September 22nd ETA for AMD FX processors

Page 13 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

statikregimen

Junior Member
Sep 19, 2011
4
0
0
AMD invited a bunch of journalists to there home office in TX on a preview of bulldozer.

Amungst the people who were invited, i know a handful.

They went to the invite and toured the facility.
Also as a bonus they got to check out an attempt breaking the WR.

They accomplished the GHZ record, but thats all.

Im not allowed to speak into details, but i know enough about the cpu to make some choices.

I know they run a lot cooler then the PH2 = bonus.
They all overclock like crazy = bonus.

But at the end of the day, the guys there were going WOW, so they broke the world record... but how come nothing else followed from it?

No CB, no Wprime, no sysmark, no passmark, no 3dmark even...

True given Wprime is intel favored... but Cinebench, and 3dMark.. :whiste:

If im not mistaken a 3dmark is not favored for either intel or AMD.
And AMD has always had a good run on 3dMark06.

At the very least im wondering if BD can beat the PH2 scores which are currently out there @ LN2 cooling.

Thanks for this....I was aware of the event, just not that they didn't show any benchmaks nor there was any public reaction from anyone who attended (figured it was all under NDA). Either way, my opinion remains with those who think it is best we just wait and see. I certainly will not buy one if it is slower than my 1100t, which I really don't see that happening. It certainly can't be too bad, if they're moving all of their upcoming products to BD modules. Would be corporate suicide, if performance is outright pathetic.

Additionally, it is a totally new architecture and with that, sometimes requires some OS-level (and compiler; probably bios, too) optimization, to truly realize the potential. I feel that BD architecture will take a healthy 3-6 months before things start looking good for it. However, due to everybody's high hopes, AMD's aggressive marketing and constant delays, I fear if it comes out of the gates sluggishly, it will falter before it has a chance to prove itself.
 

krumme

Diamond Member
Oct 9, 2009
5,956
1,596
136
OK, enough crap has been thrown at Andy Glew for one day! Here's a post that should enlighten you as to just how ridiculous this reaction is. I have to say, you have a very naive view of the world if that sounded unprofessional to you. That's the voice of experience talking, and it sounds reminiscent and somewhat regretful, but not at all the attitude of a broken man. If your children cannot recognize the validity of his statements, perhaps they should be taught that the way to avoid regrets is not to play it safe, but to take risks with their life.

Andy Glew did his undergraduate research on out-of-order microarchitectures. His name is on several patents involving register renaming, which is in use by every out-of-order CPU existent today. He used this expertise while working at Intel 1991-1996 on the P6 architecture. This became the first OOO x86 processor ever, and it went into the Pentium Pro and started Intel's foothold in the server market.

In 1996, after helping to start Intel's Microprocessor Research Lab, Andy left Intel to go back to grad school. After the birth of his daughter in 1999, he quit grad school to return to Intel, where he worked on virtual machines and speculative multithreading, and also contributed to the Willamette processor, the first Pentium 4. After some frustration with the work environment at Intel, he left to work for AMD 2002-2004 (some of their best years), where he contributed the fundamental ideas behind Bulldozer, multithreading and multiclustering. But he found the work environment at AMD not much better than at Intel, and left in June 2004 to write a book on computer architecture (http://www.pvcmuseum.com/cpu/andy-glew-proposed-amd-k10-architecture.htm ).

After 2005, he ended up back at Intel (rehires at Intel are reportedly very rare... he's been taken back twice!) working on security, supercomputers, and graphics. Inventing was in his blood though, and in 2009 he left Intel yet again to join a company called Intellectual Ventures. Recreating Thomas Edison's invention factory was a lifelong dream of his, and he has nearly 100 patents in his name, most related to computer architecture, but with some nifty off-the-wall randomness, like the one on "System providing video compression/encoding for communications across a network". Also known as, the webcam!

So that's where he is now. Pretty impressive CV, if I do say so myself. But evidently, he said one lukewarm thing about Bulldozer, and all that life history suddenly means nothing. People who cannot plausibly counter the truth of his statements now find all manner of fault with his credibility, casting aspersions on his character and the manner of his conduct. But his opinion still stands, and I think his opinion holds a thousand times the weight of the opinions of those that seek to discredit him.

You dont have to lecture me. And Andy dont have to lecture anybody any more. Fortunately.

I dont care if he wrote on the subject of BD or SB, and I dont care if he worked for Intel or AMD. His letter is a long spewing of bitterness and anger against others. Its disrespectfull. And it sadly just show the grim picture behind the nice facade of an brilliant history and achiewements. He would have been better off, not writing this stuff. What a short satisfaction for the long term bitterness that will haunt him. What a waste to write such a bad story about oneselves. Sad.

And the point is then for this thread: Taking this as any evidence for BD performance is nonsense and wrong.

When we see Johans review of BD here at anand, we will know if, and to what degree, BD achieved its goal.
 

krumme

Diamond Member
Oct 9, 2009
5,956
1,596
136
Can someone say something more about the the L1 & L2 BTB? - what do we know? - and what can be the consequences?
 

gramboh

Platinum Member
May 3, 2003
2,207
0
0
AMD invited a bunch of journalists to there home office in TX on a preview of bulldozer.

Amungst the people who were invited, i know a handful.

They went to the invite and toured the facility.
Also as a bonus they got to check out an attempt breaking the WR.

They accomplished the GHZ record, but thats all.

Im not allowed to speak into details, but i know enough about the cpu to make some choices.

I know they run a lot cooler then the PH2 = bonus.
They all overclock like crazy = bonus.

But at the end of the day, the guys there were going WOW, so they broke the world record... but how come nothing else followed from it?

No CB, no Wprime, no sysmark, no passmark, no 3dmark even...

True given Wprime is intel favored... but Cinebench, and 3dMark.. :whiste:

If im not mistaken a 3dmark is not favored for either intel or AMD.
And AMD has always had a good run on 3dMark06.

At the very least im wondering if BD can beat the PH2 scores which are currently out there @ LN2 cooling.

Thanks for sharing this info, this speaks volumes about the performance of BD for most enthusiast type people posting on AT. Server CPU only!
 

intangir

Member
Jun 13, 2005
113
0
76
You dont have to lecture me.

Yes, yes I did. I was compelled to. You needed it. It seems it might not have benefited you much, but I could not let your abhorrent statements stand uncountered and rest easy.

His letter is a long spewing of bitterness and anger against others. Its disrespectfull. And it sadly just show the grim picture behind the nice facade of an brilliant history and achiewements. He would have been better off, not writing this stuff. What a short satisfaction for the long term bitterness that will haunt him. What a waste to write such a bad story about oneselves. Sad.

Absolute drivel. Everyone can read that passage for themselves. There was no bitterness towards others expressed in the passage quoted. The tone was most certainly not angry at all. If you read it in that passage, I have to chalk it down to unfamiliarity with the language, or bias on your part. Disrespectful? I respectfully disagree. It was written in quite sympathetic, considerate language. And no, he would not have been better off not commenting on these issues. I think it is much better to shine a light on the realities of the situation. Hiding the truth benefits nobody. Even taking the insights into the project's history out of consideration, the reactions to it form a valuable litmus test, I think, as to who is biased or not on these forums.

And the point is then for this thread: Taking this as any evidence for BD performance is nonsense and wrong.

When we see Johans review of BD here at anand, we will know if, and to what degree, BD achieved its goal.

Yes, time will tell. That was my favorite yearbook quote in high school, and I think it rings more true as the years roll by. Mark my words, time will tell.
 

Vesku

Diamond Member
Aug 25, 2005
3,743
28
86
I'm confident they will price BD well based on actual retail stock performance. Which means server cpu with OC capability at desktop prices. Now if they would only start selling them...
 

aigomorla

CPU, Cases&Cooling Mod PC Gaming Mod Elite Member
Super Moderator
Sep 28, 2005
21,065
3,572
126
I know you are sponsored Aigo and I sure don't want you to think I am disrespecting you, but I hope you can appreciate why I have a ton more respect for the at-home DIY bonafide enthusiast who is sponsoring himself when he decides to push his chip to 5GHz on dry ice or LN2 or some such versus the guy who may push his chip even higher but it isn't actually his chip to begin with and if it burns up in the process he's also not out $1k retail on it (not too mention the LN2 is free because its provided at the OC-fest and so on).

No i totally agree with you which is why i have no quams on the rules of no ES's on submissions.

Also its funny because when i first started getting into computers, the overclockers were young, very young... they were starting the field in a sense..

You goto these competitions now, and everyone is old... why? because only the old people can actually afford nice hardware to get noticed by such companies to get sponsored for such events. :p :biggrin:

But that being said... i hold no records minus how many people i got to migrate to water and post view count on my watercooling stickie!!

those are probably the only records i would want to keep on the PC side...
I like to guide... not lead... leading is such a pain because there is always someone out there who wants to stab u in the back.
 

Phynaz

Lifer
Mar 13, 2006
10,140
819
126
No one until release ():)
(I'm not going to buy it until I see the outlook, Zambezi, Vishera, and NG)

That not an answer. Once reviews are out, which one will you trust?

I suspect you can't answer until after you read them and pick the one that tells you what you want to hear.
 

LOL_Wut_Axel

Diamond Member
Mar 26, 2011
4,310
8
81
You are not talking about the same IPC.

He talks about the static theoretical IPC, you are talking about a derailed IPC of an application after the code has run. (i mention derlaied because you don't know the number of instructions it did and wether it is more or less in the program compared to others).

Deeper pipelines don't really affect theoretical ipc much. If nothing goes wrong (no certain branches) the deeper pipeline will only come a few cycles later in a good ooo environment.


I'm also not sure if we know for certain BD cores are that much weaker as you seem to imply. You have to see the core's with their design goals. So the core is running at higher frequencies, its the design goal. And i doubt SB will be 30% faster than BD in the uncommon and obsolete single threaded benchmark. When looking at the typical desktop workload between 2-4threads I don't see BD very far below SB (maybe 5-10%).

If it wasn't 30% faster in single-threaded AMD wouldn't have any need to make a CPU with four modules (eight integer cores) so they can make up for it in multi-threaded. If it had decent IPC, they'd limit it to three modules for desktop. Remember that AMD needs the highest ASP they can get, and having the smallest die size they possibly can will help in them getting more dies per wafer, too.

Unfortunately, everything keeps pointing towards Bulldozer not having a decent IPC increase from K10.5.

For pretty much all desktop workloads, single-threaded performance is important. The best example of why having a high-IPC, high-frequency Quad-Core over a low-IPC, high-frequency Eight-Core is that the Quad-Core will do well in all scenarios, whereas the Eight-Core will only truly shine on multi-threaded. If you're looking for a good video encoding, rendering, or compiling CPU, then I'd recommend looking forward to Bulldozer. Otherwise, no.

Also, your whole line on IPC makes no sense. It's more important to see IPC that's demonstrable in most CPU-agnostic applications running in single-threaded. Theoretical is one thing, but the real-world is another. For example, AMD said Llano had 6% higher IPC than Phenom II. In reality, that's 3-4%.
 
Last edited:

Idontcare

Elite Member
Oct 10, 1999
21,110
64
91
But that being said... i hold no records minus how many people i got to migrate to water and post view count on my watercooling stickie!!

those are probably the only records i would want to keep on the PC side...
I like to guide... not lead... leading is such a pain because there is always someone out there who wants to stab u in the back.

Props for that :thumbsup:
 

NostaSeronx

Diamond Member
Sep 18, 2011
3,811
1,289
136
Can someone say something more about the the L1 & L2 BTB? - what do we know? - and what can be the consequences?

Branch Prediction said:
To predict and accelerate branches, AMD Family 15h processors employ a combination of nextaddress
logic, a 2-level branch target buffer (BTB) for branch identification and direct target
prediction, a return address stack used for predicting return addresses, an indirect target predictor for
predicting indirect jump and call addresses, a hybrid branch predictor for predicting conditional
branch directions, and a fetch window tracking structure (BSR). Predicted-taken branches incur a 1-
cycle bubble in the branch prediction pipeline when they are predicted by the L1 BTB, and a 4-cycle
bubble in the case where they are predicted by the L2 BTB. The minimum branch misprediction
penalty is 20 cycles in the case of conditional and indirect branches and 15 cycles for unconditional
direct branches and returns.
The BTB is a tagged two-level set associative structure accessed using the fetch address of the current
window. Each BTB entry includes information about a branch and its target. The L1 BTB contains
128 sets of 4 ways for a total of 512 entries, while the L2 BTB has 1024 sets of 5 ways for a total of
5120 entries.
The hybrid branch predictor is used for predicting conditional branches. It consists of a global
predictor, a local predictor and a selector that tracks whether each branch is correlating better with the
global or local predictor. The selector and local predictor are indexed with a linear address hash. The
global predictor is accessed via a 2-bit address hash and a 12-bit global history.
AMD Family 15h processors implement a separate 512- entry indirect target array used to predict
indirect branches with multiple dynamic targets.
In addition, the processors implement a 24-entry return address stack to predict return addresses from
a near or far call. Most of the time, as calls are fetched, the next return address is pushed onto the
return stack and subsequent returns pop a predicted return address off the top of the stack. However,
mispredictions sometimes arise during speculative execution. Mechanisms exist to restore the stack to
a consistent state after these mispredictions.

Not sure if this is what you are wanting

Also, your whole line on IPC makes no sense. It's more important to see IPC that's demonstrable in most CPU-agnostic applications running in single-threaded. Theoretical is one thing, but the real-world is another. For example, AMD said Llano had 6% higher IPC than Phenom II. In reality, that's 3-4%.

You are as slow as the slowest thread in SMP and Llano's CPU Math Capabilities doubled~ or so(225 to 500 per core)<-- not that big

So, if you have macro op fusion on Bulldozer where you didn't in Stars, theoretical IPC based on Functional Units is the max possible IPC so you have 2 ALUs/2 AGUs + any gains you get from Macro-op fusion

Multithreaded Applications that use Symmetric multiprocessing have dependent threads and independent threads OoO helps but it is really dedicated by the functional units and what feeds it and the clock rate(IPS measurement)

If you have large IPC and poor OoO and poor clockrate and threads are really light and long you are going to have issue
Bulldozer has average IPC with a improved OoO and improved Clockrate headroom and threads can be light or big and you can handle just fine

That not an answer. Once reviews are out, which one will you trust?

I suspect you can't answer until after you read them and pick the one that tells you what you want to hear.

I would look for my applications and if in my applications Bulldozer is an improvement I'll make a plan on if I should wait for a later stepping, if I should buy now or if I should wait for revision on the architecture level if there is no improvements in my applications then I would wait until a CPU achieves my certain expectations

It is less that I trust a reviewer/benchmark more if they have a proven reliability
 
Last edited:

beginner99

Diamond Member
Jun 2, 2009
5,315
1,762
136
J
I don't think i would advice people to buy a single Core Zacate @ 1.5Ghz above a dual core Atom @ 1.8Ghz with HT.

Yeah and I said beefy single core. Do you think Zacate is a beefy single core? No, it is almost as bad as atom.
The only beefy single core available is the new SB single core Pentium. And I would sure choose that over a dual core atom.

Just look at the new AMD switchable graphics review and there at the Starcraft 2 Benchmark. AMD APU just sucks so bad and the only reason for that is it's meager IPC / single-threaded performance. (the APU has 4 cores compared to the 2 of the i5...)

My take-away message:
single-core IPC matters a lot. I don't want to advocate single-core CPUs but I would always choose half the beefy cores over double "so-so" cores.
 

Riek

Senior member
Dec 16, 2008
409
15
76
Yeah and I said beefy single core. Do you think Zacate is a beefy single core? No, it is almost as bad as atom.
The only beefy single core available is the new SB single core Pentium. And I would sure choose that over a dual core atom.

Just look at the new AMD switchable graphics review and there at the Starcraft 2 Benchmark. AMD APU just sucks so bad and the only reason for that is it's meager IPC / single-threaded performance. (the APU has 4 cores compared to the 2 of the i5...)

My take-away message:
single-core IPC matters a lot. I don't want to advocate single-core CPUs but I would always choose half the beefy cores over double "so-so" cores.


Bobcat is a beefy core compared to Atom. (the difference between bobcat and Atom is larger than SB core vs thuban core)
If you belief the only beefy core is SB then you greatly misjudge SB or you have a strange concept of beefy core.

Comparing an SB core to Atom cores... yeah... they give the similar performance in 2-4 threaded scenarios? so how do you compare them then? I said performance between 2-4 threaded matters!!!
So if you have a atom serie that performs on average better between 2-4 threads than the SB you are comparing to, you are better of using that Atom structure. (useless discussion though.)
 

beginner99

Diamond Member
Jun 2, 2009
5,315
1,762
136
Bobcat is a beefy core compared to Atom. (the difference between bobcat and Atom is larger than SB core vs thuban core)

Bobact is a low power cpu just as atom. it is "slow by intention". Anyway there are both not even close to "beefy" in my terms hence atom vs SB

Comparing an SB core to Atom cores... yeah... they give the similar performance in 2-4 threaded scenarios? so how do you compare them then? I said performance between 2-4 threaded matters!!!
So if you have a atom serie that performs on average better between 2-4 threads than the SB you are comparing to, you are better of using that Atom structure. (useless discussion though.)
That's exactly were I disagree. If the CPU with more cores (but worse single core IPC) is barley faster in multi-threaded applications then the cpu with less cores (but better single core IPC) is always the better choice for a typical user.
 

sm625

Diamond Member
May 6, 2011
8,172
137
106
Celeron 827 is a single core 1.5GHz, 1.5MB cache, 17W chip. We'll probably never see a review of it but if it was clocked down to its lowest power state it would probably outrun and outlast a bobcat. Then of course there is the dual core version, the 847, which is just a bit slower but has the same TDP. The 22nm version of this chip should be quite the little beast. Yet all these chips are vapor afaik because no one ever reviews them.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
64
91
My take-away message:
single-core IPC matters a lot. I don't want to advocate single-core CPUs but I would always choose half the beefy cores over double "so-so" cores.

I think this message really can't be understated.

It's not just about IPC, and its not just about core count.

There is a minimum IPC and minimum core count that are needed.

I don't know what the minimum IPC would be, that would seem to be end-use specific, but from an OS response-time deadlock factor it would seem reasonable to set the minimum needed core count as two (2).

This much seems to be universally true regardless of platform and usage model - be it a cellphone or a big-iron server.
 

HopJokey

Platinum Member
May 6, 2005
2,110
0
0
I think this message really can't be understated.

It's not just about IPC, and its not just about core count.

There is a minimum IPC and minimum core count that are needed.

I don't know what the minimum IPC would be, that would seem to be end-use specific, but from an OS response-time deadlock factor it would seem reasonable to set the minimum needed core count as two (2).

This much seems to be universally true regardless of platform and usage model - be it a cellphone or a big-iron server.

While this may be true I feel that increasing IPC is much more difficult than adding cores from a technical standpoint. In between the two would be scaling frequency.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
64
91
While this may be true I feel that increasing IPC is much more difficult than adding cores from a technical standpoint. In between the two would be scaling frequency.

I don't know how one would go about quantifying "more difficult", but in terms of the expense of doing A vs. B there is no question that it comes down to a business question of "pay more now in R&D to develop higher IPC core logic" versus "pay more later in manufacturing expenses by having a larger die-size owing the increased core count".

Naturally there is no universally correct answer, its a trade-off that must be done at a business strategy level in the company and is one that is based on cash flow, business objectives, and risk management.
 

LOL_Wut_Axel

Diamond Member
Mar 26, 2011
4,310
8
81
Bobact is a low power cpu just as atom. it is "slow by intention". Anyway there are both not even close to "beefy" in my terms hence atom vs SB

Yeah, but putting Atom and Bobcat in the same terms when it comes to architectures I think is wrong. Both have a 2-issue design, but Bobcat gets a huge boost from out-of-order execution instead of in-order like the Atom. The Atom does have HyperThreading, so how they end up is with Bobcat having much better single-threaded performance and only a bit better multi-threaded performance.

Anyway, yeah. They're both much slower than Sandy Bridge, and than Phenom II even.
 

Gundark

Member
May 1, 2011
85
2
71
In Europe it's Sept.22nd. 10 or so hours more to wait. It's killing me, especially when it's a very high chance that nothing will happen.
 

Borealis7

Platinum Member
Oct 19, 2006
2,901
205
106
is it going to be
"I, for one, welcome our new Bulldozing Overlords."
or
"show's over. nothing to see here, folks."

guess we'll know soon enough.