• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

Bulldozer and the future

Associate
Joined
19 Mar 2010
Posts
66
First thing let me state that Bulldozer/Zambezi is a terrible CPU for desktop use. It was a marketing mistake to name it as FX and also to push the "8 core" thing.

But..Is it really an awful architecture full stop? I've read posts saying AMD is dead and people linking the 10% staff cuts to the failure of the FX cpus.

From what I can see it is going to be a fantastic server chip. This then leads to the question, what next for desktops, given how bad Zambezi is? The anwer I believe is the Fusion/APU path.

The next big architectural change for AMD is to change the memory addressing of their GPUs to x86. This is a huge shift and will allow the GPU to address the same memory as the CPU. The direction that computing seems to be taking is a combination of a few heavy hitting cores teamed up with many light-wieght cores.

The APU is this realised, we need to stop thinking of GPU cores as graphic processors but more as vector processors that can be used for graphic processing.

Intel are aproaching this from the angle of reducing an x86 core and the stacking them up - Knights Corner (AFAIK) which was the Larabee project. AMD is coming at it from the other side and bring GPUs into the x86 space.

So Bulldozer fits in with this picture in the long term. It will require a massive change in operating systems and compilers to leverage the APU concept.

AMD wanted to be seen to keep a competetive foot in the current CPU centric desktop. They've played a marketing game and tried to create a "halo" product that reminds people they are there as an alternative. Unfortunately they badged the wrong with the FX badge.

tl;dr I don't think Bulldozer is a failure and the end of AMD but it was a marketing disaster.

Commence flaming!

references:
http://nextbigfuture.com/2011/08/amds-next-generation-gpu-architecture.html

http://en.wikipedia.org/wiki/Intel_MIC

http://blogs.amd.com/fusion/2011/03/01/exposing-the-phantom-x86-bottleneck/
 
Bulldozer is not a marketing failure, 8 cores and 4.2Ghz (yes I know that's not strictly accurate) is great for marketing, I mean how could it possibly be slower than Intel's 3.3Ghz quad core 2500K? ;)

Anyway theres a few things which will make Bulldozer much faster in the future; schedulers (Windows 8), higher thread counts (with most game engines supporting mobiles and consoles which need high thread counts to be fast, this will trickle down (or up :p) to PC games, plus other software is getting to use many threads as well), production tweaks (B3, higher clocks and lower power), architecture tweaks (Piledriver), motherboard tweaks (most of the reviews used a CHIVF which was shown to be significantly slower in some cases compared to a UD5), and the increasing use of GPGPU (reduces the bottleneck of only having four FP cores). And maybe some other stuff i've forgotten :p

But ATM for 90% of users it's pretty bad, which is why I've gone for a 2500K...
 
Last edited:
I think AMD are onto something with the APU for me that is where AMD should be investing their cash instead of trying to wrestle the performance crown away from Intel for pure vanity (it's just not required).

Personally I would like to see a tweaked Phenom II X4 APU on 32nm, unless AMD know something that we don't about future chips they should bin Bulldozer completely as its too power hungry and too large to manufacturer cheaply enough to sell at a competitive price.
 
Personally I would like to see a tweaked Phenom II X4 APU on 32nm, unless AMD know something that we don't about future chips they should bin Bulldozer completely as its too power hungry and too large to manufacturer cheaply enough to sell at a competitive price.

They already have, it's called Llano (well it's missing the L3 cache, but the cores are tweaked Phenom II cores). The bulldozer architecture is not too large or power hungry, the only reason Zambezi is power hungry is because of TSMC's (um, I mean GloFo's) 32nm process and the only reason it's large is the 16MB L3 cache. Trinity (aka Llanos successor) runs Piledriver, which is a tweaked Bulldozer, and looks better than Llano in every way.
 
Last edited:
People saying that AMD is dead and other nonsense drivel like that don't know what they're talking about.

The architecture is good.... the reason why it had a pretty lackluster launch was because of lack of optimizations, a couple poor design decisions (like the L1 cache and latency) and Global Foundaries inability to get their 32nm node down. These things can easily be fixed for Piledriver and future chips.

They already have, it's called Llano (well it's missing the L3 cache, but the cores are tweaked Phenom II cores). The bulldozer architecture is not too large or power hungry, the only reason Zambezi is power hungry is because of TSMC's 32nm process and the only reason it's large is the 16MB L3 cache. Trinity (aka Llanos successor) runs Piledriver, which is a tweaked Bulldozer, and looks better than Llano in every way.

They're using GloFo's 32nm process.
 
I think AMD are onto something with the APU for me that is where AMD should be investing their cash instead of trying to wrestle the performance crown away from Intel for pure vanity (it's just not required).

Personally I would like to see a tweaked Phenom II X4 APU on 32nm, unless AMD know something that we don't about future chips they should bin Bulldozer completely as its too power hungry and too large to manufacturer cheaply enough to sell at a competitive price.

Firstly Llano is a tweaked Phenom 2 on 32nm, and its a dead end.

Secondly Bulldozer is DESIGNED TO BE AN APU ARCHITECTURE, its an architecture that has had many many changes to it and a LOT of them are in mind for APU's, the first APU based on the second gen of Bulldozer is seemingly going to launch in January.

You're also completely wrong about being power hungry.

Sandybridge is a VERY efficient, much later generation version of an architecture, Bulldozer is the FIRST generation of a new architecture, it has a LOT of room to grow into, the first generation of a new architecture is rarely particularly efficient. Anyway, Sandy bridge is a just under 1 billion transistor chip on a very mature 32nm process, and is rated to 95W at what 230ish mm2 in size. Bulldozer is a 2 billion transistor chip, rated at 125W, at 315mm2, on a very immature process. Sorry but Sandy bridge with 2.27billion transistors, which is what SB-E is, is both rated for higher power and significantly larger than Bulldozer.

Power wise, Bulldozer is very good, which I said from day one. Performance per W isn't great, overall power usage for a 2billion transistor chip on a very new process, is pretty god damned good.


The staff cuts have smeg all to do with Bulldozer, that is a fact, the next several generations including Trinity which is almost done, which has second gen Bulldozer cores, will not be cut.

Desktop performance is NOT horrible with a Bulldozer, anyone who says so, is daft, its as simple as that. Again I'll point out Anandtech's review, which isn't glowing, the page with only actual windows applications being tested. Bulldozer beats the 2500k in over half of them, and beats the 2600k in 3-4 of them. For the first generation of a chip, that is both incredibly good, and also would lead to it being truly impossible saying it would be horrible to use.

Its horrible in synthetic benchmarks...... no one runs synthetic benchmarks day in and day out, in real use bulldozer is significantly more competitive with a 2500/2600k than in synthetic benchmarks, I haven't seen a review that doesn't show this.
 
Does ARM moving into the desktop / x86 space have any impact here?

I would imagine that they will squeeze AMD before they squeeze intel. Anyone have opinions on what will happen when / if a third company enters the fray?
 
As much as I hate Intel I do have to admit AMD have some challenges they face ahead. in order to 'optimise' applications for anything they produce they have to get more developers on board, provide dedicated compilers etc etc. The way it currently stands (unfortunately) is that intel are holding all the cards - It's their compilers and optimisations that are being used most, hence why AMD will have to work hard to introduce anything new (i.e APU's) into the market.

come on AMD - we still love you! :)
 
However, it does lag behind in games.

The majority of which are GPU limited until you get to two cards, in which case money is obviously no object and the debate unnecessary.

Judging by the talk around here you'd think this was the P4 Celeron or something. Not that I think BD is spectacular or anything, but we have to put its performance in context.
 
Firstly Llano is a tweaked Phenom 2 on 32nm, and its a dead end.

Secondly Bulldozer is DESIGNED TO BE AN APU ARCHITECTURE, its an architecture that has had many many changes to it and a LOT of them are in mind for APU's, the first APU based on the second gen of Bulldozer is seemingly going to launch in January.

You're also completely wrong about being power hungry.


I thought Llano was an Athlon II hence the lack of L3 cache. :confused:

As for a power hungry Bulldozer I could have sworn I read the top model which is performance competitive with a I5 2500K is rated by AMD at 125 watts which put aside a 95 watt TDP 2500K is 30% more power, is that wrong?
 
Back
Top Bottom