Phenom B3 stepping black edition full Reveiw @ FiringSquad

[Killrose]

Full Reveiw of the B3 Phenom

FiringSquad

 

[zach0624]

looks like amd is going in the right direction and the TLB bug seems fixed. However many of the tests they ran were limited by other parts of the system but in the ones that weren't phenom put up some impressive numbers at a low a price point.

 

[Sylvanas]

Seems AMD have officially launched triple cores, the B3's and 65W quads today. Would have liked to have seen more investigation with OCing with the FS review....like what NB speed/multiplier was used when OC'd, memory settings (unganged/ganged etc). I will be purchasing these 9850's when they come in stock here...ETA early April.

techreport got 2.9-3ghz- good stuff.

 

[harpoon84]

Its a step in the right direction, it appears overclocking headroom has improved slightly with B3, and of course the overblown TLB bug is finally put to rest. Unfortunately it appears the biggest bottleneck in Phenom performance - the L3 frequency, hasn't been improved from B2. Until AMD can get the L3 running in sync with the core clock they won't be truly competitive against C2Qs.

I feel its priced a little on the high side as well, since the Q6600 price will drop to ~$230 after April pricecut. I think ~$200 would be a more appropriate price for the 9850BE.

 

[Sylvanas]

Originally posted by: harpoon84
Its a step in the right direction, it appears overclocking headroom has improved slightly with B3, and of course the overblown TLB bug is finally put to rest. Unfortunately it appears the biggest bottleneck in Phenom performance - the L3 frequency, hasn't been improved from B2. Until AMD can get the L3 running in sync with the core clock they won't be truly competitive against C2Qs.

I feel its priced a little on the high side as well, since the Q6600 price will drop to ~$230 after April pricecut. I think ~$200 would be a more appropriate price for the 9850BE.

Indeed, I would assume we will have to wait until 45nm for a Sync'd core/L3 frequency. Some users on XS in this thread have tested the NB frequency up to 2.4ghz+ and it does indeed provide significant benefits, I would find specifics but it is a 62 page thread and the mass of information is simply staggering- but the potential is there. I will post my experiences with the Northbridge OC when I receive my 9850.

 

[Cookie Monster]

So only the phenom 9850 has its NB/L3 frequency up from 1.8GHz to 2GHz?

 

[Sylvanas]

Originally posted by: Cookie Monster
So only the phenom 9850 has its NB/L3 frequency up from 1.8GHz to 2GHz?

I assume so, all the reviews have only really tested the 9850 from what I've read.

AT's review is up

 

[jones377]

Is it possible to change the clockspeed of the L3/NB independently of the CPU cores and/or HT?

 

[Idontcare]

Originally posted by: jones377
Is it possible to change the clockspeed of the L3/NB independently of the CPU cores and/or HT?

I am curious about this too.

I am also curious whether anyone has reported overclocking a Phenom with vaporphase cooling.

Do they get substantially more overclocking when cooled to -40C?

I've got my vapoLS set aside waiting for stock to come in so I can find out.

 

[Rubycon]

Tri-Athlon?

 

[heyheybooboo]

Originally posted by: jones377
Is it possible to change the clockspeed of the L3/NB independently of the CPU cores and/or HT?

Yup is my understanding - either in AOD or the BIOS. Best I can tell, however, is that as CPU clock speed goes up you still want to adjust L3/NB down for max stability.

BUT ... results are all over the place. Some folks present 2.4GHz NB with low volts - others require a slight bump. Some have shown 2.5-2.6GHz NB with big jump in volts (and special cooling)

Some folks run 'synchronous' - others don't. By that I mean that they try and keep the hypertransport and northbridge multipliers the same. Some folks have shown increased stability by keeping not only the hypertransport and northbridge multipliers the same, but the cpu multi as well.

Hard to find any overall consistency as of yet. Starting off it seems folks drop the ram divider, reduce the hypertransport and northbridge to x8 and begin cranking up the cpu fsb/multi.

Once they hit their fsb wall they start upping the cpu volts. The cpu volt wall seems to be 1.43-1.45v with some reporting .025v+ vdroop.

This is where the unlocked multi of the BEs come into play. As an example ...

250 x 10.5 = 2.625GHz
220 x 12 = 2.640GHz

With the hypertransport and northbridge to x8 you have a range difference of 1760-2000MHz between two different cpu multis. Bumping the hypertransport and northbridge each to x10 and the range becomes 2200-2500MHz depending upon cpu multi.

If you are using the higher multi on the BE it allows you to keep the hypertransport and northbridge speed down, reduce their voltages and help keep EMI down.

That's my theory and I'm stickin' to it

 

[v8envy]

Something is bugging me about that review. In all the game tests a 2.8 ghz OC on the Phenom tied a stock clock Q6600 only when GPU limited (which was very often) but otherwise trailed significantly.

That doesn't add up. The B2 Phenom (with TLB fix disabled) only has a 10-15% IPC disadvantage vs a Q6600 based on CPU intensive tests at the same clock rate. There is no way a 16% faster clock Phenom is losing to a stock clocked Q6600 by 10% in games (e.g, UT3). Something else is afoot at the circle K.

 

[Cookie Monster]

NB/L3 frequency can be OCed independently to the core clock frequency.

How to OC the NB/L3 clock.

 

[Sylvanas]

Originally posted by: Cookie Monster
NB/L3 frequency can be OCed independently to the core clock frequency.

How to OC the NB/L3 clock.

Great post. It seems all the review sites are using a rather quick and unscientific way to OC these Phenom's...its not simply like C2D where you whack the multiplier up and the voltage and off you go. There's NB + HT +individual core clock speed that seems neglected in every review which impact a lot on performance.

 

[CTho9305]

Originally posted by: Cookie Monster
NB/L3 frequency can be OCed independently to the core clock frequency.

How to OC the NB/L3 clock.

Does that mean you can use dividers to lower the frequency to 200 * 8 (with FID) / 8 (with DID) and run the chip at 200MHz for ultra-low-power operation?

 

[exar333]

God the Phenom still looks like crap. Everyone is reviewing these and stating they are so much better than the B2s, when all that has been fixed is the TLB issue.

Another review by Hothardware illustrates the power consumption of the Phenom 2.5 (load) @ 346 and the Q6600 (stock-load) @ only 255 (I am not really even going to mention the QX9650 being 206).

This reminds of me when the P4E's were using 2x as much power and every person and their brother complained about the power usage and heat.

I built by first Intel rig EVER last year and always loved AMD (still do) but the Phenom is still crap. It is no wonder they won't up the clockspeed on these guys; I would be curious to see the power usage at 2.8 and 3.0.

edit:SP


Hothardware - System Power Consumption

 

[CTho9305]

Originally posted by: ExarKun333
God the Phenom still looks like crap. Everyone is reviewing these and stating they are so much better than the B2s, when all that has been fixed is the TLB issue.

Another review by Hothardware illustrates the power consumption of the Phenom 2.5 (load) @ 346 and the Q6600 (stock-load) @ only 255 (I am not really even going to mention the QX9650 being 206).

This reminds of me when the P4E's were using 2x as much power and every person and their brother complained about the power usage and heat.

I built by first Intel rig EVER last year and always loved AMD (still do) but the Phenom is still crap. It is no wonder they won't up the clockspeed on these guys; I would be curious to see the power usage at 2.8 and 3.0.

edit:SP


Hothardware - System Power Consumption

Why would idle power consumption differ for B2 with the TLB patch enabled vs disabled? If the system is idle, why would it matter what the TLB does? Hot Hardware shows a 6 watt difference... yet the load power didn't really change. That makes absolutely no sense to me. The load power, for a real application, should be lower with the TLB patch enabled (because the CPU spends more time waiting for data from memory), right?

 

[exar333]

Originally posted by: CTho9305

Originally posted by: ExarKun333
God the Phenom still looks like crap. Everyone is reviewing these and stating they are so much better than the B2s, when all that has been fixed is the TLB issue.

Another review by Hothardware illustrates the power consumption of the Phenom 2.5 (load) @ 346 and the Q6600 (stock-load) @ only 255 (I am not really even going to mention the QX9650 being 206).

This reminds of me when the P4E's were using 2x as much power and every person and their brother complained about the power usage and heat.

I built by first Intel rig EVER last year and always loved AMD (still do) but the Phenom is still crap. It is no wonder they won't up the clockspeed on these guys; I would be curious to see the power usage at 2.8 and 3.0.

edit:SP


Hothardware - System Power Consumption

Why would idle power consumption differ for B2 with the TLB patch enabled vs disabled? If the system is idle, why would it matter what the TLB does? Hot Hardware shows a 6 watt difference... yet the load power didn't really change. That makes absolutely no sense to me. The load power, for a real application, should be lower with the TLB patch enabled (because the CPU spends more time waiting for data from memory), right?

I wonder if power is still being used in the L3 cache with the TLB patch enabled? Please dont flame me for this remark, just speculation on my part. Maybe the cpu can manage the power better with the cache enabled and either use or reduce power as needed. Anyone have more information?

 

[aigomorla]

i just cant see how AMD expects to compete against the Q6600 even.

And even intels own Q9450 series is having a hard time catching up to a Q6600 G0 in overclocking numbers.

Which leads me to think, AMD has issues or is best to be aiming in a OEM market. But i still think they would get killed by a kentsfield in that market even.

 

[exar333]

Originally posted by: aigomorla
i just cant see how AMD expects to compete against the Q6600 even.

And even intels own Q9450 series is having a hard time catching up to a Q6600 G0 in overclocking numbers.

Which leads me to think, AMD has issues or is best to be aiming in a OEM market. But i still think they would get killed by a kentsfield in that market even.

If they did a straight die-shrink of the Q6600 to 45nm and kept the bus speed and multiplier the same, they would be insane overclockers. The problem is the low multiplier keeping the 45nm quads down, but thats nothing new. Intel is forcing us to shell out the cash for the extreme cpus...

now only if Intel followed AMD's lead and offered the unlocked multipliers as no extra charge...*wink*

 

[Viditor]

Originally posted by: aigomorla
i just cant see how AMD expects to compete against the Q6600 even.

And even intels own Q9450 series is having a hard time catching up to a Q6600 G0 in overclocking numbers.

Which leads me to think, AMD has issues or is best to be aiming in a OEM market. But i still think they would get killed by a kentsfield in that market even.

You are speaking of overclockers, which is less than 1% of the market...I really don't think AMD is too worried about that.
Where the B3 Phenoms should do exceedingly well is in the sweet spot (mainstream sales).

 

[dmens]

Originally posted by: Viditor
You are speaking of overclockers, which is less than 1% of the market...I really don't think AMD is too worried about that.
Where the B3 Phenoms should do exceedingly well is in the sweet spot (mainstream sales).

lol yeah, call whatever market segment where phenom is the least behind the "sweet spot".

and even at that price point AMD has no clear cut advantage in any metric, so why would it do exceedingly well?

 

[harpoon84]

Originally posted by: Viditor
You are speaking of overclockers, which is less than 1% of the market...I really don't think AMD is too worried about that.
Where the B3 Phenoms should do exceedingly well is in the sweet spot (mainstream sales).

The official Q6600 price will drop to $224 in April, if AMD wants to truly compete in the mainstream quads they'll have to drop Phenom prices again. The only way Phenom will sell well is to be cheaper than the Q6600. Even if sales numbers are healthy, will the margins be?

 

[Viditor]

Originally posted by: harpoon84

Originally posted by: Viditor
You are speaking of overclockers, which is less than 1% of the market...I really don't think AMD is too worried about that.
Where the B3 Phenoms should do exceedingly well is in the sweet spot (mainstream sales).

The official Q6600 price will drop to $224 in April, if AMD wants to truly compete in the mainstream quads they'll have to drop Phenom prices again. The only way Phenom will sell well is to be cheaper than the Q6600. Even if sales numbers are healthy, will the margins be?

That's always the way it is...margins vs competition. Will AMD make enough margin when they drop the price, will Intel make enough margin @$224?
Without knowing the costs (R&D, materials, yields, marketing, shipping, etc...), then we really can't know the answer to your question...
For example, while Intel's R&D costs are MUCH higher than AMD, they have been shipping smaller dies. Which has the better margin? Who knows...though lately I give the nod to Intel.
I have a few spreadsheets which guesstimate it, but at the end of each quarter I can guarantee you I am never exactly correct (though I'm usually within a few points).

I agree that AMD will remain aggressive on their pricing and we should probably continue to see the Phenom lead in the value quad core sector...

However, I would remind you that AMD can temper their pricing because of one very remarkable first...
For the very first time EVER (and I can't believe I haven't seen more comments on this), AMD actually has the clearly better chipset and platform!
Intel has always been the leader in this segment, with occasional challenges by Via and nVidia...but AMD? Don't make me laugh...
780G is a true breakthrough and it's importance should be well noted (and considered a selling point for Phenom).

 

[Idontcare]

Originally posted by: ExarKun333

Originally posted by: CTho9305

Originally posted by: ExarKun333
God the Phenom still looks like crap. Everyone is reviewing these and stating they are so much better than the B2s, when all that has been fixed is the TLB issue.

Another review by Hothardware illustrates the power consumption of the Phenom 2.5 (load) @ 346 and the Q6600 (stock-load) @ only 255 (I am not really even going to mention the QX9650 being 206).

This reminds of me when the P4E's were using 2x as much power and every person and their brother complained about the power usage and heat.

I built by first Intel rig EVER last year and always loved AMD (still do) but the Phenom is still crap. It is no wonder they won't up the clockspeed on these guys; I would be curious to see the power usage at 2.8 and 3.0.

edit:SP


Hothardware - System Power Consumption

Why would idle power consumption differ for B2 with the TLB patch enabled vs disabled? If the system is idle, why would it matter what the TLB does? Hot Hardware shows a 6 watt difference... yet the load power didn't really change. That makes absolutely no sense to me. The load power, for a real application, should be lower with the TLB patch enabled (because the CPU spends more time waiting for data from memory), right?

I wonder if power is still being used in the L3 cache with the TLB patch enabled? Please dont flame me for this remark, just speculation on my part. Maybe the cpu can manage the power better with the cache enabled and either use or reduce power as needed. Anyone have more information?

CTho9305 is right, you are thinking about it backwards. The TLB patch basically forces data to the L3 instead of staying resident in the L2...kind of like turning off the L2 but not nearly so severe.

So L3 is powered on and if anything getting updated more than usual...but because the individual cores themselves ought to be idling more often with the TLB patch than without the patch (that is what causes the lower performance) the overall average power consumption ought to be lower for the B2 patched system.

CTho9305,
As to the 6W delta...I'd call that noise. The PSU efficiency itself could be meandering a mere 6W and still be well within specs. 20W (10%) would raise my eyebrow, but not a mere 6W discrepancy.