AMD Zambezi news, info, fans !

[tiborrr]

I am in dire need of Giga 990FXA-UD7 beta bios G2. Some one please share. Any BIOS that can disable Dozer Modules.

[tbone8ty]

yeah tigerdirect and newegg have the 8120 off and on. its on as of now lol

edit: to late lol its definitly off

[tiborrr]

Really need that bios badly, I have 40L here on standby and i'm stuck at 6.98GHz with FX-4100:

[kaktus1907]

Source

[Blue Storm]

@tiborrr: did you try to deactivate one module? I was stuck at ~6,65 GHz with FX-6100 and three modules, but with deactivated two modules I reached 7,652 GHz

//eh sorry, I didn't see your first post on this page...so, it's really bad

[The Stilt]

Source

I would like to know how they managed to run with three cores...

[Mechromancer]

Source

This is useful information. A Zambezi BF3 rig isn't a bad idea. BD can handle high resolutions as well as seen in previous tests.

[informal]

I would like to know how they managed to run with three cores...

I guess they disabled half modules in bios(so 2M active out of 4) and used affinity setting in task manager to make the game run only on cores #0,1,2 but not #3.

[FlanK3r]

someone with Sabertooth under LN2? At Sabertooth I can break 7 GHz with FX (and 2 active modules)

[demonkevy666]

http://www.ocaholic.ch/xoops/html/mo...id=537&page=10

I was wondering why street fighter 4 here looks so good on Bulldozer then I found this

http://www.pcgameshardware.com/aid,6...n-detail/News/

[FlanK3r]

Good review of FX-8150 with patchs at Techpowerup
logo.jpg

http://www.techpowerup.com/reviews/AMD/FX8150/

[RussC]

Wow, those power consuption #s are shocking. I dont remember them being quite that bad. Oh my my.....

RussC

>Good review of FX-8150 with patchs at Techpowerup

[tbone8ty]

Good review of FX-8150 with patchs at Techpowerup
logo.jpg

http://www.techpowerup.com/reviews/AMD/FX8150/

great review, like the OC results for bulldozer for metro 2033, beats 2600k at 4.8ghz

but damn power consumption blows!

[BeepBeep2]

Wow, those power consuption #s are shocking. I dont remember them being quite that bad. Oh my my.....

RussC

>Good review of FX-8150 with patchs at Techpowerup

great review, like the OC results for bulldozer for metro 2033, beats 2600k at 4.8ghz

but damn power consumption blows!

I'd been saying this since release. To get the same performance with BD that you do with Phenom II you have to use more power and its a node smaller. Essentially, I was told to shut up and the blame was put on GloFo. System power consumption minus ~70w for other components, you're at about 170-175w for CPU power consumption with the Bulldozer CPU and 90-95w for i7 2600K.

In essence, the BD CPU consumes 90% more power than a 2600K and does not win any tests except for:
Pov Ray 3.7b by 2.9%
7-Zip 32M by 6.5% (Loses to 2120 in real-world compression time)
WinRAR Benchmark by 22% but the 2600K's HyperThreading is not working because of Core Parking bug that was also seen on BD when MS released patch the first time (Only matches in real-world compression time)
TrueCrypt by a healthy 25%.

The rest of the benchmarks are losses between 5 and 50% against 2500/2600K while being 90% less efficient ...too many people mistake TDP for power consumption. They are not the same thing...TDP is a measure of heat that a heatsink needs to be able to dissipate, and power consumption is never listed.

Their CPU is at 4.7 GHz and 1.41v: At idle, it uses almost as much as a stock 1100T at full load.



But who knows, maybe Bulldozer is only "good" at 16 cores and less than 3GHz, where it is at the server market competing with 2GHz 8 core intel CPUs that do more work with less power than it anyway.

[FlanK3r]

there is problem with 32nm and GF, because I tested 3 FXs and between the worst and the best is about 60W!

[demonkevy666]

there is problem with 32nm and GF, because I tested 3 FXs and between the worst and the best is about 60W!

am I the only who think bulldozer would not have been a power hog for it's die size on 45nm O.o!?!?!

[Tomasis]

architecture and power usage are different things.

It needs to remember to keep discussing those topics separately

[BeepBeep2]

architecture and power usage are different things.

It needs to remember to keep discussing those topics separately

So GloFo's 32nm SOI is up to 90% worse than Intel's 32nm bulk and around 33% worse than GloFo 45nm SOI? (~50% shrink, same power consumption between FX-8150 and 1100T CPUs, single core turbo broken for 1100T CPUs making FX ST/gaming performance look better than it should)

That certainly can't all be true.

I mean, you could compare Llano 3850/3870K to FX-4100 but there are several problems with that...
1. Llano can be under volted at stock voltage to hell and back, most cases around 1.1v @ about 3 GHz and post spectacular stock performance/watt vs stock Phenom II X4

2. FX series can barely run stock speeds at stock voltage (Think about 3.6 @ 1.25v).

3. Llano has GPU on die, uses different socket and different (weaker/less efficient) VRM components on cheaper boards and doesn't turbo / overclock worth bullcrap.

4. Llano is missing L3 cache, and it's added ~5% performance despite having performance similar to same clocked Phenom II counterparts (assuming you take advantage of Llano's IMC).

5. FX is more voltage tolerant, feed it with a stupid 1.6v+ on water and it will scale to oblivion as long as it doesn't trip your power breaker or catch your motherboard on fire. Llano just stops around 1.45-1.5v

[Tomasis]

why cannot it be true? thats is the problem, isnt?

Did you read the news that amd ditches SOI and sold off remaining to GF.

Does it not tell something to you?

For you is it supposed to cheapier, easier, faster to produce every lower nm for cpu?

[BeepBeep2]

why cannot it be true? thats is the problem, isnt?

Did you read the news that amd ditches SOI and sold off remaining to GF.

Does it not tell something to you?

For you is it supposed to cheapier, easier, faster to produce every lower nm for cpu?

Okay, first of all AMD did not "ditch SOI".

AMD is running Trinity's successor on 28nm bulk, and AMD changed agreement with GlobalFoundries from "pay per die" to "pay flat, non-variable price per wafer". Furthermore, the 28nm CPUs will still be made at GlobalFoundries. IMO the whole 28nm s*** is just to stop gap as far as low-power oriented products (read: APU) vs Intel's upcoming 22nm.

Where the hell do you think AMD will make their 22nm CPUs too? ...on TSMC 20nm bulk? SOI offers better power efficiency, less steps to manufacturing and higher performance over bulk, and by the time we get to ~16nm there will be no choice but to run everything SOI.

Intel has done well with bulk but they have a lot more R&D money too. You can't blame 90% worse efficiency at the same node with better technology though on the fab alone.

[Tomasis]

ok so did you read Alpha article? point me to article re 16nm will be everything on SOI? Intel, tmsc will turn to SOI?

are you sure that zambezi and llano are manufactured under the same way? for me Llano doesnt clock well. My old Phenon II on 45nm clocks better, like 500mhz at least.

Due lower voltage tolerance, I suspect Llano is manufactured at different way, like "cold treatment". Brazo doesnt clock high either, but is suitable as low watt cpu. I dont think it depends on built in gpu, imho. At 32nm, Llano is supposed clock better than Phenom II.

Without L3, it makes easier to produce cpus also. Producing high density caches never been Amd/Gf strong point compared to Intel's. Hence more transistors.

Thats why Im excited to see Trinity, it will be interesting!

[AbortRetryFail?]

ok so did you read Alpha article? point me to article re 16nm will be everything on SOI? Intel, tmsc will turn to SOI?
. . . .

We are simply reaching a point in design where soon, copper interconnects can no longer be used, hence, the development of photronics or "optical transistors".

Intel has successfully demonstrated photronics in their labs using SOI layers. Of course, there may well be other advances and changes, but from what I've seen they like the overall properties.

There is something in the insulating layers that prevents 'leakage' and allows for the use of different wave lengths of light.

As far as the Alpha article, it seemed to me that TS responded to the question on SOI by saying it was "on-board" for all product at 28nm, and that the roadmap had not changed.

I might be reading things differently than others but it seems some folks are parsing his statements into some things that just are not there.

[BeepBeep2]

ok so did you read Alpha article? point me to article re 16nm will be everything on SOI? Intel, tmsc will turn to SOI?

are you sure that zambezi and llano are manufactured under the same way? for me Llano doesnt clock well. My old Phenon II on 45nm clocks better, like 500mhz at least.

Due lower voltage tolerance, I suspect Llano is manufactured at different way, like "cold treatment". Brazo doesnt clock high either, but is suitable as low watt cpu. I dont think it depends on built in gpu, imho. At 32nm, Llano is supposed clock better than Phenom II.

Without L3, it makes easier to produce cpus also. Producing high density caches never been Amd/Gf strong point compared to Intel's. Hence more transistors.

Thats why Im excited to see Trinity, it will be interesting!

1. Parasitic Capacitance becomes a larger problem as nodes shrink. Maybe you didn't know, but a big reason Intel is using their "3D Gate" is to combat it, meaning they can stick with bulk silicon instead of having to move to the more expensive SOI just yet. Of course, they would be at an even greater advantage if they moved to SOI on 22nm (they would have a huge advantage). Parasitic Capacitance is directly related to parasitic delay which in turn I believe is closely related to gate delay.

Unless more semiconductor companies move to similar 3D tri-gates or very least FinFET, I assure you that they will be using SOI by 16nm...

2. Yeah, you're right. However Llano has a GPU on the die, weaker motherboards and VRM, also missing 200 pins. (Not sure how they pulled that off.)

3. I suspect the same thing, however Llano is not the same core as Phenom II and that is the reason it has 10% IPC increase in some applications. Still, there is a lot more to look at here than GlobalFoundries side.

4. Yes, but in this case it was left out because of cost/die size. I'm not sure how many % of power consumption is used purely by L3, but Llano could easily have matched a Phenom II X4 CPU running 200 MHz higher if it had L3, IMO.

5. I'm excited to see Trinity as well. Hopefully Piledriver brings substantial improvements, as it currently looks right know that it won't be "run cold".

[demonkevy666]

1. Parasitic Capacitance becomes a larger problem as nodes shrink. Maybe you didn't know, but a big reason Intel is using their "3D Gate" is to combat it, meaning they can stick with bulk silicon instead of having to move to the more expensive SOI just yet. Of course, they would be at an even greater advantage if they moved to SOI on 22nm (they would have a huge advantage). Parasitic Capacitance is directly related to parasitic delay which in turn I believe is closely related to gate delay.

Unless more semiconductor companies move to similar 3D tri-gates or very least FinFET, I assure you that they will be using SOI by 16nm...

2. Yeah, you're right. However Llano has a GPU on the die, weaker motherboards and VRM, also missing 200 pins. (Not sure how they pulled that off.)

3. I suspect the same thing, however Llano is not the same core as Phenom II and that is the reason it has 10% IPC increase in some applications. Still, there is a lot more to look at here than GlobalFoundries side.

4. Yes, but in this case it was left out because of cost/die size. I'm not sure how many % of power consumption is used purely by L3, but Llano could easily have matched a Phenom II X4 CPU running 200 MHz higher if it had L3, IMO.

5. I'm excited to see Trinity as well. Hopefully Piledriver brings substantial improvements, as it currently looks right know that it won't be "run cold".

No HT link to NB = less pins. Pci-express link uses less pins "I think".
What ever 10-20 watts is about L3 caches power usage, judging from Propus vs Deneb.

[undone]

Stepping C0 FX ES
http://setiathome.berkeley.edu/sah/s...hostid=5747954