So many times that I was explaining how the Athlon was superior to the Pentium at the time
but for most consumers they wanted to see the "Intel Inside" sticker and always asked "How many ghz?"
trying to explain how AMD was more efficient and a better designed CPU wasnt easy
then Intel switched to lower speed Core2 CPUs
and changed their marketing to focus on the efficiency of the chip
what helped was that consumers could grasp that Core2 was two CPUs instead of one
so of course two CPUs are better than one
very simple concept to grasp
which overcame the GHZ being lower
not trying to insult the general public but when it comes to something complex like computers
you need a simple concept for marketing
ei more ghz is better,more CPUs are better
and Intel was smart enough to take advantage of that
plus having Intel commercials on TV including during the Superbowl
like you said AMD didnt have the budget to compete with that kind of advertising
even though in the performance enthusiast market Athlon beat Pentium up to Core2
for the general public Intel was the leader
really didnt matter who had the "performance crown"
even though AMD had the fastest desktop processors Intel outsold them
That is what AMD needs now is a simple sellable concept to market to the general public
even if they dont have the fastest CPU it doesnt matter
More cores anyone?
 
Upvote 0 Downvote
thinking about it
AMD is in a great marketing position with Bulldozer
generally at the same price points you will get more cores
FX 4100 vs I3-2100 4 vs 2
FX 6xxx vs I5-2400 6 vs 4
FX 8xxx vs I7-2500/2600 8 vs 4
plus generally higher clock speeds for the BD
so more cores and higher clock speeds
something the general public can understand
now take the money from R&D that you are not using to compete with anymore
and put it into your marketing/advertising budget
and AMD even has a cool name "Bulldozer" to market
have some TV commercials with a real Bulldozer running over computer towers
with some sort of funny spokesperson driving it
doesnt need to be anybody famous
think Dell with the Dell dude years ago
so now two simple concepts to market plus a strong brand image with the name Bulldozer
doesnt matter if Sandybridge beats BD in benches
general public never reads benchmarks
really AMD is in a better marketing position then they have been at in years
 
Upvote 0 Downvote
It comes down to human error and software being buggy. But you don't get the nice square cores and ram blocks without a fair amount of hand setting. Tons of input goes into the place and route of the logic blocks or when you look at the pictures of die it would be a jumbled mess.

It's a juggling act for the management team to allocate engineers for design, verification and test. I hear numbers like 5 verification per designer, but that probably needs to double as the software gets more complex.
Click to expand...

Heh, a "jumbled mess" is what my VLSI design project looked like. My lab partner & I spent quite a lot of time designing just a basic clocked inverter cell, with a fanout of 2, in a delay chain with taps so that you could get 2^n different delays out of it. Extremely simple circuit but we did all the step & place grunt work ourselves once we had the basic cell design, then of course we had to design the output drivers around each output bonding pad and then account for their delay, the clock drivers, etc etc. Just routing Vdd & ground took a few days..

I was amazed the thing actually worked when it came back from TI (who fabbed our designs for us as part of an arrangement with Texas A&M). Some of the other teams were much more ambitious, designing ALUs, crossbar switches, etc and they had a pretty low success rate. Didn't really matter since the purpose of the class was to learn the basic concepts, but still for bragging rights, we were happy ours worked 😀.

IIRC the testing & validation complexity goes up exponentially with the amount of logic on the chip, so I think it's amazing that these later CPUs work as well as they do. Or maybe their "errata" list goes up exponentially as well 😛..
 
Upvote 0 Downvote
Maybe you can, if you have enough computing power and expert software. I think Intel probably does, given the size of their R&D and process budgets. Probably IBM too. And at least up until recently, IBM shared their process research with their fab club which included AMD.

However as somebody mentioned, there seems to be a lot of wasted space in BD.
Click to expand...



"AMD says just as we have gone beyond the single-core era, we are rapidly coming to the end of the multi-core era and are now entering the new Heterogeneous Systems Era which will feature parallelism and power efficient GPUs."

They will use Opteron chips for highend market until such time that their APU is mature
I seriously doubt they intend to continue selling opteron chips in the long run wouldn't make sense .
I suppose they may keep selling them ,I just don't see the reasoning in buying them after 2014.
I don't see AMD improving their high end beyond using it as a test bed for the APU
A new era is unfolding no use beating a dead horse
 
Upvote 0 Downvote
More ghz, newer brand, more cache, more cores and lots of advertising are the things needed to get attention of general public.

In india many of my friends/classmates don't even know that amd is a company that make cpus, they only know intel is the one and it is because of advertising. (i was also one of them).
Even few months ago i wasn't aware of new cpu of amd (that is bd, when i wasn't a part of toms) i thought that phenom2 was the latest from amd. This is all because of lack of advertising on tv, newspapers, radios etc. (still haven't seen any advertisement of bd on tv)
 
Upvote 0 Downvote
Sorry but when it comes to multiple tasks, a CPU cannot be beaten. If its a single task (like F@H) then a GPU will be massivley better. But a GPU has its place as does a CPU. And considering Quick Sync, we can see that GPUs are not always better at everything.

both cpu and gpu will work in unison The CPU may be better at some tasks but so is the GPU better at other tasks


They sold "well" because the yields form GF were pretty bad to start. A new arch on a new process (one AMD has no control over) means pretty bad yields to start.

Intel on the other hand has a mature process, much better yields and can also make many more CPUs than GF can. Thats why Intel cannot "offload" all thir CPUs as fast. They have way more. Thats always been the case when it came between AMD and Intel. Intel can make more at a time.
Click to expand...

Intel doesn't unload more high end cpu's they are too expensive Intel sells a lot of 2600k and 2500k but their high end
are low volume sellers. They made so many 2500k and 2600k that there is now a backlog.
due market saturation. As remarkable as the 2500k and 2600k are they're not Intel's high end.
BD sold well because a lot of people have the Small D syndrome kind of like little boys and v8 cars

 
Upvote 0 Downvote
thinking about it
AMD is in a great marketing position with Bulldozer
generally at the same price points you will get more cores
FX 4100 vs I3-2100 4 vs 2
FX 6xxx vs I5-2400 6 vs 4
FX 8xxx vs I7-2500/2600 8 vs 4
plus generally higher clock speeds for the BD
so more cores and higher clock speeds
something the general public can understand
now take the money from R&D that you are not using to compete with anymore
and put it into your marketing/advertising budget
and AMD even has a cool name "Bulldozer" to market
have some TV commercials with a real Bulldozer running over computer towers
with some sort of funny spokesperson driving it
doesnt need to be anybody famous
think Dell with the Dell dude years ago
so now two simple concepts to market plus a strong brand image with the name Bulldozer
doesnt matter if Sandybridge beats BD in benches
general public never reads benchmarks
really AMD is in a better marketing position then they have been at in years
Click to expand...

I think your right so does AMD they will market more BD models And I bet they all sell out too,except the 4 core models.
All the kids want 8 core power .
 
Upvote 0 Downvote
"AMD says just as we have gone beyond the single-core era, we are rapidly coming to the end of the multi-core era and are now entering the new Heterogeneous Systems Era which will feature parallelism and power efficient GPUs."

They will use Opteron chips for highend market until such time that their APU is mature
I seriously doubt they intend to continue selling opteron chips in the long run wouldn't make sense .
I suppose they may keep selling them ,I just don't see the reasoning in buying them after 2014.
I don't see AMD improving their high end beyond using it as a test bed for the APU
A new era is unfolding no use beating a dead horse
Click to expand...




I'm sorry to say this but Amd should not bet on this alone! A small company like Amd should never put all their egg's in one basket!
 
Upvote 0 Downvote
Status
Not open for further replies.

TRENDING THREADS

Latest posts

Moderators online

Share this page

COMPANY
RESOURCES
FOLLOW
Top Bottom