Phenom B3 stepping reviews

[voodoo_papa]

looking at picking up a new amd CPU to replace my old one, but would like one with the B3 stepping.

Do all of the Agena core CPU's have B3 stepping?

Im looking at this in particular
http://www.newegg.com/Product/Product.aspx?Item=N82E16819103244

thanks!

*edit* just found this. I guess newegg hasnt stocked them yet

"The new Phenom processor family on B3 stepping includes four models: 9550, 9650, 9750 and 9850 Black Edition"

 

[gpippas]

Nothing too taxing. But the old buisness pc is a single core 1.8ghz skt 754 sempron. 3 years ago it was pretty fast for work computer and never really got bogged down. But with software moving along as it does its really starting to show its age. So my reasoning for 4 cores for a work machine is that it will likely last another 3/4 years if not more.

I found in my experiance work machines last longer with more cores rather than high clock speeds.

 

[homerdog]

Interesting, but for a general use PC I'd rather buy a cheap dual core and upgrade in a couple of years if necessary. Saves on the power bill too.

 

[blackpanther26]

I wonder what the 9850 E would do if it's NB was set to 2.2-2.4GHz instead of 2.0GHz. I still think there is more hidden performance in the Phenom than what meets the eye.

 

[Mathos]

I wonder what the 9850 E would do if it's NB was set to 2.2-2.4GHz instead of 2.0GHz. I still think there is more hidden performance in the Phenom than what meets the eye.

Dont know, I'm tempted to get some money together, toss this 9600BE on ebay or something, and get a 9850BE just to not have to deal with the TLB bios crap, so I can do a bit less hassle with my OCing, and try a few others BIOSs that I know are stable but have the TLB fix so I don't use em.

 

[shabodah]

it would be interesting to see the northbridge overclocked as well

I wonder what the 9850 E would do if it's NB was set to 2.2-2.4GHz instead of 2.0GHz. I still think there is more hidden performance in the Phenom than what meets the eye.

^^This is what I've been saying all along. The 6400 x2 performs as it does because the IMC is runing at 3.2ghz. If you all recall, socket AM2 could not perform as well as 939 until the IMC was at 2.4ghz. Thus, the X3 & X4 chips are basically running into the same issue Intel is with the FSB. Sure, they have far more bandwidth to the chipset, but the bandwidth to the memory system is VASTLY hindered compared to what it could be. Honestly, I was quite upset to find out that AMD had decided to use two 64bit memory controllers instead of two 128 bit memory controllers. Sure, they can read and write at the same time, but the overall bandwidth is at best the same as an x2 of the same IMC frequency.

 

[yipsl]

I would go with a 9850. Currently the X3's are sheduled for OEM only. The 9850 also has the NB set at the original speed of 2ghz rather than the downclocked 1.8ghz.

Yes, that's what decides it for me. I did a bit more reading today on overclocking, and it seems it can hit 2.8 stable with regular voltage. With voltage changes (something I'm not inclined to do on a 780G board), it can only hit 2.9 stable. 3.0 still seems like a leap of faith.

I'll be hovering on Newegg once April 14th arrives. I want a 9850 and 4 gigs of performance DDR2 1066! Maybe then I won't be quite so CPU limited.

*edit* just found this. I guess newegg hasnt stocked them yet

"The new Phenom processor family on B3 stepping includes four models: 9550, 9650, 9750 and 9850 Black Edition"

April 7th is the release date on AMD's website:

http://www.amd.com/us-en/Corporate/VirtualPressRoom/0,,51_104_609,00.html?redir=CPPR01#phenom

 

[bmadd]

so how long does it usaully take for release to go to other countries? Cause im in australia so that means a good 2 weeks more which puts it at the end of april

 

[zenmaster]

I wonder what the 9850 E would do if it's NB was set to 2.2-2.4GHz instead of 2.0GHz. I still think there is more hidden performance in the Phenom than what meets the eye.

Likely Very Little.
The Increase from 1.8 to 2.0 (Just over a 10% boost) yielded about a 1% increase in perfomrance.
http://xbitlabs.com/articles/cpu/display/phenom-x4-9850_4.html#sect0

The reason the IMC is not higher is likely due to Chip limitations.
I recall reading about a numberof ES Samples having the IMC burn up when folks upped the voltage enough to try and hit 2.2 or 2.4. (2.0 Was the Default on the ES Samples.)

Most likely the 9850 are partially hand-picked to so they can easily support 2.0NB on safe voltage.

I doubt "AMD" is "Hiding" performance, but doing the best they can to produce the best chip.
If it was simply a matter of setting a higher default NB, I have no doubt they would do so if it was that simple.

 

[jimmysmitty]

Likely Very Little.
The Increase from 1.8 to 2.0 (Just over a 10% boost) yielded about a 1% increase in perfomrance.
http://xbitlabs.com/articles/cpu/display/phenom-x4-9850_4.html#sect0

The reason the IMC is not higher is likely due to Chip limitations.
I recall reading about a numberof ES Samples having the IMC burn up when folks upped the voltage enough to try and hit 2.2 or 2.4. (2.0 Was the Default on the ES Samples.)

Most likely the 9850 are partially hand-picked to so they can easily support 2.0NB on safe voltage.

I doubt "AMD" is "Hiding" performance, but doing the best they can to produce the best chip.
If it was simply a matter of setting a higher default NB, I have no doubt they would do so if it was that simple.

Well this does make sense if you look at what the NB spped does. Its mainly the speed of the memory controler. So while it will increase the memory bandwidth and performance easily for synthetic benchmarks, the real worl wont see the same boost in performance.

Mor than likely due to the fact that there are very few memory bandwidth sensative programs out there. Though this would benefit the server, it wont benefit gaming and such in the same way.

End users like ourselves who mainly game and do video/audio would benefit if there was a way to increase the speed between the PCIe and CPU/memory. PCIe2.0 increaed the bandwidth of the interface but the speed of the link between the CPU has not increased. Once that does PCIe2.0might perform better than PCIe 1.1

Either way the Phenoms show that they are not that bad of a chip. But still the performance on a clock per clock basis is not there yet and I am still afraid not all will OC the same as the ES samples. I just hope the retail samples don't ship with a 1.8GHz NB again. Twice in a row would just be wrong.

*edit*
The X-bit labs only got theirs to 2.7GHz stable with a voltage increase using only the multiplier. Anand got theirs to 2.8GHz same without a voltage increase and techreport got to 3.1GHz stable with a voltage of 1.56v. So far its turning out that the B3's don't OC much better and all have different ranges they can and will go to. This almost points that the OC "disability" of the Phenoms may be limited by the architecture and not the stepping. When we have more we might know.

 

[albundy2]

3.1ghz wow i didn't think that would even be doable in a wet dream.

 

[Reynod]

http://www.anandtech.com/cpuchipsets/showdoc.aspx?i=3272&p=1

http://www.legitreviews.com/article/682/1/

http://www.pcper.com/article.php?aid=539&type=expert&pid=12

http://www.hothardware.com/Articles/AMD_Phenom_X4_9850_B3_Revision/

http://www.techreport.com/articles.x/14424/1

http://www.xbitlabs.com/articles/cpu/display/phenom-x4-9850.html

http://www.hothardware.com/Articles/AMD_Phenom_X4_9850_B3_Revision/

The legit article was interesting:

When it comes to benchmarks, the move from 2.50GHz to 2.97GHz increased performance by 25%, which is great considering the overclock was only 18%.

So it seems to scale very well.

zenmaster thanks for the link to Xbit where they upped the IMC speed and noted not much difference.

The latency on the L3 is still crap though.

Bolting on a heap of L3 doesn't seem to do little when compared to a much faster (and larger) L2 idea Intel is using.

Still ... overall the results are very positive for AMD ... moving in the right direction.

 

[gpippas]

Increasing the NB speed also increases the frequency of the L3 cache if I remember correctly (might be wrong). Thats where the increased performance comes from. It should also decrease the latency. Like jimmysmitty said the increased memory bandwidth won't show instant results.

zenmaster

Xbit did show the performance increase was only about 1% for a 10% overclock. However we don't know how well the NB really scales with increased frequency. So for a 30% increase in NB speed to 2.5ghz it is possible that you could see a 5% performance increase. Which when you consider a 9850BE is approx 15% slower than a Q6600 at stock speeds that is a good chunk of performance difference to be made up.

Like you said though obviously AMD can't get it faster otherwise they would.

 

[Mathos]

Increasing the NB speed also increases the frequency of the L3 cache if I remember correctly (might be wrong). Thats where the increased performance comes from. It should also decrease the latency. Like jimmysmitty said the increased memory bandwidth won't show instant results.

zenmaster

Xbit did show the performance increase was only about 1% for a 10% overclock. However we don't know how well the NB really scales with increased frequency. So for a 30% increase in NB speed to 2.5ghz it is possible that you could see a 5% performance increase. Which when you consider a 9850BE is approx 15% slower than a Q6600 at stock speeds that is a good chunk of performance difference to be made up.

Like you said though obviously AMD can't get it faster otherwise they would.

NB/IMC performance scaling depends on core speed too. Sad part is either way, IMC performance doesn't start to increase a lot until the cores are above 2.6. And with Phenom still not clocking well past that you see the problem? Damn thing doesn't scale good till you get past 2.6ghz and can't currently get past it with the current design.

 

[epsilon84]

The legit article was interesting:

When it comes to benchmarks, the move from 2.50GHz to 2.97GHz increased performance by 25%, which is great considering the overclock was only 18%.

So it seems to scale very well.

That's odd, theoretically speaking you can't achieve better than 1:1 scaling ratio with increased clcokspeeds, I wonder if LR increased the NB/L3 as well whilst overclocking? That could be a possible explanation for the 'superlinear' scaling.

If you go by Tech Reports's overclocking benchmarks, between 2.5GHz and 3GHz Cinebench performance increases 18% from a 20% clock increase.

Mathos, I'm not sure what you mean by the 2.6GHz number? Its not like there is some magical switch beyond 2599.99MHz that suddenly allows Phenom to achieve superlinear scaling...

 

[computertech82]

Nice finds there. I was hoping a bit better performance (like actually BEATING the X2 6400+ once and for all....oh well). At least it's in the right direction, if the quoted prices are correct, they may not be too bad of a deal to get (providing you have a much slower cpu already or it's your first system). If the black edition can actually make it 3ghz (beating the x2 6400+), at least it would be a upgrade for the older boards; providing the bios gets updated for it.
By what I see though, seems the OLD Q6600 still has a slight lead over the new cpus at stock, and if overclocked, the q6600 really takes the lead over the new amd cpus. Maybe the 45nm cpus will be better?

 

[Reynod]

Yes I agree ... that bit seemed odd.

Once it is cranked above 2.8 Ghz the chip does seem a lot better across the board though.

Power dissipation becomes an issue when overclocked.

No more than an overclocked Q6600 I suppose ... Both monstrous.

I can't help but think how much better it could be if the IMC ran at the core speed (or closer to it) and it had a bigger L2 cache instead of the L3.

Still ... things seem to be looking up.

 

[Mathos]

That's odd, theoretically speaking you can't achieve better than 1:1 scaling ratio with increased clcokspeeds, I wonder if LR increased the NB/L3 as well whilst overclocking? That could be a possible explanation for the 'superlinear' scaling.

If you go by Tech Reports's overclocking benchmarks, between 2.5GHz and 3GHz Cinebench performance increases 18% from a 20% clock increase.

Mathos, I'm not sure what you mean by the 2.6GHz number? Its not like there is some magical switch beyond 2599.99MHz that suddenly allows Phenom to achieve superlinear scaling...

It has a wierd scaling curve, starts to dip down at 2.6, then after that it shoots back up again to the original lines linear rise. But for example, my 2.6core 2.4nb/imc is equal to say about a 2.8ghz Phenom, going by benchies. But I say thats mostly due to the higher L3 cache speed.

 

[EXT64]

That's interesting. If it does yield such a big increase, I wonder why they aren't upping the HT speed to full. Are they saving that for the 45nm so they look really good?

 

[homerdog]

That's interesting. If it does yield such a big increase, I wonder why they aren't upping the HT speed to full. Are they saving that for the 45nm so they look really good?

I somehow doubt that AMD is holding back right now.

 

[Mathos]

I somehow doubt that AMD is holding back right now.

Think you're right, its very rare for the IMC to be able to run stable above 2.0/2.2 on the phenoms when it comes down to it. Though may be how they come back on the deneb 45nm though.

 

[EXT64]

Oh no, I agree it isn't as simple as upping the frequency on the current hardware. I meant designing it to run at the higher HT link as many current mothers claim to support.

 

[MU_Engineer]

^^This is what I've been saying all along. The 6400 x2 performs as it does because the IMC is runing at 3.2ghz. If you all recall, socket AM2 could not perform as well as 939 until the IMC was at 2.4ghz. Thus, the X3 & X4 chips are basically running into the same issue Intel is with the FSB. Sure, they have far more bandwidth to the chipset, but the bandwidth to the memory system is VASTLY hindered compared to what it could be. Honestly, I was quite upset to find out that AMD had decided to use two 64bit memory controllers instead of two 128 bit memory controllers. Sure, they can read and write at the same time, but the overall bandwidth is at best the same as an x2 of the same IMC frequency.

There are a few possible reasons why AMD didn't put two 128-bit memory controllers into the Phenom. Two 128-bit memory controllers would require four channels using available 64-bit DDR2 or DDR3 modules. A current Socket AM2 or Socket 1207 chip only has 480 pins for two channels and using four would require doubling the RAM control pin count, requiring a new socket. AMD has promised backwards compatibility, particularly on Socket 1207, so this is a no-go. Requiring four modules for peak performance would also get a bit cumbersome while just transitioning to DDR3 also gives a big increase in bandwidth, so that probably puts dampers on the dual 128-bit IMC plan as well.

 

[Reynod]

Thanks MU ...

Is memory bandwidth the real issue at present thou? ... isn't it the cache latency on that L3 pig of a thing MU ... and the prefetch logic ... plus the core2 is essentially a wider design?

Look at the Scott Wasson original Barcy article for a good brief on that whole latency issue.

Tends to explain the major reason the single socket performance can't catch Intel.

On the multisocket front the thing scales quite well.

An ugly duckling ... some may say but I think of it more like comparing a Train engine to a prime mover.

Intel's prime mover can tow one or two trailers at speed but runs out of steam on a steep hill and has to drop back a few cogs (but picks up down the other side quickly). It is a lot faster and more maneuverable.


AMD's train engine can't go as fast but can pull a bigger load because it can easily insert more engines in a line ... the couplings between them allow one engine to control the rest easily ... more engines more load.

VIA's go cart has a whipper snipper motor.

Intel's new atom is a glow plug engine from my model aeroplane.

I see some of the other threads used car analogies ... not going there again with the whole Jap vs US Muscle cars ... I like Detroit Iron !!

 

[lsdouglas]

I'm interested if my board would support "Deneb", i currently have an Asus M3A32-MVP Deluxe motherboard running a Phenom 9600 (B2).