Rumour: Bulldozer 50% Faster than Core i7 and Phenom II.

Page 51 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.
Status
Not open for further replies.

Bearach

Senior member
Dec 11, 2010
312
0
0
SB can take a lot of punishment

1687033.png

I am not denying that Sandy Bridge is a good clocking chip, what I'm saying is that it wasn't a realistic overclock. How long would SB last on that vcore?

While that is "reachable" by some peoples' standards, Would you seriously run your chip at 1.75v 24/7?

The post was relating to how well SB can clock and I believe it should have been to actual running values, and not showing how far you can go unstable.
 

Bearach

Senior member
Dec 11, 2010
312
0
0
I might be sounding aggressive, but what i'm trying to point out is that although what i can attain as 24/7 stable might be out of reach for the majority it's still a quality the chip posses as long as the consumer has the ability to keep it in a stable enviroment. So it shouldn't be discredited just because it's out of reach o the majority. And yes i'm well aware that enthusiasts make up a very little sum of Intel and AMD revenue.

Anyway i think i'm dragging the thread off topic.

What you posted though wasn't actually your 24/7 stable overclock and that's the only reason I had issue with it.

So in effect you also posted what was out of reach for you too.

Speculating where your stable clock is, is just that. It could be just a little more higher and then you need way too much voltage to push it further, we don't know... so that 900MHz headroom is a big unknown.

Nice talking to you though, and excellent clocking of that chip and it was at least nice to see that the system was benchmark stable for you.
 

Arkadrel

Diamond Member
Oct 19, 2010
3,681
2
0
I just want SB to give Intel a run for their money and start producing some fantastic CPUs!


Freudian slip?

I just want SB... first 4 words :)
your brain goes... what do I want? SandyBridge.

Supposed to say you want bulldozer/AMD to give intel a run for its money right?


Also unless Im miss informed, I thought most people couldnt get stable 5ghz with their SBs... only very few got that far on air (lucky ppl with good chips).
 
Last edited:

smakme7757

Golden Member
Nov 20, 2010
1,487
1
81
What you posted though wasn't actually your 24/7 stable overclock and that's the only reason I had issue with it.

So in effect you also posted what was out of reach for you too.

Speculating where your stable clock is, is just that. It could be just a little more higher and then you need way too much voltage to push it further, we don't know... so that 900MHz headroom is a big unknown.

Nice talking to you though, and excellent clocking of that chip and it was at least nice to see that the system was benchmark stable for you.

Fair enough, point taken. I appreciate the civilised debate, it was nice talking to you as well! :).


Freudian slip?

I just want SB... first 4 words :)
your brain goes... what do I want? SandyBridge.

Supposed to say you want bulldozer/AMD to give intel a run for its money right?


Also unless Im miss informed, I thought most people couldnt get stable 5ghz with their SBs... only very few got that far on air (lucky ppl with good chips).

Yea i was supposed to write BD :). As for what i want, i go where the performance is, i have no brand loyalty although Corsair is preferred as i've only had good experiences with both their products and customer service.

5Ghz is no doubt the magic number and on most forums i frequent there is only a few not reaching 5Ghz, with most going much higher than that. As always though it's chip dependant and your milage will vary. But like i always say your possiblities are greater with more knowledge: Stay in school kids!
 
Last edited:

maddie

Diamond Member
Jul 18, 2010
5,152
5,540
136
So I will do also some speculations on pricing (street prices):

4M/8 core BD 300-600 USD - top performing x64 CPU
3M/6 core BD 200 USD competing with SB 2600 ~ on par
2M/4 core BD 100 USD competing with core i3/i5

That is also the reason why AMD starts first with the 8 core BD and the 4/6 core Bulldozer CPUs will follow later when the process has gained better yields.

The big problem for AMD will be that the performance crown will not last long. When Intel pulls out the Sandy Bridge EN with 8C/16T, the crown goes back to Intel and I do not know if AMD wants to compete with the extreme parts at all. If they want they could issue a 12 core.

And regarding overclocking I am sure that the Bulldozer parts will break all records regarding the absolute GHz number. But regarding overclocking it is always important how much overclock you can get on air and without increasing the core (or any other) voltage. Everything else is just meaningless.

I want resilient computers even if the room temperature goes up. I do not overclock my PCs (except the old Celeron 300 -> 450 one) and I am an enthusiast user. Sure some are interested in that and do overclock their CPUs but this is not important business wise.
Isn't that the reverse of what normally happens?
 

Ajay

Lifer
Jan 8, 2001
16,094
8,113
136
When Ivory Bridge and Z68 come out and will beat IMO bulldozer 8 core 16 with HT.

I think 8 core can beat 12 core bulldozer,, just imo. :D

From TechPowerUp

Here are the first pictures of Intel's new high-end CPU socket, the 2011-pin land grid array (LGA2011). A selection of pictures of an unannounced motherboard by MSI made it to the internet. LGA2011, coupled with a new chipset, the Intel X68 Express, will drive the company's new high-end and enthusiast-grade processors that feature 6, 8, or 12 cores, and quad-channel DDR3 memory controllers. At first sight, the LGA2011 is huge! Its retention clip looks to be completely detachable by unhooking the retention bars on either sides. With all LGA sockets till date, you needed to unhook one retention bar, letting you open the retention clip along a hinge.

A consumer dodeca-core CPU :eek:
 

PreferLinux

Senior member
Dec 29, 2010
420
0
0
OEMs dont care about overclocking, so thats a non issue (since thats by far the biggest market).

40-50% is a big claim.
i7-2600k comes at 3.4Ghz with turbo up to 3.8ghz.
a 50% overclock would mean 5.1Ghz with turbo of up 5.7ghz.

I doubt most people that buy a i7-2600k get anywhere near 5.1ghz, but yes the Sandy Bridges do overclock pretty well.

The Bulldozers might too...no way to know until their released.
Also even non-OEMs, that buy cpus dont always overclock or wish too.
Its probably just a small group of us gamers.. that enjoy doing it.


if it isnt, it ll sell for lower. Why shouldnt they be able to sell it cheaper?
150mm^2 4"core" bulldozer vs Sandy bridge 4 "core" of 216mm^2.

Its bound to be cheaper to produce (~150mm^2 vs 216mm^2).

As long as you get the performance you pay for, its a great product for the end user.
Yes, if you are using the same manufacture. As Intel is doing it for themselves, I doubt AMD will get it cheaper from a company that is putting on their markup.

So I will do also some speculations on pricing (street prices):

4M/8 core BD 300-600 USD - top performing x64 CPU
3M/6 core BD 200 USD competing with SB 2600 ~ on par
2M/4 core BD 100 USD competing with core i3/i5

That is also the reason why AMD starts first with the 8 core BD and the 4/6 core Bulldozer CPUs will follow later when the process has gained better yields.

The big problem for AMD will be that the performance crown will not last long. When Intel pulls out the Sandy Bridge EN with 8C/16T, the crown goes back to Intel and I do not know if AMD wants to compete with the extreme parts at all. If they want they could issue a 12 core.

And regarding overclocking I am sure that the Bulldozer parts will break all records regarding the absolute GHz number. But regarding overclocking it is always important how much overclock you can get on air and without increasing the core (or any other) voltage. Everything else is just meaningless.

I want resilient computers even if the room temperature goes up. I do not overclock my PCs (except the old Celeron 300 -> 450 one) and I am an enthusiast user. Sure some are interested in that and do overclock their CPUs but this is not important business wise.
Beating 8.2 GHz on a Cedar Mill Netburst Celeron that is designed purely for extremely high clocks? Maybe, seeing the AMD record is ~7 GHz, but not likely.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
63
91
Beating 8.2 GHz on a Cedar Mill Netburst Celeron that is designed purely for extremely high clocks? Maybe, seeing the AMD record is ~7 GHz, but not likely.

They did 7GHz with 45nm but you don't think hitting 8.2GHz is reasonable on 32nm?
 

PreferLinux

Senior member
Dec 29, 2010
420
0
0
Intel did 8.2 GHz on 65 nm. But yes, I think it is possible, but then I'd expect SB-E to do better too, probably.
 

HW2050Plus

Member
Jan 12, 2011
168
0
0
Isn't that the reverse of what normally happens?
Right if you think about that a larger part is more likly to contain a failure. But on the other hand you have to consider also the selling price. You have a 75% higher failure rate but three times of the selling price. That is why this makes sense with a new process that does not yet have mature yields.

In addition since the yields are not enough to fully fill the market they sell still lower performing parts with their mature 45 nm line. This is another reason why AMD does it this way.

However the first issue with higher failure rate is the reason why AMD makes Interlagos as a MCM.

Beating 8.2 GHz on a Cedar Mill Netburst Celeron that is designed purely for extremely high clocks? Maybe, seeing the AMD record is ~7 GHz, but not likely.
This is extremly likly. AMD record is 7 GHz. Now just add the design based clock advantage of 30% (22/17) to this and you have already 9.1 GHz without the gain from a much better process (32 nm, high-k, etc.). And that is what I predict: You will see 9+ GHz from AMD Bulldozer parts in extreme overclocking. The Bulldozer parts will already emerge with higher clocks than Cedar Mill. And it is a design for extremly high clocks similar to Cedar Mill.

Therefore though there is no guarantee to get there BD has the best preconditions to break that record and the 9 GHz mark.

The lowest gate delay of any processor still remains the Prescott core based Pentium 4's. That's irregardless of any process generation.
You mean gate count per pipeline stage.

There are two factors which make up the maximum frequency:
1.) Gate Fan Out per pipeline stage. With Bulldozer this dropped from 22 to 17 FO4.
2.) Gate Delay and that is a parameter dependend on process parameters (improved to 32 nm, improved with high-k) and temperature.

The frequency you get is 1/(GateFOPerStage * GateDelay).
With Bulldozer all parameters dropped! The first parameter alone was dropped by 30%! It is unclear how much the second parameter dropped.

Sure with Cedar Mill the first one is even lower but the second is higher. And Bulldozer is the first CPU since Cedar Mill (and predecessors) which comes close to it regarding the first parameter.
 

Anexate

Member
Feb 8, 2011
34
0
0
www.facebook.com
June 20th - begins the shipments of AMD FX-series high-end desktop microprocessors

AMD's Bulldozer:
-4 eight-core AMD FX8000
-2 six-core AMD FX6000
-2 quad-core AMD FX4000
 

IntelUser2000

Elite Member
Oct 14, 2003
8,686
3,787
136
You mean gate count per pipeline stage.

There are two factors which make up the maximum frequency:
1.) Gate Fan Out per pipeline stage. With Bulldozer this dropped from 22 to 17 FO4.
2.) Gate Delay and that is a parameter dependend on process parameters (improved to 32 nm, improved with high-k) and temperature.

If the gate delay metric quoted was FO4, then its a process independent metric. Prescott was supposedly at 12.5 FO4 delays at 90nm. Penryn was 18 or something.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
63
91
The lowest gate delay of any processor still remains the Prescott core based Pentium 4's. That's irregardless of any process generation.

Congrats on making Elite Member! :thumbsup:

You were long over-due for the recognition IMO, and well deserve the distinction.
 

Dresdenboy

Golden Member
Jul 28, 2003
1,730
554
136
citavia.blog.de
There are two factors which make up the maximum frequency:
1.) Gate Fan Out per pipeline stage. With Bulldozer this dropped from 22 to 17 FO4.
2.) Gate Delay and that is a parameter dependend on process parameters (improved to 32 nm, improved with high-k) and temperature.

The frequency you get is 1/(GateFOPerStage * GateDelay).
With Bulldozer all parameters dropped! The first parameter alone was dropped by 30%! It is unclear how much the second parameter dropped.

According to GF slides, they expect a 40-50% advantage in perf/watt compared to the current 45nm process (although I'm not sure if they already include the Low-K update for 45nm):
GloFo_32nmSOIBlk_675.jpg

10x0901ub234weasf.jpg

So this is one variable. Another is gate delay. But I don't have GF's gate delay numbers for their processes at the likely operating voltage range, but it should roughly be 70% of their 45nm SOI gate delay (linear scaling factor).

Llano should have ~35M T per core, 110M incl. L2, so ~75M for the L2 and other stuff (e.g. power gating ring).
BD has 213M T per module, minus 150M for the L2+rest there remain 63M T for the 2 cores.
Both for BD and Llano AMD stated 0.8 to 1.3V. There is even more extensive clock gating in BD, which might reduce dynamic core power even more (maybe 20% vs. K10).
http://pc.watch.impress.co.jp/img/pcw/docs/430/341/html/13.jpg.html

Assuming nearly constant uncore power consumption (reduced by process, increased by # of T and clock frequency), a BD module on a X8 might have 150% of the power budget of a Thuban core and a BD core 75% respectively. With less L2 power (IIRC it is an 8T design in BD) but likely more relative leakage we might just look at roughly estimated core power relations:
BD : 63M T at 150% power (2 cores)
BD : ~32M T at 75% power (1 core avg)
hypothetical 10h 6C at 32nm: ~35M T at 100% power
10h: ~35M T at 166% power (inverse of 40% lower perf/watt)
So if a Thuban core has about 13.3W TDP (95W-15W uncore), it would use ~8W in 32nm at the same clock. This would be 100%.
A BD module at the same voltage and maybe 20% higher clock (FO4 reduction!) might use ~12W. 4 Modules would use ~50W then. Now scaling back to 80W again (20W per module) could add even another 20% of clock frequency headroom (about 45% combined). Thus a 4GHz Zambezi at 95W is not unlikely.
Could someone please check my numbers?
 

Martimus

Diamond Member
Apr 24, 2007
4,490
157
106
According to GF slides, they expect a 40-50% advantage in perf/watt compared to the current 45nm process (although I'm not sure if they already include the Low-K update for 45nm):
GloFo_32nmSOIBlk_675.jpg

10x0901ub234weasf.jpg

So this is one variable. Another is gate delay. But I don't have GF's gate delay numbers for their processes at the likely operating voltage range, but it should roughly be 70% of their 45nm SOI gate delay (linear scaling factor).

Llano should have ~35M T per core, 110M incl. L2, so ~75M for the L2 and other stuff (e.g. power gating ring).
BD has 213M T per module, minus 150M for the L2+rest there remain 63M T for the 2 cores.
Both for BD and Llano AMD stated 0.8 to 1.3V. There is even more extensive clock gating in BD, which might reduce dynamic core power even more (maybe 20% vs. K10).
http://pc.watch.impress.co.jp/img/pcw/docs/430/341/html/13.jpg.html

Assuming nearly constant uncore power consumption (reduced by process, increased by # of T and clock frequency), a BD module on a X8 might have 150% of the power budget of a Thuban core and a BD core 75% respectively. With less L2 power (IIRC it is an 8T design in BD) but likely more relative leakage we might just look at roughly estimated core power relations:
BD : 63M T at 150% power (2 cores)
BD : ~32M T at 75% power (1 core avg)
hypothetical 10h 6C at 32nm: ~35M T at 100% power
10h: ~35M T at 166% power (inverse of 40% lower perf/watt)
So if a Thuban core has about 13.3W TDP (95W-15W uncore), it would use ~8W in 32nm at the same clock. This would be 100%.
A BD module at the same voltage and maybe 20% higher clock (FO4 reduction!) might use ~12W. 4 Modules would use ~50W then. Now scaling back to 80W again (20W per module) could add even another 20% of clock frequency headroom (about 45% combined). Thus a 4GHz Zambezi at 95W is not unlikely.
Could someone please check my numbers?

It depends on how old those slides are. As mentioned here: http://forums.anandtech.com/showthread.php?t=2092185&highlight GF's low-k update over 45nm is not nearly as aggressive as they had planned. The Low-K material they had planned to use could not survive the packaging process, so they had to cut down on using it, and mostly used the same material they used at 45nm (which was actually pretty good).

I don't expect the Low-K to be much better than it was for 45nm at GF, since they made the decision to just change one metal layer to allow for better yields.
 

Idontcare

Elite Member
Oct 10, 1999
21,110
63
91
Thus a 4GHz Zambezi at 95W is not unlikely.

That is my expectation as well.

I would not be surprised if AMD ultimately reaches a retail SKU clocked at 4.5GHz at some point on 32nm, especially if they keep the 140W TDP SKU going at 32nm.
 
Status
Not open for further replies.