AMD Zambezi news, info, fans !

[FlanK3r]

Attachment 121184

I think, power consumption is mistake of editors (some used bad BIOS version). OBR has more experinece, than most of editors from all world, look at his results. Still could be some problems in BIOS of some motherboards.

[dess]

Let me ask, again, what could be the rationale behind all this 2x2-way INT CMT thingy, that they've preferred it over 1x4-way SMT? Easier to design and leaving the rest to the clocks, hoping for the best?

SMT could also be implemented on the scalar INT domain (FlexFP is so, already), probably on the same or even lesser footprint... That way they wouldn't have limited peak IPC, unlike now. Who cares if the 2nd thread per module brought lesser %, if the "100%", so the ST performance was much better!?

Well, perhaps the cost of R&D for that would have been just too must for them.

Anyway, it's perhaps even bigger of a marketing fail to call it a 8-core. People are not that stupid, no matter what some marketing duds think. For a modest 4C/8T CPU it would've looked much better...

[mav2000]

Yeah, Bulldozer is an epic failure, I built a 990FX system all ready to go hoping that Bulldozer would deliver but apparently didn't. I have no more hope for AMD when it comes to CPU's anymore. and I doubt their next generation would be any good as well. The gaming benchmarks were the worst though, and power consumption was insane despite that its slower than the i7 2600K, not worth it. Not going to buy Bulldozer anymore.

Tell me about it...same boat, different place....this sucks.

[mav2000]

Yeah, Bulldozer is an epic failure, I built a 990FX system all ready to go hoping that Bulldozer would deliver but apparently didn't. I have no more hope for AMD when it comes to CPU's anymore. and I doubt their next generation would be any good as well. The gaming benchmarks were the worst though, and power consumption was insane despite that its slower than the i7 2600K, not worth it. Not going to buy Bulldozer anymore.

Tell me about it...same boat, different place....this sucks.

[The Stilt]

Atleast the IMC is completely 'twisted' on BD

No trouble doing DDR-2400 24/7 (unofficial divider on Crosshair V) and much much higher with better cooling (due the fact that the minimum NB to DRAM ratio is 1:1 and high MEMCLK stresses the NB much more).

Shamino did DDR-2904 on a not_so_good A0-stepping BD (back in April).

I did not find any trouble doing DDR-2700 (Memtest stable) on Crosshair V once the bios was fixed.

"Underclocked" CPU, northbridge at minimum speed, DDR-2700:

Attachment 121187

CPU with a bit better clocks (NB too):

Attachment 121186

The bandwidth is ok I guess

[savantu]

What we have been promised:
-higher or equal IPC (AMD HC22 presentation, AMD bulldozer blog,JF blog and posts)
-more cores(33% more)
-more clocks
-overall 30-50% more performance in the same TDP envelope than previous design

What we got as a final product:
-lower IPC,ranging from 10 to 25%!
-more "weak threads" that in many workloads barely beat X6 at lower clock ( weak threads are direct result of much lower IPC than K10even 4.2Ghz Turbo is not enough to offset this)
-higher clocks with much higher power draw under load (this can be GloFo's mistake for all we know)
- overall around 10-15% more performance vs 1100T

End result: we practically got 8 Bobcat cores (which are sometimes slower per clock than real Bobcat!),higher power draw than 1100T which can clock rather high (with astronomically high power draw when you do that).

With AMD's projected IPC improvements(now to be taken with much salt) ,even Steamroller will not be able to beat Llano which is around 7% faster than Thuban at around same clock while having no L3 cache at all.This is major fail I'm afraid. For desktop at least. For server I suppose they can do fairly well in SOME workloads,but overall they will be just a tiny bit better than where they were with MC. 2B transistors and this is what we get...

Oh my...Loooking at benchmarks this morning, I was remembering all the discussions with you over the past 18 months about BD....Well, we can't be both right at the same time.


Hardware Canucks did some pretty good comparisons :

http://www.hardwarecanucks.com/forum...review-20.html

PS : You should change your signature, someone was right and you and others were terribly in the wrong.

[FlanK3r]

Stillt: nice man, I need better RAMs for 2900 MHz :-D, But I found some drops with higher RAM clocks at 8-10-8 or 8-11-8 against 8-9-8 in overall bandwith (the best was about 2220 8-9-8 at clock about 5 GHz)

[Andi64]

Yay! Nice! DDR3 2900Mhz... oh, wait! No.

AMD_FX-8150-23.jpg

[FlanK3r]

because 7-7-7, 9-11-9 . Poor timings in second situation. Easy 7-9-6 1866 MHz for me at FX. An then u will see diferences. Of course, its more for overclockers and enthusiast than for daily users. But, we are at XS

[Allubz]

Has anyone got a clue on the overclock potential of the FX-8120? Is it binned, or as good as the FX-8150? In the latter case I'd consider buying a Zambezi to OC it to 5GHz 24/7.

[Frontl1ne]

Yay! Nice! DDR3 2900Mhz... oh, wait! No.

There were some performance improvements to be had if the testing was done with better memory (3DMark11 and Mafia 2):

http://www.madshrimps.be/articles/ar...#axzz1aY74DCwx

[PerryR]

Yeah, Bulldozer is an epic failure, I built a 990FX system all ready to go hoping that Bulldozer would deliver but apparently didn't. I have no more hope for AMD when it comes to CPU's anymore. and I doubt their next generation would be any good as well. The gaming benchmarks were the worst though, and power consumption was insane despite that its slower than the i7 2600K, not worth it. Not going to buy Bulldozer anymore.

Yeah, the fact that they would even bother to market this towards gamers is amusingly insulting. In their sad attempt at trying to be different from Intel, AMD shoots themselves in the foot.

[spursindonesia]

I don't think so,
the biggest problem with BD is the low single thread performance, low IPC.

so using softwares that can only take advantage of 1-4 cores will always make BD look BAD, and unfortunately that's the case with many popular softwares.

Speaker of the TRUTH ...

[Kobaltrock]

I know folks had really high expectaions for BD, I did too!
But, if you look at things from a rational point of veiw this chip doesn't look all that bad...
Yes, it's different, and it loses badly in ST performance. Keep in mind, this is a radical change from what we've seen in previous arch's.
It may take a little time for software/firmware tweaks (or even a respin) to make this chip shine!

It loses in MT performance as well, the X6 even beat it in some MT tests.
The current line of BD will never shine, there is just not enough tweaking that can be done to save it, unless you want the BD-zip'o'matic, then your golden!

What really shocks me, is the fact that I see alot of known AMD fans trashing it too....

None of this is really new to me. I hope it doesn't happen again, but I rode Commodore Amiga out of existence, and to this very day I still think it was a serious loss
from a tech point of veiw...

I'm less dissapointed in the chips performance, and more dissapointed in some of the responses I've seen.

I'll buy one as soon as I can, and push the snot out of it!
Maybe I'll be riding AMD out of business just like I did the Amiga....

I'd hope that reguardless of which camp your in, you'd realize that a world without AMD wouldn't be good for anyone.

Just because people here are more or less PO'ed at how all this was handled, doesn't mean we want AMD to go under. Competition is good.
Lying to the consumer is not, they brought this onto themselves. I don't blame anyone for jumping camps after this fiasco.

This would be like Apple releasing a iPad 3, that was slower than a iPad 2, yet they were telling world + dog that it will be faster, (aka IPC increases!)

The only way AMD can save face is to lower the price by $50 across the board, then, the sting wouldn't hurt so much.

[crazydiamond]

flank3r (or anyone) can you pls compare thuban 24/7 settings (somewhere around 4.2ghz/2.8ghz/1600mhz) vs BD 24/7 settings whatever they might be? win7 64bit if you can

[Barr3l Rid3r]

where about hypers on this crap?

[El Gappo]

Atleast the IMC is completely 'twisted' on BD

No trouble doing DDR-2400 24/7 (unofficial divider on Crosshair V) and much much higher with better cooling (due the fact that the minimum NB to DRAM ratio is 1:1 and high MEMCLK stresses the NB much more).

Shamino did DDR-2904 on a not_so_good A0-stepping BD (back in April).

I did not find any trouble doing DDR-2700 (Memtest stable) on Crosshair V once the bios was fixed.

"Underclocked" CPU, northbridge at minimum speed, DDR-2700:

Attachment 121187

CPU with a bit better clocks (NB too):

Attachment 121186

The bandwidth is ok I guess

Sick man

[badatgames18]

yep.. people have been saying 2400+ is very doable if imc is under cold.. even 3k from what i've heard

[radaja]

@leeghoofd in your review at madshrimps in the stock clock test section you said this

And yes we are finally in 2500K territory people. Reminder though : thats 4 cores and 4 hyperthreaded ones for the 2500K versus the "8" cores of the FX-8150.

i think you made a typo about the 4 cores and 4 hyperthreaded,should be just 4 cores.

[BeepBeep2]

From SA forum,new image:


3dmark06 ~3034pts,lower than Phenom I 9850
C11.5 ~2.86pts, 2.5x lower than 1100T Thuban
Super Pi.... ah who cares anymore lol
The chip this Chinese guy has is evidently a seriously borked ES 6 core Zambezi @ 2.6ghz.

PS now the scores are going further down with each new picture,we are in K7 range now^^.This is clearly terrace testing that ES

Anyone remember this post? This was April.
Add 33% cores theoretical score... = 3.8 Pts 8 core at 2.6 Ghz 3.6/2.6 = 1.384 x 3.8 = 5.26 pts, or 5.7 at 3.9 Ghz (all core turbo)

You know how hard we laughed at that leak?
We made fun of it, talked about how terrace must have been the source, how the source must have "ran all the tests at once"

...and look where we are today with much more optimized bioses...
Scoring a 6.0 instead of a 5.7.

[El Gappo]

yep.. people have been saying 2400+ is very doable if imc is under cold.. even 3k from what i've heard

Doesn't 'need' to be cold



Super easy.

Need to have a play with he higher strap.

[Leeghoofd]

Can you work over 2800 NB Gappo ( no cold) I only could succeed with cold...

[Sentential]

Man the spin just doesn't stop for this hunk of :

#1 Its super effective against 2600k
-No

#2 Well it overclocks a lot! 5ghz on air is common!
-No

#3 Well its memory speeds are insane!
-Gimme a freaking break

[jazkat]

im gutted and totally unimpressed with bd but with there r&d budget who would of thought they would stand a chance against the big blue?

has amd commented on WHY bd isnt what it was supposed to be yet?

[The Stilt]

Can you work over 2800 NB Gappo ( no cold) I only could succeed with cold...

Based on the binning I did the average on the FX-8150´s was around 2700MHz with ~1.33V.
These were equivalent to the retail cpus, however I believe the retails will be atleast a bit better.

Naturally some of them liked the VDDNB higher some lower, depending if the chip was a "keeper" or a leaker

NCLK seems to be the only limiter for MEMCLK on aircooling.
On K15h the NCKL domain cannot work at lower speeds than the MEMCLK (so the minimum is 1:1).
High MEMCLK frequency stresses NB much more than lower frequencies. Therefore it appears that the NB is doing worse than it actually is.
Surely cold helps reaching higher MEMCLK frequencies, how much that is impossible to say until there are chips that do 2700MHz+ NCLK at the same MEMCLK frequency (on air).

It seems that less compute units / cores activated will make the NCLK go further.
Atleast the average FX-4100 does better in the terms of NB overclockability.

There will be a special bios build for Crosshair V quite soon.
It will feature all of the gimmicks found in previous "ES" builds + some new improvements.
Some of the timing parameters and NB related stuff will be improved to make it fly even higher in proper conditions