Why BD failed? AMD Ex-Employee speaks out!

[Horrux]

why BD faild? this is why:
BD die shot should look very much like this
http://cache.futurelooks.com/wordpr...1/10/AMD_Bulldozer_Review_FX-8000-500x201.png
but instead it looks like this
http://www.xbitlabs.com/images/news/2011-10/amd_bulldozer_orochi_die_floorplan.jpg

they've got 800 million transistors for in/out, logic, NB, wasted space, etc. so some how the have managed to waste a whole phenom 2 x4's worth of transistors on what should really take about, what, 150mill tansistors? also why is the "unified" L3 cache separated into 4 sections with such massive gaps?
maybe it was a learning curve using mostly software to design the chips, who knows? what is clearly evident is that they are not using the space available to them efficiently.

quoting some estimations on die area by X-bit:
"AMD publicly said that each Bulldozer dual-core CPU module with 2MB unified L2 cache contains 213 million transistors and is 30.9mm2 large. By contrast, die size of one processing engine of Llano processor (11-layer 32nm SOI, K10.5+ micro-architecture) is 9.69mm2 (without L2 cache), which indicates that AMD has succeeded in minimizing elements of its new micro-architecture so to maintain small size and production cost of the novelty.

As a result, all four CPU modules with L2 cache within Zambezi/Orochi processor consist of 852 million of transistors and take 123.6mm2 of die space. Assuming that 8MB of L3 cache (6 bits per cell) consist of 405 million of transistors, it leaves around whopping 800 million of transistors to various input/output interfaces, dual-channel DDR3 memory controller as well as various logic and routing inside the chip.

800 million of transistors - which take up a lot of die space - in an incredibly high number for various I/O, memory, logic, etc. For example, Intel's Core i-series "Sandy Bridge" quad-core chip with integrated graphics consists of 995 million."

This horror of truth gives me hope for piledriver. Even AMD should be able to fix that some, and increase performance, or decrease its size, or increase cache, or put a quad-channel DDR3 controller on there without increasing its size... SOMETHING DAMMIT. Or is it DAAMIT?

 

[bucketface]

you can't compare a diagram from a slide shot with a real core print....

ok well is this good enough...
http://images.anandtech.com/reviews/cpu/intel/sandybridge/review/die.jpg
almost no wasted space, seriously AMD could take some pointers from just analysing this image.
i'm not hating, just dissapointed and would like to see AMD actually bring something competitive to the game.

 

[Horrux]

^^^
Wow. That's beautiful. And this is coming from not so much an AMD fan as an Intel-hater.

 

[laszlo]

ok well is this good enough...
http://images.anandtech.com/reviews/cpu/intel/sandybridge/review/die.jpg
almost no wasted space, seriously AMD could take some pointers from just analysing this image.
i'm not hating, just dissapointed and would like to see AMD actually bring something competitive to the game.

i really don't know why is that space named "wasted" as no reviewer got the dye schematics in detail from AMD; can you point one? as i i think is still under NDA or unknown...

 

[bucketface]

Wow. That's beautiful. And this is coming from not so much an AMD fan as an Intel-hater.

I'm not a fan of any company, i do however like and try to support the underdog... unfortunately AMD is making it rather hard to do so at the moment. I really do hope that they manage to iron out the problems by the time Steamroller comes because thats when i'm looking to make my next big upgrade. until then i think i'll just get a 1055t.

i really don't know why is that space named "wasted" as no reviewer got the dye schematics in detail from AMD; can you point one? as i i think is still under NDA or unknown...

i wouldn't know i'm no expert after all, I am making some assumptions based off a limited knowledge of these things. That "wasted" space just looks like it's just being used to link bits up, maybe it's doing more though.... I think only some AMD engineers would know exactly what it's for, thing is SB has almost none of it and seems the better for it, saving a considerable amount of diespace.

Edit*
actually theres alot of die area used for the hypertransport bus, NB, misc I/O and mem controller, i guess about 400mill transistors. compared to SB's IO, etc which looks to be maybe 150mill transistors?.
i'm thinking it's all for compatability with AM3(+) socket.
Anyway this is all just my wild speculations.

 

[brandonwh64]

This is what I mean. Ok, no more off topic - PMs or GN please.

Maybe you should take comments better? He wasn't trying to troll you.

 

[NdMk2o1o]

Maybe you should take comments better? He wasn't trying to troll you.

He blatantly was for 3-4 of his last post and i'm not the only one who thinks so by the looks of things, however I am sure that qubit will gladly listen to what you have to say in PM/GN instead of bringing the thread off topic yet again.

 

[brandonwh64]

He blatantly was for 3-4 of his last post and i'm not the only one who thinks so by the looks of things, however I am sure that qubit will gladly listen to what you have to say in PM/GN instead of bringing the thread off topic yet again.

I was just stating the obvious.

Back on track then... Bulldozer will need to go through drastic changes on die to combat the effects it is getting publicly. As another member has mentioned before is that AMD had plenty of time to benchmark bulldozer to intel SB chips for a while now.

 

[alexsubri]

Why can't we all just get along?

anyways, I am more anxious to see AMD FX-8170 come out, I doubt it will consume less power, but I want to see it's performance against the FX-8150. Wasn't it stated that FX-8150 was more of a server and desktop CPU? I don't know if I am saying this correctly.

 

[Horrux]

anyways, I am more anxious to see AMD FX-8170 come out, I doubt it will consume less power, but I want to see it's performance against the FX-8150. Wasn't it stated that FX-8150 was more of a server and desktop CPU? I don't know if I am saying this correctly.

Yes, it looks like AMD pulled the design directly from server parts. Always a bad move, but AMD is on a budget.

 

[Steevo]

The best thing AMD could do is to pull out 2/3 of their cache and put in a logic co-processor to do more branch prediction and thread handling at the hardware layer, run it 2X the core speed and love the results.

 

[Super XP]

The best thing AMD could do is to pull out 2/3 of their cache and put in a logic co-processor to do more branch prediction and thread handling at the hardware layer, run it 2X the core speed and love the results.

Does this require a complete design overhaul? If not perhaps they may do something with Piledriver to change the playing field.

 

[Steevo]

Does this require a complete design overhaul? If not perhaps they may do something with Piledriver to change the playing field.

Design? Perhaps, depends on the hardware already in place and if they have room to add extra traces to interface between parts. If not does it really matter? The chip sucks at hardware level now, what could possibly make it worse?

They could have just added two more cores to the X6 and been better off with the die shrink.


There comes a point in time where you just have to call a turd a turd, and no amount of polishing will make it a gold nugget.

 

[HalfAHertz]

Yes, it looks like AMD pulled the design directly from server parts. Always a bad move, but AMD is on a budget.

See those 4 HT links? They're used for inter-socket communication in 2 and 4 socket server configs. So yes it is safe to say that BD was a purely server centric design from the get go.

 

[btarunr]

The reason DEC Alphas were always much faster than anything else is they designed each transistor by hand.

Good luck doing that with 2 billion transistors.

See those 4 HT links? They're used for inter-socket communication in 2 and 4 socket server configs. So yes it is safe to say that BD was a purely server centric design from the get go.

See these two QPI links?

They are for daisy-chaining two/multiple LGA1366 sockets, when this silicon is used in Xeon 3000 or 5000 series. In 1P/Core i7 series, Link 1 is rudimentary.

Just like AMD's silicon. Both AMD and Intel design common silicon for their enterprise and client products.

 

[qubit]

Why did Bulldozer underwhelm?

Here's a couple of articles from Charlie Demerjian with a technical breakdown as to why Bulldozer is so "meh", as he puts it.

Bulldozer is finally here, and the numbers all show that it isn’t going to to be taking any performance crowns. That said, in a few areas it closes the performance gap with Intel’s CPUs, but is lagging in quite a few others.

The story of Bulldozer and why it does what it does, both good and bad, can be summed up as death by 1000 cuts. There isn’t really any high point to the architecture, nor are there any really low points. To make matters worse, there isn’t any obvious smoking gun as to why things ended up so, well, meh. What you can get now, what you should have been able to get, and what you will be able to get from this new architecture is a long and complex story. Lets get started.

Part 1

Part 2

Reading both parts makes for depressing reading. The last paragraph of part 2 especially so:

With luck, those rough spots will be polished away with the coming of Piledriver, and the intended performance will shine through. If not, the next three generations are toast too, and we doubt AMD can survive that long without a competitive core. Will AMD bleed out from the 1000 cuts, or will the bandaids be applied in time? Ask me again after CES

 

[nt300]

Design? Perhaps, depends on the hardware already in place and if they have room to add extra traces to interface between parts. If not does it really matter? The chip sucks at hardware level now, what could possibly make it worse?

They could have just added two more cores to the X6 and been better off with the die shrink.

There comes a point in time where you just have to call a turd a turd, and no amount of polishing will make it a gold nugget.

I've been away too long. When people plug in there FX 8150 or any Bulldozer CPU it does what it was meant to do right. I don't see calling something that is factually innovative a turd. Yes Bulldozer definitely needs some fine tuning and polishing, which is why we all hope Piledriver offers the performance Bulldozer was suppose to offer.

AMD's previous design is old and needs to be dumped into the garbage. What they need to do now is focus on making Bulldozer/Piledriver/Steamroller/Excavator the design it was meant to be for high performance desktop chips.

 

[Horrux]

I've been away too long. When people plug in there FX 8150 or any Bulldozer CPU it does what it was meant to do right. I don't see calling something that is factually innovative a turd. Yes Bulldozer definitely needs some fine tuning and polishing, which is why we all hope Piledriver offers the performance Bulldozer was suppose to offer.

AMD's previous design is old and needs to be dumped into the garbage. What they need to do now is focus on making Bulldozer/Piledriver/Steamroller/Excavator the design it was meant to be for high performance desktop chips.

But aren't Intel's i7 direct descendants of the original Pentium Pro of 12 years ago?

 

[CDdude55]

I've been away too long. When people plug in there FX 8150 or any Bulldozer CPU it does what it was meant to do right. I don't see calling something that is factually innovative a turd. Yes Bulldozer definitely needs some fine tuning and polishing, which is why we all hope Piledriver offers the performance Bulldozer was suppose to offer.

AMD's previous design is old and needs to be dumped into the garbage. What they need to do now is focus on making Bulldozer/Piledriver/Steamroller/Excavator the design it was meant to be for high performance desktop chips.

Well it's not really ''factually innovative'', if it was innovative it wouldn't be as crappy as it is. Granted, i do commend them on trying a new design, i'm glad they tried something different instead of tweaking the K10 design over and over.

I just hope they improve BD soon.

 

[Damn_Smooth]

Well it's not really ''factually innovative'', if it was innovative it wouldn't be as crappy as it is. Granted, i do commend them on trying a new design, i'm glad they tried something different instead of tweaking the K10 design over and over.

I just hope they improve BD soon.

You and me both man. I don't mind skipping BD, but I hope something comes out that runs on AM3+ and is decent. ASAP.

 

[erocker]

I've been away too long. When people plug in there FX 8150 or any Bulldozer CPU it does what it was meant to do right. I don't see calling something that is factually innovative a turd. Yes Bulldozer definitely needs some fine tuning and polishing, which is why we all hope Piledriver offers the performance Bulldozer was suppose to offer.

AMD's previous design is old and needs to be dumped into the garbage. What they need to do now is focus on making Bulldozer/Piledriver/Steamroller/Excavator the design it was meant to be for high performance desktop chips.

Yes, let's build an awesome innovative car of the future and sell it. It's slow and can't really drive on many roads but let's sell it anyways. People will buy it right up, they love our cars.

 

[cadaveca]

I just hope they improve BD soon.

meh. if they could up clocks to 4.6 GHz, 5.0GHz turbo on air..that'd be good eonugh, no? That seems realisitic with process refinements, anyway. PhenomII 955 to 980 went about that far, but did take quite some time.

 

[Frick]

Yes, let's build an awesome innovative car of the future and sell it. It's slow and can't really drive on many roads but let's sell it anyways. People will buy it right up, they love our cars.

Like Citroën..

 

[Steevo]

Yes, let's build an awesome innovative car of the future and sell it. It's slow and can't really drive on many roads but let's sell it anyways. People will buy it right up, they love our cars.

You Fanboi !!!!!!

Wait, what?....


Intel did jump back 4 YEARS in CPU design to what they were doing with Pentium 3 to make the leap forward to "Core". AMD had a good design with 5-7% slower IPC with K8, then they fucked it up with Phenom, and not just refining the existing design.


The best car analogy to the new "FX" is the old 4-6-8 GM engines, they were never quite sure which cylinder to fire, so either they fired them all, or none. Good idea, but one about 20 years too soon for the actual production. Same way here, lets step back and make a bold statement.

90% of users would rather have a 6Ghz high performance dual core.


Wait, what....... most software is only written for 1-2 threads? Well, we are onto something here. By dammed, instead of changing the reads for the cars, lets change the cars for the road.

 

[Super XP]

But aren't Intel's i7 direct descendants of the original Pentium Pro of 12 years ago?

You mean from the Pentium 3. Don't know why Intel released the crapy Pentium 4 when the Pentium 3 was 100x much better. Then Intel finally scrapped the P4, took the P3 and with innovation created the Conroe, and now we have the i7's