Just installed a AMD FX 8350. My thoughts!!!

[inf64]

Number of ALUs don't define a core. But the fact that it has 4 FP units(2 threads each) has a lot to do with it performing like i7(QC with SMT). If it had beefier FP unit it would have performed much better in many workloads(not that i7 level of performance is bad though). The problem is that even with 4 of these FP "cores" and 8 integer cores(yes they are "true" cores) AMD is hitting the TDP ceiling on 32nm node. Proves that GloFo's process is a complete turd.

 

[Idontcare]

Core count doesn't matter, they could have one core or 24 cores and it wouldn't matter. The problem is single-threaded performance, hands down, followed by performance/watt for both single and multi threaded apps.

If you look at the negativity associated with bulldozer it is not about price or multi-threaded performance, it is about power consumption and single-threaded performance.

The irony is that this is exactly what Intel addressed in going from Prescott to Conroe in 2006. Intel boosted single-threaded performance and lowered power consumption...and bam, stole the mindshare and the market.

Six years later AMD attempts to reverse that lesson learned WTF AMD?

The real irony though is something that few people probably have the background to appreciate, and that is the fact AMD sunk >$5B into a strategy that was supposed to be the future (fusion) in which the ridiculously parallelizable codebase was suppose to be running on the iGPU portion of the APU anyways, meaning the plan back then was to keep just a few x86 cores around to do the heavy-lifting with single-threaded and difficult-to-parallelize code while porting the easily parallelized stuff over to the fusion part of the core.

And yet they went and abandoned that plan almost after day (if we run the clock backwards to the timeframe when bulldozer itself was being roughed out on a drawing board) as they decided all the way back then that they would pursue a weaker-IPC but moar-coars approach even for the x86 side of the APU

So again, WTF AMD? Even the marketing guys were left wondering "WTF AMD?" as the only thing that made sense to them was the lie they were told which was that IPC was going to increase.

 

[mrmt]

So again, WTF AMD? Even the marketing guys were left wondering "WTF AMD?" as the only thing that made sense to them was the lie they were told which was that IPC was going to increase.

Were their market guys feed with wrong information or did they get something wrong from engineering presentations?

 

[Sheep221]

Number of ALUs don't define a core. But the fact that it has 4 FP units(2 threads each) has a lot to do with it performing like i7(QC with SMT). If it had beefier FP unit it would have performed much better in many workloads(not that i7 level of performance is bad though). The problem is that even with 4 of these FP "cores" and 8 integer cores(yes they are "true" cores) AMD is hitting the TDP ceiling on 32nm node. Proves that GloFo's process is a complete turd.

GF is spinoff company, the FABs and technology they have was built by AMD before, so it's not necessarily a fault of GF, it's likely to be bad design rather than manufacturing. You are right the FP thing tho

 

[Idontcare]

Were their market guys feed with wrong information or did they get something wrong from engineering presentations?

Assuming JFAMD wasn't just straight-up lying (I beleive he was the one who was lied to), he claimed to have had multiple interactions with engineers, including follow-up conversations solely for the purpose of ensuring he understand what it was they were telling him, and yet somehow he was still misled to believe/report/publicize that IPC would not decrease with bulldozer.

Considering that it was right near the same timeframe that AMD was getting bulldozer silicon back from the fab that the BoD decided it was time to fire Dirk, I get the distinct impression that JFAMD wasn't the only person within AMD who was misled for a good number of years on the pre-release hype of bulldozer.

 

[Idontcare]

GF is spinoff company, the FABs and technology they have was built by AMD before, so it's not necessarily a fault of GF, it's likely to be bad design rather than manufacturing. You are right the FP thing tho

The thermals and power-consumption of 32nm Llano versus those of 45nm deneb compel me give credence to the argument that something is/was rotten in Denmark (no offense Shintai ) when talking about the process-node underpinnings of bulldozer.

 

[Sheep221]

Similar did intel with IB, using cheaper materials.

 

[mrmt]

Considering that it was right near the same timeframe that AMD was getting bulldozer silicon back from the fab that the BoD decided it was time to fire Dirk, I get the distinct impression that JFAMD wasn't the only person within AMD who was misled for a good number of years on the pre-release hype of bulldozer.

While I agree that AMD was fooled top down by the Bulldozer teams, I don't think Fruehe deserves too much leniency.

His discourse contradicting AMD own slides, his insulting manner to everyone that contested his "facts", even after Dirk was fired, his rebuttals of the leaked benchmarks, everything points out not for someone that was misled, but to someone that first could not grasp what engineering was really saying and after someone trying to lie his way out of the situation.

I think Scali gave a blunt but straight view on what JF-AMD really did.

http://scalibq.wordpress.com/2011/1...e-sensible-thing-and-comes-up-with-an-excuse/

 

[KompuKare]

The thermals and power-consumption of 32nm Llano versus those of 45nm deneb compel me give credence to the argument that something is/was rotten in Denmark (no offense Shintai ) when talking about the process-node underpinnings of bulldozer.

Ah, Dresden is over 500KM south of the Danish border (using Google Maps' route: Dresden to Rodbyhavn which looks like the closest point).

Edit Hm, ok I've been corrected. Classic prose on forums, whatever next? Now that I think of it, those words did sound familiar.

 

[ShintaiDK]

Ah, Dresden is over 500KM south of the Danish border (using Google Maps' route: Dresden to Rodbyhavn which looks like the closest point).

Its something that comes from Shakespeare and is said by Marcellus. And its actually means Hamlet who is a Prince of Denmark and not the country as such.

IDC forgot one word as well. Correct is: "Something is rotten in the state of Denmark".

 

[Idontcare]

Ah, Dresden is over 500KM south of the Danish border (using Google Maps' route: Dresden to Rodbyhavn which looks like the closest point).

Its an idiom meaning something doesn't add up.

something is rotten in (the state of) Denmark.
Prov. something suspicious is going on. (From shakespeare's play Hamlet.)

 

[frozentundra123456]

Core count doesn't matter, they could have one core or 24 cores and it wouldn't matter. The problem is single-threaded performance, hands down, followed by performance/watt for both single and multi threaded apps.

If you look at the negativity associated with bulldozer it is not about price or multi-threaded performance, it is about power consumption and single-threaded performance.

The irony is that this is exactly what Intel addressed in going from Prescott to Conroe in 2006. Intel boosted single-threaded performance and lowered power consumption...and bam, stole the mindshare and the market.

Six years later AMD attempts to reverse that lesson learned WTF AMD?

The real irony though is something that few people probably have the background to appreciate, and that is the fact AMD sunk >$5B into a strategy that was supposed to be the future (fusion) in which the ridiculously parallelizable codebase was suppose to be running on the iGPU portion of the APU anyways, meaning the plan back then was to keep just a few x86 cores around to do the heavy-lifting with single-threaded and difficult-to-parallelize code while porting the easily parallelized stuff over to the fusion part of the core.

And yet they went and abandoned that plan almost after day (if we run the clock backwards to the timeframe when bulldozer itself was being roughed out on a drawing board) as they decided all the way back then that they would pursue a weaker-IPC but moar-coars approach even for the x86 side of the APU

So again, WTF AMD? Even the marketing guys were left wondering "WTF AMD?" as the only thing that made sense to them was the lie they were told which was that IPC was going to increase.

I was a great fan of fusion initially. I thought, wow, you can game on an off the shelf system without upgrading it. Well, you can, sort of, but a 75.00 discrete card still is vasltly superior after all these years of development. Fusion was also late to be implemented and the igp never was quite good enough IMO to sufficiently compensate for the mediocre CPU performance. I remember being excited when fusion finally came out, only to be shocked that the cpu was not any kind of new architecture. They just got stuck in limbo of "jack of all trades and master of none". CPU performance was clearly inferior to intel, and the igp was more than needed for everyday use, but not enough for really heavy gpu tasks.

 

[Centauri]

Why are you condemning a product you know you've never owned?

 

[AnandThenMan]

Why are you condemning a product you know you've never owned?

It's the fanboy way.

 

[AtenRa]

Core count doesn't matter, they could have one core or 24 cores and it wouldn't matter. The problem is single-threaded performance, hands down, followed by performance/watt for both single and multi threaded apps.

Bulldozer was designed for MultiThreaded loads, core count and performance/watt. If you compare PileDriver to Core i7 Sandybridge (both second generation 32nm products) you will find that PD is faster in MT loads and performance/watt is acceptable(faster while consuming more).

http://www.anandtech.com/bench/Product/697?vs=287

I have said that before, the problem with AMD is they lack behind in manufacturing not in design.

If you look at the negativity associated with bulldozer it is not about price or multi-threaded performance, it is about power consumption and single-threaded performance.

The irony is that this is exactly what Intel addressed in going from Prescott to Conroe in 2006. Intel boosted single-threaded performance and lowered power consumption...and bam, stole the mindshare and the market.

Six years later AMD attempts to reverse that lesson learned WTF AMD?

Yes, that is because after all that time, since the first dual core Desktop CPU, desktop workloads just starting to be Multithreaded. We where stagnated with a 10% IPC increase per year and 15-20% increase in performance due to being oriented to single threaded increases.

Im not against the single thread increases but everyone knows that is the most difficult thing to do. It is very easy to come to negative returns trying to get even 1% more IPC.

That was one of the most important reasons to go MultiCore and MultiThreaded.

The real irony though is something that few people probably have the background to appreciate, and that is the fact AMD sunk >$5B into a strategy that was supposed to be the future (fusion) in which the ridiculously parallelizable codebase was suppose to be running on the iGPU portion of the APU anyways, meaning the plan back then was to keep just a few x86 cores around to do the heavy-lifting with single-threaded and difficult-to-parallelize code while porting the easily parallelized stuff over to the fusion part of the core.

And yet they went and abandoned that plan almost after day (if we run the clock backwards to the timeframe when bulldozer itself was being roughed out on a drawing board) as they decided all the way back then that they would pursue a weaker-IPC but moar-coars approach even for the x86 side of the APU

The FUSION through the HSA is still their future and they are committed to it. The only problem i see here is the fact that it takes more time for this to give them what they were targeting for.
Yes, they spend 5-bill and this investment haven't payed off as of yet, but i believe we see the first steps in to the near future with Llano and Trinity.

So again, WTF AMD? Even the marketing guys were left wondering "WTF AMD?" as the only thing that made sense to them was the lie they were told which was that IPC was going to increase.

I dont want to play the IPC advocate for the BD but as you and I and a lot of us know here, IPC is Application Driven. You can have an IPC increase in one application and a decrease or not in another.
Granted IPC have fallen in the majority of desktop Legacy code with Bulldozer but it was increased in SIMD and others.

 

[frozentundra123456]

Bulldozer was designed for MultiThreaded loads, core count and performance/watt. If you compare PileDriver to Core i7 Sandybridge (both second generation 32nm products) you will find that PD is faster in MT loads and performance/watt is acceptable(faster while consuming more).

http://www.anandtech.com/bench/Product/697?vs=287

I have said that before, the problem with AMD is they lack behind in manufacturing not in design.





Yes, that is because after all that time, since the first dual core Desktop CPU, desktop workloads just starting to be Multithreaded. We where stagnated with a 10% IPC increase per year and 15-20% increase in performance due to being oriented to single threaded increases.

Im not against the single thread increases but everyone knows that is the most difficult thing to do. It is very easy to come to negative returns trying to get even 1% more IPC.

That was one of the most important reasons to go MultiCore and MultiThreaded.



The FUSION through the HSA is still their future and they are committed to it. The only problem i see here is the fact that it takes more time for this to give them what they were targeting for.
Yes, they spend 5-bill and this investment haven't payed off as of yet, but i believe we see the first steps in to the near future with Llano and Trinity.





I dont want to play the IPC advocate for the BD but as you and I and a lot of us know here, IPC is Application Driven. You can have an IPC increase in one application and a decrease or not in another.
Granted IPC have fallen in the majority of desktop Legacy code with Bulldozer but it was increased in SIMD and others.

If I counted correctly, sb was faster in 17 of the 25 non gaming benchmarks and of course faster in every gaming benchmark. Not a very good showing for amd in my book.

 

[AtenRa]

If I counted correctly, sb was faster in 17 of the 25 non gaming benchmarks and of course faster in every gaming benchmark. Not a very good showing for amd in my book.

Take off the Synthetics and you have a draw in real applications(FX faster in the majority of MT apps).

 

[mrmt]

I have said that before, the problem with AMD is they lack behind in manufacturing not in design

Maybe you can tell us what secret sauce they can use in order to sell 315mm^2 chips for 110USD and still make money.

 

[tulx]

Core count doesn't matter, they could have one core or 24 cores and it wouldn't matter. The problem is single-threaded performance, hands down, followed by performance/watt for both single and multi threaded apps.

While I agree that single-threaded performance is very important in gaming (though with a downward trend), as a personal user, I really couldn't care less about performance/watt. Even an OC'd 8150 with SLi GTX 480's would consume less power than some other house appliances. Moreover, the electricity bill is (at least where I live) the least significant one. I spend much more money for heating or diesel. Even if my current electricity bill doubled, I wouldn't really notice it much in the monthly balance.

Performance/watt is definitely a factor for a server farm or maybe an internet-cafe owner (should those still exist), but not for a private PC gamer with two machines at home.

 

[inf64]

Maybe you can tell us what secret sauce they can use in order to sell 315mm^2 chips for 110USD and still make money.

Because it costs AMD around 50$ per die to make it.

 

[mrmt]

Because it costs AMD around 50$ per die to make it.

Probably more, but even if you are righr 50 usd isn't near enough to cover shipment, insurance and the retailer margin.

That's why they got 38% gross margins last quarter

 

[AtenRa]

Maybe you can tell us what secret sauce they can use in order to sell 315mm^2 chips for 110USD and still make money.

They were selling Phenom II X6 346mm2 dies for 110 USD and made a profit, why they dont with a smaller die at a cheaper manufacturing process ??? 32nm SOI HKMG Gate First is 10-20% cheaper.

You cant directly compare Intel against AMD in manufacturing, they are so different in so many ways. Different manufacturing processes, different OPerational EXpenses and more.

 

[mrmt]

They were selling Phenom II X6 346mm2 dies for 110 USD and made a profit, why they dont with a smaller die at a cheaper manufacturing process ??? 32nm SOI HKMG Gate First is 10-20% cheaper.

You cant directly compare Intel against AMD in manufacturing, they are so different in so many ways. Different manufacturing processes, different OPerational EXpenses and more.

Phenom X6 was always a very small part of AMD product mix, bulldozer is not.

I'd like you to back up your claim. Glf 32nm is 10% cheaper than what?

Third I'm not comparing manufacturing proccesses, but talking about AMD own financial statements

 

[mrmt]

Btw, I made a mistake when reading AMD numbers. 38% wasn't the gross margins for the CPU business but for the entire company, which means that the CPU group gross margins should be even lower, some 35%-36% before the Llano markdown and around 28% after.

Atenra, Inf64 and other Bulldozer fans will have a tough job explaining how such a marvelous, efficient and competitive architecture is generating these atrocious margins.

 

[guskline]

Funny, how FX 8350 just came out and the barrage of naysayers continues. I HAVE an FX 8350 and I HAVE a FX8150 (which is for sale since I now own the 8350). I have tried to honestly compare the 8150 to the 2500k. I said the 8150 was behind the 2500k but OC'd made the gap close. Now the 8350 is overall better and neck and neck with a 3570k. Just my opinion, but hey what do I know! I ownly own 2 2500k rigs and a FX 8350 rig. Intel is ahead in single threaded aps but the margin really narrows for multi threaded aps. Vishera isn't the golden egg to put AMD ahead but it makes a respectable showing, addressing the Bulldozer.

BTW, I still marvel at the speed of my 2500ks in gaming considering their price. At least AMD has given an option with the Vishera.