Welcome Guest ( Log In | Register )

Outline · [ Standard ] · Linear+

 AMD Bulldozer & Bobcat

views
     
haris
post Oct 13 2011, 09:11 PM

Getting Started
**
Junior Member
172 posts

Joined: Dec 2005
From: KL


I think all this shouting and face palming over benchmark result is due to misunderstanding and over expectation.

1. This is like a reboot of CPU architecture for AMD. Is like 1 step of many to reposition the CPU by de emphasize FPU performance. Because at the end of the road Bulldozer architecture will become one of the Fusion. I.e more emphasis on floating point calculation on the GPU.

2. That is why AMD has 4 year roadmap. And the fusion thing should happen on next year Piledriver roadmap. It sucks this year on gaming benchmark, so much pitty for early adopters. As always!

3. Bulldozer, being the first in the roadmap also means software support is not here yet. Anand tech review bring up some thread scheduling issue on the different cores of Bulldozer with Windows 7. Which should be solved by Windows 8, next year. Support in Linux kernel should be sooner (just my opinion) since Linux is popular as server.

4. Which bring this to my next point. AMD need to "cari makan" also right? With Intel being so powerful in the main stream performance CPU market. Will be very high risk for AMD to go head to head with Intel in that particular market. If "untung sabut" then everything is good. If "untung batu". How then? So this time around (again this is my opinion) they put a CPU that will sell good at that price to the servers market. Hedging that server market (with premium margin) will support further effort to tune the architecture for Desktop in the next iteration.

5. Untill then, looks like AMD trying to promote the Bulldozer to enthusias as over clocking friendly CPU. Really have no idea whether this is good or not. But personally I won't have some thing that hot and eats lots of electricity in my house!

So I hope what I wrote above can give you guys a better perspective. Do understand that there is a lot of disappointment (me too!). But this has to happen before AMD really can get competitive again.

Unfortunately this means it sucks to be early adopter OR it is another wait. But we have a similar wait with Phenom -> Phenom II right? So what is another year.

Cheers!
biggrin.gif



haris
post Oct 13 2011, 09:47 PM

Getting Started
**
Junior Member
172 posts

Joined: Dec 2005
From: KL


As long as price per performance is okay. Which is what we are seeing how AMD is competing nowadays.

For sure new technology from Intel won't be cheap. At least we know Bulldozer is around USD250 today. Next year will perhaps go down to sub USD200.

Definitely there will be people who will go for top performance no matter what is the cost. And I salute those guys with deeps pocket for driving the technology up. While letting us normal people do the exact same thing with cost effective processor.


haris
post Oct 14 2011, 10:21 AM

Getting Started
**
Junior Member
172 posts

Joined: Dec 2005
From: KL


QUOTE(billytong @ Oct 14 2011, 08:21 AM)
Price/performance? It seems Intel are better now.  With i5 2500K selling just above RM600. IMO, if AMD wanna hit the price/performance mark, + given their shit power consumption, FX-8150 should have sell @ Rm500+ not RM900+.
*
Ah! taking my post out of context much?

I was replying to TristanX about incoming intel new processor arch and not comparing to the current midlle or top performing Intel processor.

Definitely AMD Bulldozer is not doing well at the moment. You can read my earlier post with 5 point underlining the current issue.

IMHO, we should look at this situation objectively without much frothing in the mouth. A non competitive AMD also means that Intel will monopolize the x86 processor market once again. Perhaps we will see another US$1000 processor soon. Perhaps than the people with mouth frothing now will sober up a little bit or probably have mouth frothing issue with Intel CPU pricing?

Anyway, the CPU market is not a charity. Hence it is not a good idea to push consumer to buy AMD when AMD is not performing. Let's not reward them (referring to AMD) when they screwed up.

1. If you (as PC enthusiast or the master geek of your community) need to buy a CPU in the short term. By all means, buy Intel. As their price and performance ratio is the best at the moment.

2. If someone ask for recommendation for short term purchase, recommend the same.

3. But a bit of education or being informative is good. Just to let people know what is the current situation with regards to Intel and AMD competition in the space. As well as letting people know the direction of AMD new Architecture (i.e. shooting for multiple core performance while not so hot on single core performance and the fusion stuff next year which should improve things).

In respond to other comment about seller being dishonest. Well nothing much we can do about that right? Seller are always selling to make quick bucks. The only defence for that is education (i.e. consumer not supposed to just walk in and buy stuff without doing any research). Hence my 3rd point above. I know that much of us here in LYN are supposed to be reference point of people around us. So do our best ler.

Cheers!
blink.gif

---
Edit: Just to say that at this point I am so damn jealous to see so many stars on other poster biggrin.gif. I am long time user of LYN myself but never being chatty so much in the forum. And just post whenever there is a reason. Like this debacle with AMD Bulldozer. I know that everyone here is being fair. And we all want to see a verry competitive and high performing CPU market. But hey, a little bit flame here and there add spice to discussion!

This post has been edited by haris: Oct 14 2011, 10:40 AM
haris
post Oct 14 2011, 11:13 AM

Getting Started
**
Junior Member
172 posts

Joined: Dec 2005
From: KL


QUOTE(haris @ Oct 13 2011, 09:11 PM)
I think all this shouting and face palming over benchmark result is due to misunderstanding and over expectation.

1. This is like a reboot of CPU architecture for AMD. Is like 1 step of many to reposition the CPU by de emphasize FPU performance. Because at the end of the road Bulldozer architecture will become one of the Fusion. I.e more emphasis on floating point calculation on the GPU.

2. That is why AMD has 4 year roadmap. And the fusion thing should happen on next year Piledriver roadmap. It sucks this year on gaming benchmark, so much pitty for early adopters. As always!

3. Bulldozer, being the first in the roadmap also means software support is not here yet. Anand tech review bring up some thread scheduling issue on the different cores of Bulldozer with Windows 7. Which should be solved by Windows 8, next year. Support in Linux kernel should be sooner (just my opinion) since Linux is popular as server.

4. Which bring this to my next point. AMD need to "cari makan" also right? With Intel being so powerful in the main stream performance CPU market. Will be very high risk for AMD to go head to head with Intel in that particular market.  If "untung sabut" then everything is good. If "untung batu". How then? So this time around (again this is my opinion) they put a CPU that will sell good at that price to the servers market. Hedging that server market (with premium margin) will support further effort to tune the architecture for Desktop in the next iteration.

5. Untill then, looks like AMD trying to promote the Bulldozer to enthusias as over clocking friendly CPU. Really have no idea whether this is good or not. But personally I won't have some thing that hot and eats lots of electricity in my house!

So I hope what I wrote above can give you guys a better perspective. Do understand that there is a lot of disappointment (me too!). But this has to happen before AMD really can get competitive again.

Unfortunately this means it sucks to be early adopter OR it is another wait. But we have a similar wait with Phenom -> Phenom II right? So what is another year.

Cheers!
biggrin.gif
*
Sorry for replying my own post. Just found more technical information that might give more clues on the state of Bulldozer today. Probably this will be my last post about the matter.

Quoted from ArsTechnica user comment

QUOTE
A poster at
http://www.xtremesystems.org/forums/showth...1-Threaded-Perfhttp://www.xtremesystems.org/forums/sho ... eaded-Perf

showed large gains in performance when disabling the second integer core in each module. This indicates that one major problem is that Windows is not efficiently assigning threads leading to cache thrashing. The review site Anandtech showed as much as a 10% performance improvement with the (still alpha) pre-release Windows 8 OS. Anand's N x N queen test also shows that Bulldozer has worse branch prediction than the previous generation (branch prediction is even more important for Bulldozer because it has a longer pipeline). Considering that AMD won't talk about branch prediction, I assume that pipeline flushing is a major problem and likely the one being corrected with piledriver. Further, the fact that piledriver is being released so soon, I believe that AMD already knew about performance problems, but instead counted on clockspeed to make up the difference until the fixed design could be released (given that major architecture changes take years to design rather than months). There have also been hints that maybe the decode unit isn't wide enough.

Anand also states that cache latencies are terrible (between 25 and 125% slower than Deneb or Sandybridge processors). Target speeds for Bulldozer were supposed to be 30% faster than Deneb. Both of these problems can be blamed on Globalfoundries poor 32nm process. Cache latencies can be inceased and clockspeeds lowered to give higher yields (AMD has already said that Globalfoundries is responsible for poor Llanos production).

It is likely that AMD saw the branch prediction (and possibly decoder) needed to be reworked, but decided that the improved clockspeed (because Bulldozer was designed to give as close to linear clockspeed performance increases instead of the tapered performance seen in Sandybridge) would be sufficient, but then a few months later (and too late to stop) they found that yields were too poor, thus a sub-par chip was launched.

 

Change to:
| Lo-Fi Version
0.0447sec    0.67    7 queries    GZIP Disabled
Time is now: 28th November 2025 - 10:23 AM