AMD "Piledriver" refresh of Zambezi - info, speculations, test, fans

[tbone8ty]

He applied Microsoft patches... I thought with pile driver scheduling was fixed? Or is that with steamroller?

6.82 for the 8350... Not bad what did 8150 get for 11.5 at that clock?

[Shadowized]

http://hardforum.com/showthread.php?t=1722207

results are within 1% of orbs, so I guess he was right similar to with his old BD benchs, quite frankly though I really have to question what the hell AMD have been upto seeing as in CB 11.5 I was getting around 7.4 points with my 1090T @ 4ghz and that thing is going on 2.5+ years old now.

[tbone8ty]

Lol they didn't do sheet! Hence they r firing 30% of their workforce

[informal]

The Hardocp results are quite good for x264. The poster is using a 32bit build which is around 16% slower in pass1 versus x64 build. 2nd pass is not affected that much(if at all). So here are 4.3Ghz Bulldozer x64bit results:

Results for x264.exe r2200
x264 Benchmark: 64-bit
==========================

Pass 1
------
encoded 11812 frames, 47.11 fps, 7754.23 kb/s
encoded 11812 frames, 45.39 fps, 7754.43 kb/s
encoded 11812 frames, 46.00 fps, 7754.17 kb/s
encoded 11812 frames, 45.79 fps, 7754.20 kb/s

Pass 2
------
encoded 11812 frames, 11.04 fps, 8002.11 kb/s
encoded 11812 frames, 10.92 fps, 8002.10 kb/s
encoded 11812 frames, 10.91 fps, 8002.12 kb/s
encoded 11812 frames, 11.06 fps, 8002.12 kb/s


System Details
--------------
Name AMD FX-8120
Codename Zambezi
Specification AMD FX(tm)-8120 Eight-Core Processor
Core Stepping OR-B2
Technology 32 nm
Stock frequency 4100 MHz
Core Speed 4305.3 MHz

Northbridge AMD 770 rev. 00
Southbridge AMD SB700 rev. 00

CAS# latency (CL) 7.0
RAS# to CAS# delay (tRCD) 7
RAS# Precharge (tRP) 7
Cycle Time (tRAS) 20
Memory Frequency 420.0 MHz (1:2)
Memory Type DDR3
Memory Size 4096 MBytes
Channels Dual

Windows Version Microsoft Windows 7 (6.1) Ultimate Edition 64-bit Service Pack 1 (Build 7601)

FX8350:

^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Results for x264.exe r2200
x264 Benchmark: 32-bit
==========================

Pass 1
------
encoded 5906 frames, 47.22 fps, 7779.53 kb/s
encoded 5906 frames, 47.78 fps, 7779.53 kb/s
encoded 5906 frames, 47.15 fps, 7779.53 kb/s
encoded 5906 frames, 47.91 fps, 7779.53 kb/s

Pass 2
------
encoded 5906 frames, 15.36 fps, 8002.11 kb/s
encoded 5906 frames, 15.38 fps, 8002.11 kb/s
encoded 5906 frames, 15.37 fps, 8002.11 kb/s
encoded 5906 frames, 15.34 fps, 8002.11 kb/s


Is there a way to change the frames? I just ran it and hoped for the best. I also got the x64 version working and will try that next

In pass 1 , 4Ghz 8350 in 32bit build (~16% less efficient than x64) matches 4.3Ghz FX8120 in x64 build. Pretty good result which means PD's IPC in this particular test is more than 7% higher than Bulldozer's (closer to 10%). User will upload x64 results soon so we will know the exact improvement.
In pass2 the improvement is massive even if PD uses 32bit binaries. PD @ 4Ghz scores 15.35pts while BD @ 4.3Ghz scores 11.05pts. I rechecked the BD numbers since another user on techarp forum posted a QC BD results and they are roughly 2x lower then FX8120@4.3Ghz,corelating well with 2x less modules/cores. So in summary pass2 is massively better on PD for whatever reason.

edit:
For comparison with 1090T,here is what it scores in same benchmark( as response that they didn't do "sheet" )

Results for x264.exe r2200
x264 Benchmark: 32-bit
==========================

Pass 1
------
encoded 5906 frames, 26.34 fps, 7779.70 kb/s
encoded 5906 frames, 26.39 fps, 7779.70 kb/s
encoded 5906 frames, 26.35 fps, 7779.70 kb/s
encoded 5906 frames, 26.29 fps, 7779.70 kb/s

Pass 2
------
encoded 5906 frames, 11.24 fps, 8003.26 kb/s
encoded 5906 frames, 11.22 fps, 8003.26 kb/s
encoded 5906 frames, 11.19 fps, 8003.26 kb/s
encoded 5906 frames, 11.24 fps, 8003.26 kb/s

Results for x264.exe r2200
x264 Benchmark: 64-bit
==========================

Pass 1
------
encoded 11812 frames, 31.58 fps, 7754.14 kb/s
encoded 11812 frames, 31.69 fps, 7754.05 kb/s
encoded 11812 frames, 31.63 fps, 7754.11 kb/s
encoded 11812 frames, 31.84 fps, 7754.16 kb/s

Pass 2
------
encoded 11812 frames, 10.94 fps, 8001.97 kb/s
encoded 11812 frames, 10.91 fps, 8002.01 kb/s
encoded 11812 frames, 10.97 fps, 8001.98 kb/s
encoded 11812 frames, 10.95 fps, 8001.96 kb/s


System Details
--------------
Windows Version Microsoft Windows 8 (6.2) 64-bit (Build 8250)
DirectX Version 11.0
Number of processors 1
Number of threads 6 (max 6)
Name AMD Phenom II X6 1090T
Codename Thuban
Specification AMD Phenom(tm) II X6 1090T Processor
Package Socket AM3 (938)
HT Link speed 1995.8 MHz
Stock frequency 3200 MHz

FX8350 vs 1090T,both stock and running same binaries in x264 test. Pass 1 is 49% faster on FX8350 while pass 2 is 40% faster on FX8350. Massive improvement.

[slaveondope]

results are within 1% of orbs, so I guess he was right similar to with his old BD benchs, quite frankly though I really have to question what the hell AMD have been upto seeing as in CB 11.5 I was getting around 7.4 points with my 1090T @ 4ghz and that thing is going on 2.5+ years old now.

Ya I know. Won't be getting rid of the best AM3 chip any time soon myself.

Lol they didn't do sheet! Hence they r firing 30% of their workforce

Their stock is showing it too. Waiting to hear of a buyout in the coming months.

[Shadowized]

For comparison with 1090T,here is what it scores in same benchmark

if you compare both at stock yeah I can see that 40% figure, however this is xtremesystems and a lot of people would be ashamed to leave a gem such as the 1090T at stock speeds, most are all within the 3.6 - 4.4ghz range judging by that old Thuban thread we've got kicking around.

[informal]

Then you apply the same logic to FX8350,right? It can OC to ~5Ghz on water cooling as reports tell us,so in that case we have an avg 4Ghz(from your 3.6-4.4 range) for X6 vs 5Ghz on FX8350. Both OCs are 25% above the stock for both CPUs. This won't change the advantage of 40-50% that 8350 sees in stock vs stock case (since we OC both CPUs by same % amount).

[tbone8ty]

still waiting on power consumption numbers....

anandtech said piledriver is more about power efficiency than anything else.

though still wondering if u really need to use microsofts patches for bulldozer on piledriver for win7?

[vario]

<--informal
You know your nitpicking here right ?
From these early data it seems its still going to be WORSE in MOST lightly threaded scenarios than X6.
Dont get me wrong, i would be real happy to have a reason to upgrade my ancient X6, im almost hating myself now for not being able to cope with blue in my system.
And X6 was a weak competition to blue when it got out already.

[informal]

I'm just comparing X6 to FX8350. One chip was not frequency oriented while the other one is. The fact that you could actually push X6 to 4Ghz was a great thing ,but why should we ignore the OC potential of BD/PD? It offers great OC at the expense of high(er) power draw. You still get the increased performance versus X6, 25+% in MT area and not lower performance in ST tests(25% higher clock on PD should be more than enough to cover the IPC deficit it could have versus Thuban). No matter how you look at it, FX8350 is more than worthy successor to X6 ,much better than FX8150 was when it launched(it was not that bad either).

Oh ,let's not forget the complete ISA support PD core offers,even more than IB does (FMA3/BMI+ XOP on top of AVX and previous SSx). For 208 bucks this is a great deal IMO ,especially when you consider the fact that it is still AM3+ compatible and that the board will likely take in the SR based FX core next year.

[vario]

Of course new instructions are great, AES and some others that might even be useful.
However after all this time, PD should be better all round cpu.From these early results im just not seeing this.
X 6 overclocks to 4.0 4.3 (stable) and NB overclocks mostly in the 3.0ghz area (and NB clocks in thuban give it some great boost in some applications like games).All this at a MUCH lower power consumption than BD at ~4.7 which seems to be a stable clock for BD`s so im pretty sure that BD wasnt "not all that bad" .And it looks like PD wont clock higher or much higher, 5.0ghz on water looks like ~4.7-4.8 on air so maybe a small bump,maybe even not that.
What im hoping for is a LOT better power consumption figures,but if its going to chew up ~300W at its highest air OC, how its better than thuban ?Its unusable for day to day operation at this specs.
Attachment 130602
This kind of situations SHOULD NOT HAPPEN,and adding insult to injury, in this test x6 did not have NB overclocked,and 8150 was consuming 50 watts more.Maybe PD will at worst match X6 and at best move long past it.
Phenom II architecture was introduced such long ago,that now AMD REALLY should be able to put something thats just better.In every regard.

[demonkevy666]

Of course new instructions are great, AES and some others that might even be useful.
However after all this time, PD should be better all round cpu.From these early results im just not seeing this.
X 6 overclocks to 4.0 4.3 (stable) and NB overclocks mostly in the 3.0ghz area (and NB clocks in thuban give it some great boost in some applications like games).All this at a MUCH lower power consumption than BD at ~4.7 which seems to be a stable clock for BD`s so im pretty sure that BD wasnt "not all that bad" .And it looks like PD wont clock higher or much higher, 5.0ghz on water looks like ~4.7-4.8 on air so maybe a small bump,maybe even not that.
What im hoping for is a LOT better power consumption figures,but if its going to chew up ~300W at its highest air OC, how its better than thuban ?Its unusable for day to day operation at this specs.
Attachment 130602
This kind of situations SHOULD NOT HAPPEN,and adding insult to injury, in this test x6 did not have NB overclocked,and 8150 was consuming 50 watts more.Maybe PD will at worst match X6 and at best move long past it.
Phenom II architecture was introduced such long ago,that now AMD REALLY should be able to put something thats just better.In every regard.

an X6 can easily pull more then 300 watts.

[informal]

Thuban heavily OCed (4+Ghz) draw A LOT of power. Power draw practically skyrockets,just like it does with BD. So no ,power draw is not an excuse. Also I doubt that highly OCed Thuban ,even with OCed NB, could be faster overall chip than PD @ 5Ghz .The IPC deficit is not uniform and varies a lot. In some very important workloads Bulldozer just destroys Thuban. Let me illustrate with some real world application workloads (numbers from techspot's i7-3770K review):

Application_01.png
8150 is 36% faster than 1100T.

Application_02.png
8150 is 28%/20% faster than 1100T.

Application_03.png
8150 is 30% faster than 1100T.

Encoding_01.png
8150 is 30% faster than 1100T.

Encoding_02.png
8150 is 31%/8.5% faster than 1100T (MTed pass and low threaded pass)

Encoding_03.png
8150 is 22% faster than 1100T.

I don't want to link the gaming results since FX8150 is at least on par with 1100T or faster in every game they tested.

[BeepBeep2]

Informal,
I disagree.

X6 had outliers, CPUs that would do 4 GHz+ at less than 1.35v, or CPUs like Infrareds that did 4.2+ with 1.44v or less...lol
These CPUs did not draw ridiculous amounts of power, there's no way.
Then we had the middle, or average, 4-4.1 GHz at ~1.45v
Then some really bad, needing 1.5v+ for 4 GHz.

So far with FX, the outliers are CPUs doing about 4.8 with 1.45v, average doing 4.6 with 1.45v, and worse doing ~4.4 with that voltage.

Nobody is pushing the voltage higher because power consumption, temperature, current draw (ESPECIALLY for a 32nm process) is already out of hand at that point, and temperatures are often completely unmanageable, in 70c range for those with good CPUs and $300+ water setups (at least 3x120 rad).
With Thuban, some of those guys had that 4 GHz+ overclock with manageable temps, and I know for one, with an average CPU, 1.44v load 4 GHz I could push 1.62v load without worrying about temperatures all that much. That was .3v over stock before power consumption went crazy, on FX that can be as little as .1v...
It isn't the same, and on top of all these behaviors of these different CPUs, I don't feel safe pushing XXXa amount of current and Y.Y voltage into a 32nm transistor vs XXXa current and same Y.Y voltage into the 45nm part.
And again you clutter a thread with handpicked charts that show the 8 core matching at least, a last gen 6 core on worse manufacturing process, or beating the 6 core by less than the 33% needed to say they really "match".

I've bought 4 Phenom II X4 955/965 over the last few months and all of them did 4 GHz with <1.45v too, where how does FX stack up in single thread and gaming?

When you shrink a node and change the uarch, you expect to see gains like Core 2 > i7, not i7 > Core 2 with more cores/threads glued on...

[FlanK3r]

btw, in CS6 Photoshop is FX better than in CS5.x (more close with 2600k).

beep:every chip is different, in OC, in temperatures. My FXs (now 5x FX-8150, 2x FX-8120) are all different, some are cold, some hotter with similar OC. The same in power consumption, between the best and worse chip is about 30W difference in load in stock!

[BeepBeep2]

btw, in CS6 Photoshop is FX better than in CS5.x (more close with 2600k).

beep:every chip is different, in OC, in temperatures. My FXs (now 5x FX-8150, 2x FX-8120) are all different, some are cold, some hotter with similar OC. The same in power consumption, between the best and worse chip is about 30W difference in load in stock!

I know, that goes for every CPU generation
But once again informal tries to shed only good light on FX and hide all it's problems.
This time I take a neutral stance instead of being upset about AMD taking a step backward...like I was at release, someone hoping AMD would close the MT (power) efficiency gap with intel and also partially close the already visible single thread gap they had with intel...(I was mad because while they improved MT performance 50% of the time, power consumption was so bad, efficiency got worse for what they had compared to 45nm, ST got worse, etc)

The problem now though is:
The process matured...
Overclocks got worse in general.
Power consumption is still hit or miss, not sure if generally better than older uarch, but this process is mature now so lets wait for 8350 reviews
In many single thread apps, this architecture does not cut it.
In many MT apps, it beats the old architecture by less than 33% - each thread giving less performance than older uarch...
Informal claims it beats or at least matches Phenom II overall gaming, but I see people getting 4.3+ Phenom II stable and highest I've seen across the net with BD was 4.8 air up until now, in which case Phenom II wins 1/2 the time and Informal's claims go from "beat or match" to "match or lose".

Look, they are fun to have and a lot of us don't care about power consumption but are just enthusiasts who want new toys so it is ok...

But anyone wanting to get work done efficiently chooses the 8-thread intel product, which pays itself off in a few months of electricity and beats the AMD product up to 80% in single thread, or power efficiency multi-thread.


I don't like the way OBR used his english sometimes, and the kind of things he writes about, his :banana::banana::banana::banana::banana: blog where he attacks people personally and uses vulgar language about those people or subject of his articles...
But he does make points once in a while and personally I am tired of Informal's relentless AMD fanboyism

[informal]

I clutter the thread with cherry picked charts?For real? The workloads are real world and FX beats old Thuban by huge margins. It does that with less resources per core which is a feat by itself. In games it's not worse. Yes,it has a very high power draw when clocks/Vcore is pushed but that doesn't mean it "sucks". The process is very leaky and we have to see if Piledriver will change much in that regard. But claiming it's a fail CPU just because it has lower IPC in some cases or because it cannot beat X6 uniformly and across the board in all benchmarks (usually crap benchmarks like super pi or wprime) is ridiculous.

Also notice I haven't mention any intel 8 thread product since FX cannot beat those.It's a fact and nobody is denying that.

PS I'm tired of ignorant people ,I just don't say it flat out .

[xVeinx]

To be honest, I hyped up Bulldozer a bit myself. I saw it perform better for certain software that does string comparisons. I also found that Matlab performs poorly on it, along with certain python functions. It is hit and miss, and I don't think Informal is being unreasonable. Nobody is arguing that Bulldozer is without it's flaws, and no one in this thread will be fooled into thinking they should buy an FX 8150 instead of a Xeon E5-2690.

[informal]

To be honest, I hyped up Bulldozer a bit myself. I saw it perform better for certain software that does string comparisons. I also found that Matlab performs poorly on it, along with certain python functions. It is hit and miss, and I don't think Informal is being unreasonable. Nobody is arguing that Bulldozer is without it's flaws, and no one in this thread will be fooled into thinking they should buy an FX 8150 instead of a Xeon E5-2690.

Exactly. And AMD knows that and they adjusted the price according to performance. We can see that pre-order price for FX8350 is 208$ on shopblt (unless they changed again). For this price it is a compelling product . Sure it will have higher power draw than most of the CPUs in its price range but it will have certain strong points and for people that need that kind of performance in their workloads it will do just fine. In games it will certainly not be as competitive as in some other stuff so gamers will go intel route . It all depends on how much one is willing to pay and what he uses his computer for.

[Darakian]

If you don't like it don't buy it. FX is designed for heavily threaded work loads and (non-gaming) work loads are becoming heavily threaded.

[tbone8ty]

BeepBeep2,
Just go away man. You're a troll and a flamer hiding behind an elitist asswipe's mask.

haha your joking right?

[BeepBeep2]

Did I do something wrong here?
Informal came in, grabbed all the charts where FX won against Phenom II from a review, did not link the ones where it lost/only matched, and furthermore we've all seen them five-hundred times and the CPU was released a year ago.

This is a thread about Piledriver, it is 11 months after Bulldozer's and he's still trying to prove to people day after day how great Bulldozer is. Then his latest retort to me includes reasonable thoughts, however he says things such as -

Yes,it has a very high power draw when clocks/Vcore is pushed but that doesn't mean it "sucks".

But that was not the reason I think Bulldozer sucks?
Maybe it sucks because it often can not beat Intel's 45nm 4 core 8T parts?
Maybe because you often need a water cooling setup worth more than the processor to cool it satisfactorily, even when barely above stock turbo voltage? (1.4v-1.475v)
Maybe it sucks because it loses to 32nm Intel 4 core 8T parts 80% of the time, and when it wins it is by a small margin?
Maybe because in gaming, if you want to upgrade from your Phenom II X4 CPU for several real-world applications, you MUST buy the 8 "core"?
...

But claiming it's a fail CPU just because it has lower IPC in some cases or because it cannot beat X6 uniformly and across the board in all benchmarks (usually crap benchmarks like super pi or wprime) is ridiculous.

Why throw SuperPi or wPrime into this? Just because I bench at HWBOT does not mean I am too stupid to realize that x87 (SuperPi) is antiquated.
wPrime is a threaded benchmark, too, that calculates sqr roots, but is useless outside competitive overclocking.
You say it doesn't lose in gaming, which isn't always true, but if you average everything out and make sure to include BF3, I guess you're right. I just don't know what you are comparing here, stock for stock or overclocked?
Also, I thought we stopped arguing about this 10 months ago?

I don't get it...you are always on this crusade to advertise all the great things about BD, and there certainly good things about it, but it just isn't the "total package" that it needs to be for a high-end/upper midrange desktop CPU...it belongs in a workstation with the Opteron name and AM3+ socket, not "FX". For all the hype AMD and its fans put on it's release, it was pathetic...

On top of all that, no matter how much you talk about the upsides of the multi-threaded performance, as soon as you look at the power consumption it gets thrown out the window...

8150 launched at $270, at the time 2600Ks were $280...I would have chosen the 2600K 5/5 times...when FX won by 10% in benches, it was using 70% more power.

BeepBeep2,
Just go away man. You're a troll and a flamer hiding behind an elitist asswipe's mask.

Cool story bro? What are you getting at here?
What am I trolling, what am I flaming, and how am I hiding, more specifically behind an elitist asswipe's mask?

AMD FX-8150 vs Intel 4C 8T = "sucks all the time"
AMD FX-8150 vs AMD X6 1100T = "why is it not better, they had 2 years and a node shrink, sometimes takes 8 threads to match 6, and sometimes more power consumption"
AMD FX-6100 vs AMD X4 9x5 = "This is not an upgrade"...

I don't know if you realized, but I bought an 8150, and also did decent testing of the Windows 7 patches and explained how they help the performance of FX...and FX is a lot of fun to overclock.

That's about it...

[BeepBeep2]

Thuban heavily OCed (4+Ghz) draw A LOT of power. Power draw practically skyrockets,just like it does with BD. So no ,power draw is not an excuse. Also I doubt that highly OCed Thuban ,even with OCed NB, could be faster overall chip than PD @ 5Ghz.

Bit tech, first hit I had on google...
Overclock settings:
FX-8150 - 4818 MHz, 1.45v, 1.2v CPU-NB, unknown CPU-NB clock, due to voltage, assume CPU-NB = <2400, HT Ref 205
1100T - 4212 MHz, 1.51v, 1.xv CPU-NB, but HT Ref 301, and HT Link 2107, so assume CPU-NB = >2100

The FX-8150 and 1100T run similar stock voltages...the power consumption goes up to ~600w with just 1.45v, at this point is already blowing up VRMs...the 1100T is horrid too, but has .06v more and draws 100w less...meanwhile 2500K/2600K are drawing 300w less than the FX-8150, about half the consumption...

Efficiency at stock vs what was competing intel solution, before power consumption blows up with voltage increase/OC, 2500K was 100% better through Gimp (Single thread), Handbrake, and 7-Zip...

Nope, doesn't suck...

I also doubt that highly OC'ed Thuban would be faster overall than PD at @ 5 GHz. Aren't you making this number a little safe for yourself, so no one can attack it?
I thought we figured that overall, BD needs to be 600 MHz higher than Phenom II to be faster overall? (Get close in ST, win noticeably in MT?)
So if PD is even 5% faster IPC vs BD, then obviously 5 GHz PD, a Thuban at 4.3-4.4 even could not be faster overall...

[FlanK3r]

Again, this is not so simply (black or white). In some reviews (I dont know, if half of reviews or less or more) was 8150 power consumption lower than 1100T at stock. Of course, theris big difference in OC load. 8150 after is near SB-E CPUs. If u saw graphs voltage/power consumption, every chip is horrible in power consumption from some point, depens at voltages. Man must find some compromise between OC and voltage. IB has great power cons., but from some voltage is worse than SB with high voltage.

[informal]

Don't worry,Beep saw the "light",now he is here to teach us a few things . It doesn't matter that I don't even use intel for comparison and that I already said it cannot compete with intel.
Beep,why don't you just take your 2500K/2600K and play with it and leave this section to "deluded fanboys" .