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

[Maxforces]

3-5% average clock to clock wont give you 100MHz, performance wont scal 100% with MHz.
even if scaling will be perfect 1:1 you need 126Mhz-210MHz to much IB 3-5% average clock to clock
4200+ 3%=4326
4200 + 5%=4410
if there is perfect scaling, as is known it is not perfect
On average you need 150-200 not 100

[FlanK3r]

look again at tab...did you seen? Or effectivity performance/power consumption. In my opinion is SB better for classic OC ratio/performance/power consumption. You can tets easily in home (or I can), test 4800 MHz SB vs 4700MHz IB and you will see. I have of course both CPUs...

[Maxforces]

OCCT

it`s all over web like this

[Manicdan]

1.25v compared to 1.38v
not sure how stable each is. or what the cooling used is. or if they did an IHS mod on the IB. or what the temps are. need that info to know if one chart has any weight.

its easy to find a review that puts things in a perspective that someone is looking for, because the OC power consumption can be anything based on how you build the PC.
i did a little review on power consumption based on temps. my 2500k can be at 234w load or 246w just from letting the temps go from 48c to 54c. being able to keep that IB cool will shave off insane wattage. im not even going to compare voltages because any OCer knows that is the biggest factor. just cause this guy "needed" 1.38v for 4.7ghz does not mean everyone else does. IB is pretty good in the mid 4ghz range, but to reach 5ghz you need to really force feed it some electricity, where SB was nearly linear all the way up to 5ghz. if you were limited to an all in one water loop (like a H80) or air cooling, i would not recommend IB for OCing.

[FlanK3r]

1.25V with E1 is very rare...I have average chip and I need 1.375V for 4.68 GHz and 100C in load . But back to Vishera, Intel forum is at otherside ...

[Oese]

I'd really like to have some benches in r13 cinebench AMD vs. Intel. If AMD really gets a +40% boost from newer cinema 4d engines, then this might be the change AMD was looking for.

If only cinebench, which represents an actuall app and new technology in contrast to SuperPi etc., would have recognized bulldozer and the correct renderpaths right at launch, the whole picture could have been different.

It would have been obvious, how powerful the new arch is besides its memory bandwidth, and it would have been obvious, that it is the software not using latest tech that slows down bully in games etc.

People would have said "ok, in games it runs in gpu limitation, and in apps i get more power with the latest tech. good processor at a good price. ok, downside is higher power consumption, but i get higher perf for that too, right?"

Now, 1 year after release, there are some very unpopular benches in linux environment that show bully can actually hold its ground, but much much more popular cinebench r11.5 which is the bench bully is worst looking in.

And the new cinebench not available anywhere...

If the rendering speed under cinema 4d R13 is truly +40% on bully/vishera, then we get 7.7 against like 8.3 CB Points R11.5 vs. R13 what is only 10% more points in R13. Now if doing the math, Intel points then must be lower in R13 CB then in R11.5... You did think the last thought about what would happen then to the AMD vs Intel discussion, not me

[Piledriver]

• FX 8120@4.5ghz : ~ 679s
• I7 2600K@4.5ghz : ~ 621s
• The thing you already know :~ 580s

This pretty much are in line with the benchmarks tom's did with piledriver vs zambezi in single thread equal clock. 15% faster plus ~2% from the L3 cache, so... please leak more.

If AMD delivers the promise of 15% improvement on top of piledriver we will be talking about ~30% performance improvement in just two years. Not bad since this is the last performance CPU from AMD


Can someone tell me how u thank someone? I'm retarded.

[Evantaur]

Can someone tell me how u thank someone? I'm retarded.

you need post count of 100 to thank someone

[informal]

IPC won't be 15% higher uniformly. You will have some workloads that will show little to no improvement(0-3%) and some seeing pretty high improvement(~15+%). On average IPC improvement will be in the range of 6-10% and clock improvement will be in the similar range(same max Turbo,higher all core Turbo and higher stock clock). On average FX8350 should be 15-16% faster than FX8150,both at stock. In some workloads the difference may be a huge 25+% though.

[Maxforces]

If the rendering speed under cinema 4d R13 is truly +40% on bully/vishera, then we get 7.7 against like 8.3 CB Points R11.5 vs. R13 what is only 10% more points in R13. Now if doing the math, Intel points then must be lower in R13 CB then in R11.5... You did think the last thought about what would happen then to the AMD vs Intel discussion, not me

each processor gets jump performance under cinema 4d R13 comparing to cinema 4d 11.5.There is no magic only for bully/vishera

PII X4 3.9ghz




r11.5- 5.27
r13- 4.41

FX 8150 4.5ghz




r11.5- 3.15
r13- 2.49

2600K 4.5ghz




r11.5 - 2.51
r13- 2.33

vishera will be faster than old SB in cinema4d r13 clock to clock, nice after almost 2 years

[Oese]

Ah thanks, i just read somewhere in this thread.

But still from your tests SB gains only 10% whilst Bully gains like 25%. What i didnt expect is the Phenom ii to gain 20% also... How come? Even this one didnt get the right optimizations in r13 even though its rather old?

Given these results, cb13 points should still go down in relation to cb11.5 for SB, whilst Vishera gets a plus..

[docwhom]

But still from your tests SB gains only 10% whilst Bully gains like 25%. What i didnt expect is the Phenom ii to gain 20% also... How come? Even this one didnt get the right optimizations in r13 even though its rather old?

I think I'm going back to AMD from SB simply because C4D is developing nicely with AMD now. Love having at least one application that goes better on this gear, finally.

[FlanK3r]

FX-8350 4.5GHZ = 7.77 points in R11.5
How much MHz up need FX-8150 Zambezi for similar score?


4715 MHz and "only" 7.64


4740 MHz and still 7.66


4786 MHz and 7.72 points


4815 MHZ and 7.76, finally


There could be a bit difference between boards and win7 settings, so we can to say, AMD Zambezi FX need +240 MHz gain for the same score as Vishera FX in Cinebench.

[Piledriver]

you need post count of 100 to thank someone

Thanks for the heads up, i gave up a couple times, but yesterday i really wanted to thank the leaks and i was going crazy, searched everywhere...

[Maxforces]

Ah thanks, i just read somewhere in this thread.

But still from your tests SB gains only 10% whilst Bully gains like 25%. What i didnt expect is the Phenom ii to gain 20% also... How come? Even this one didnt get the right optimizations in r13 even though its rather old?

Given these results, cb13 points should still go down in relation to cb11.5 for SB, whilst Vishera gets a plus..

These questions head to the manufacturer,perhaps more friendly compiler.AMD users should be happy.There is no difference between R13 and r14

[Oese]

Yeah seems like r11.5 underrated phenom ii also.

screw it why is r13/14 not released as a bench already?

[SUPERKAMES]

@Maxforces :
Can you give me the 3D scene (for testing with cinema4d r13) ? anh which renderer did you use ?
Thanks.

[Maxforces]

http://www.oyonale.com/modeles.php?lang=en&page=58
default renderer 800x600

[RussC]

Those IB numbers are really bad because of INTC screwing up the TIM. When u remove the IHS and use a good TIM, those #s come way down on CPU core and temps. Anyone that OC's a IB pulls the IHS now. IB is much closer than the chart shows with proper TIM. When INTC gets those parts back to being soldered, it will be a much better CPU comparison to SB. The core temps will drop ~25degC with good TIM/soldered IHS. There are a few threads in the INTC forum that chronicle this very well. INTC will eventually fix this issue.

So bring this back around, that comparison is not really that valid, as the IB core is really much better than the chart shows.

RussC

not., again not...

there is clock to clock comparison tab at 4200MHz (turbo off). As you can see, average is only 3.42% better than Sandy-DT.
Attachment 130248


AND OC vs power consumption vs Cinebench performance against SB-DT
Attachment 130249


PS: the difference between Zambezi and Vishera could be simillary as between SB-DT and IB-DT (so about 3-5% average clock to clock)

[informal]

And until that happens we have what we have. What-if comparisons are valid only in theory.

[FlanK3r]

it is the same as to say, FX could be good, if start clocks was 4/4.5 GHz because fail of Glofo ...

[sequoia464]

Anyone have an update on the release date? I have heard Oct 12 and Oct 23.

BLT is still showing 10-12 as an ETA for them .. http://www.shopblt.com/cgi-bin/shop/...er_id=!ORDERID!

[informal]

APUs should be launching Oct 2nd while new FX is anywhere from Oct 17th to Oct 22nd. BTW Windows 8 launches Oct 26th IIRC so it might be a clue for a launch date. It does have optimized scheduler for Bulldozer so AMD is maybe waiting for a combined launch so they can get those ~3-7% performance benefit(which is not bad at all ).

[SUPERKAMES]

Piledriver also , air cooler only.



PS:Thanks Maxforces , I'll try the Cinema4D file later , maybe on Friday .

[informal]

Thanks for that result SUPERKAMES . It looks great!