Phenom B3

[pcgamer321]

So I really want to try the new Spider Platform, especially the 790fx chipset, but I really don't like the current Phenoms. And I do not need to buy a computer now, as my current one is top notch, I was just going to use it as backup parts though because I'm paranoid or give it to my brother so don't tell me to get an Intel now.(Already have C2D btw)

Will the B3 stepping make any difference?

 

[Martimus]

No-one seems to know. I know that this doesn't help you, but I think we are just going to have to wait until they are reviewed before we know how well they perform.

 

[MarcVenice]

Well, the disabling of the cache to increase stability is said to decrease performance by as much as 10%. With it being 10% faster it could actually come close to Intels clock for clock speed, and if they made some more enhancements it would be even better. But B3's have allready been shipped out, engineering samples, if I'm not mistaken, and we haven't really heard a thing about it, so I bet performance is still lacking compared to Intel.

 

[hennethannun]

That's hard to say until it's out. AMD SAYS that the B3 will fix the TLB errata without degrading performance like the current software workaround does.

If it DOES work the way it's supposed to, then I can't imaging AMD would sit on the news. they desperately need phenom to compete, so any good news should make it's way to the public asap. Conversely, if they don't see the desired results then they will probably just say as little as possible. So the more we hear about the B3 stepping and the earlier we hear it, the more hopeful I will be...

 

[v8envy]

The general public is buying the Gateways and HPs with the current CPUs. No need to remind them and make them worry re: whether or not the machine they see at Best Buy has processor bugs or not.

So if the OEMs are happy, there's no need for fanfare. Even with the TLB cache fix would YOU buy a Phenom today? They know the enthusiast market has written the Phenom off, at least until they hit 3 ghz and/or start overclocking well. With or without bugs. The product is positioned for the mainstream/value segments. Those buyers don't read processor news. And the OEMs who have inventory would probably like as few reminders of the bug as possible. So I doubt you'll see trumpeting and fanfare once the bug is squashed -- the 9550 and 9650 should make a rather quiet appearance on store shelves.

I'm a bit surprised there are no 'try-cores' out yet, though.

 

[Martimus]

Rumor about B3 stepping Barcelona being tested out.

It isn't much, but I thought you might be interested in it considering the query.

 

[v8envy]

Rumor squashed in comments. The isoHUNT guy denied they're B3 stepping Barcelonas -- just regular old TLB-affected CPUs.

 

[taltamir]

Originally posted by: MarcVenice
Well, the disabling of the cache to increase stability is said to decrease performance by as much as 10%. With it being 10% faster it could actually come close to Intels clock for clock speed, and if they made some more enhancements it would be even better. But B3's have allready been shipped out, engineering samples, if I'm not mistaken, and we haven't really heard a thing about it, so I bet performance is still lacking compared to Intel.

I thought it was a minimum of 10%... 20% on average, with some very specific application seeing as high as 50%

But regardless of those scores, the phenom can NOT compete regardless of the bug. So as v8envy said, no need to advertise the fact that there is a bug, as most people simply do not know.

 

[hennethannun]

I think Phenom can compete without the bug. It obviously doesn't have enough speed right now to challenge the C2Es, and by the time it does nehalem will be around the door. So as v8envy said they have lost the top of he enthusiast market for another year or so.

But without the TLB phenom is more or less competitive in its price range, especially at stock speeds. and AMD has successfully undercut Intel's price for quad-core processors.

and v8envy, you make a good point about not wanting to alert the masses about the TLB, but that same argument works against you. AMD already announced the TLB errata, and most people who don't pay attention don't know anything about it. Announcing that they have fixed it can't have a worse PR effect that announcing it existed with the mass market, and it will be a big boost to all those people who want to buy phenom but won't drop money on a bugged product (for example: ME! I was seriously thinking of jumping up to phenom but the TLB errata pushed me over to wolfdale). If AMD KNOWS that B3 will fix the problem with no drawbacks, then they would be stupid to sit on that info. ergo they don't know that it will fix the problem without a performance hit. It's just a matter of whether or not AMD doesn't knows it WON'T fix the problem or if they just haven't finished engineering the new stepping and don't really know what the results will be.

 

[pcgamer321]

alright thanks guys. Seeing as there is no info I'll just wait and see before seeing what I will do.

 

[Idontcare]

The bug does not effect performance, it causes system lockups and reboots.

The workaround to the bug effects performance. Disable the workaround on a B2 and you have the same performance as a B3.

You can disable the bug workaround either by BIOS or by using the AMD created overclocking utility in windows.

So performance of B3 is same as B2, only question is do you experience lock-ups with B2 and if so then you need the bug-workaround enabled and only then do you experience the performance loss.

In any event, B3 without the bug or B2 without the bug-workaround enabled, the performance clock-for-clock still doesn't compete with Intel quads. Even when overclocked the performance doesn't compete with the lowly Q6600 at stock.

http://www.tomshardware.com/20...om_9600_black_edition/

In the end, we can't really recommend AMD's Phenom 9600 Black Edition, since the CPU draws more power than Intel's Core 2 Quad Q6600 while costing the same, offering less overclocking headroom, and thus providing lower performance overall.

The bottom line is that the Phenom 9600 Black Edition offers bad value for money. Based on our benchmark results and the direct comparison with Intel's Q6600, AMD's CPU should cost no more than $220, in our opinion.

 

[Cookie Monster]

There is still some hope. The phenoms also have this problem with the L3/NB speed stuck at 2.0GHz or 1.8GHz. That is why the current reviews show an alarmingly high latency (50~55ns?) even with the IMC of the phenoms for the L3 cache. If this can be fixed (only for AM2+ mobos) then you can get some performance out of it since AMD was intending to clock it much higher than the actual CPU frequency.

But its a big mess (mostly thnx to the TLB bug), and im not sure how AMD is going to clean it up with current roadmaps suggesting cancellation of 9900/9700 phenoms with the arrival of 9x50 phenoms in Q3.

 

[heyheybooboo]

^^^^^

What Cookie said ...

AFAIK cache speed should increase proportionately with clock speed. Being stuck at 1.8GHz seems a bigger bug to me than the TLB (which was in virtualization, right?)

 

[nerp]

Hey, last summer John McCain was considered dead and he was taking out loans because he was so broke. Look at what the passage of some time can do. . .

 

[soccerballtux]

Originally posted by: nerp
Hey, last summer John McCain was considered dead and he was taking out loans because he was so broke. Look at what the passage of some time can do. . .

He's just about dead anyhow being what, 72? or is it 62?

 

[taltamir]

Originally posted by: hennethannun
I think Phenom can compete without the bug. It obviously doesn't have enough speed right now to challenge the C2Es, and by the time it does nehalem will be around the door. So as v8envy said they have lost the top of he enthusiast market for another year or so.

But without the TLB phenom is more or less competitive in its price range, especially at stock speeds. and AMD has successfully undercut Intel's price for quad-core processors.

and v8envy, you make a good point about not wanting to alert the masses about the TLB, but that same argument works against you. AMD already announced the TLB errata, and most people who don't pay attention don't know anything about it. Announcing that they have fixed it can't have a worse PR effect that announcing it existed with the mass market, and it will be a big boost to all those people who want to buy phenom but won't drop money on a bugged product (for example: ME! I was seriously thinking of jumping up to phenom but the TLB errata pushed me over to wolfdale). If AMD KNOWS that B3 will fix the problem with no drawbacks, then they would be stupid to sit on that info. ergo they don't know that it will fix the problem without a performance hit. It's just a matter of whether or not AMD doesn't knows it WON'T fix the problem or if they just haven't finished engineering the new stepping and don't really know what the results will be.

Thats a good point. I almost bought a phenom for my AM2 board, but the TLB deterred me and I finally ended up buying a wofdale and a new motherboard for it.
and people who havent heard of it yet are probably not gonna hear it is fixed. It is not like this is gonna be announced on CNN or FOX or something...

Originally posted by: Cookie Monster
There is still some hope. The phenoms also have this problem with the L3/NB speed stuck at 2.0GHz or 1.8GHz. That is why the current reviews show an alarmingly high latency (50~55ns?) even with the IMC of the phenoms for the L3 cache. If this can be fixed (only for AM2+ mobos) then you can get some performance out of it since AMD was intending to clock it much higher than the actual CPU frequency.

But its a big mess (mostly thnx to the TLB bug), and im not sure how AMD is going to clean it up with current roadmaps suggesting cancellation of 9900/9700 phenoms with the arrival of 9x50 phenoms in Q3.

And that would be a reason to not be trumping it up so much... it might fix the TLB but nothe issues preventing them from getting higher clockspeeds.

 

[Nessism]

So what is the real world likelihood of experiencing the TLB bug anyway? I thought it only showed up under very specific configurations that are not very common.

 

[hennethannun]

Yup, by all accounts the TLB errata only shows up in a very few circumstances such as (but not exclusively) virtualization(This and this are slightly more descriptive). So it's stuff that doesn't happen very often at all for us non-server types. but that's not really the point. Phenom is a late addition to the cpu scene. It was delayed several times, launched at speeds 30% lower than were initially planned and can only barely match the clock for clock performance of it's main competitor. Given all that information, PLUS the fact that is has a serious, universal design flaw (even one that doesn't come up too often), why bother? there is already a mature, non-glitched alternative product available. That is what makes this problem such a disaster for AMD. when you show up at the party late AND sloppily dressed then you can't really expect anyone to pay much attention.

idontcare: That tomshardware article doesn't have particularly helpful data (or at least, the data is presented in a rather convoluted way), but if you look closely you will see that phenom does, more or less, compete with the Q6600 clock for the clock (obviously it would depend on what you want to do, but over all phenom is within 5% of kentsfield clock for clock). the problem is that kentsfield is now Intel's second line product AND the C2Qs have way more overhead. most Q6600s can pull 3ghz with a minimum of difficulty, while the 9600 black struggles to reach 2.9ghz. And on top of that phenom uses more power and runs hotter too.

Also, I don't think we can really comment on the performance of the B3s yet, since we haven't seen any of them. It's possible that they will fix the TLB without affecting performance, but it's equally possible that there is a serious design flaw that somehow slipped through the early stages of development. In which case it could be possible (though by no means probable) that there is no easy way to fix the problem without doing something similar to the current, performance killing, workaround on a hardware level. You're right that the B3s *should* offer un-patched B2 performance (possibly even a bit better) clock for clock without any TLB related issues, but we just don't know right now.

 

[exar333]

Nope, still will be crap.

 

[Idontcare]

Originally posted by: Nessism
So what is the real world likelihood of experiencing the TLB bug anyway? I thought it only showed up under very specific configurations that are not very common.

Likely never if you are typical desktop user who does not do virtualization.

Virtualization is likely to be done by programmers and in the server environment.

The reason you can't find performance competitive Phenom's has nothing to do with TLB bug, it has to do with TDP and speedbins.

Both of which are exhibiting major suckage at the moment compared to Intel's 65nm quad lineup, let alone their 45nm one.

The tomshardware review (linked above) kind of highlights this in a nice and easy to read manner. But every review out there on the phenom in one way or another touches on the power-consumption aspects versus the limited overclocking overhead.

 

[Nessism]

I know the Phenom's not the best but I still think it's a good bang for the buck - got mine with motherboard for $220 - recently reduced to $200 at Fry's. Intel Quad is still $270. Overclocked dual core may be an alternative but it's kind of silly to criticize the Phenom for a rare bug when the alternative is an overclocked chip.

 

[Nemesis 1]

Originally posted by: v8envy
The general public is buying the Gateways and HPs with the current CPUs. No need to remind them and make them worry re: whether or not the machine they see at Best Buy has processor bugs or not.

So if the OEMs are happy, there's no need for fanfare. Even with the TLB cache fix would YOU buy a Phenom today? They know the enthusiast market has written the Phenom off, at least until they hit 3 ghz and/or start overclocking well. With or without bugs. The product is positioned for the mainstream/value segments. Those buyers don't read processor news. And the OEMs who have inventory would probably like as few reminders of the bug as possible. So I doubt you'll see trumpeting and fanfare once the bug is squashed -- the 9550 and 9650 should make a rather quiet appearance on store shelves.

I'm a bit surprised there are no 'try-cores' out yet, though.

Normally I just read these . But I can't believe I just heard what I heard. I am going put in the Star and tribune the facts. Its a serious problem and with those kinds of performance hits with bios fix . Public needs to know,

 

[superstition]

Originally posted by: Nessism
I know the Phenom's not the best but I still think it's a good bang for the buck - got mine with motherboard for $220 - recently reduced to $200 at Fry's. Intel Quad is still $270. Overclocked dual core may be an alternative but it's kind of silly to criticize the Phenom for a rare bug when the alternative is an overclocked chip.

I don't think this is good logic, unless you're running a system that relies on four cores for best performance. What are you doing with your system primarily?

Overclocking is a non sequitur if your motherboard supports your FSB speed easily and you don't have to raise voltages. It's no different from buying a product at stock, frankly, as long as you have decent cooling. If a person can buy an Intel dual core and easily get better performance for less money, then the Phenom becomes irrelevant.

 

[Nessism]

Originally posted by: superstition

Originally posted by: Nessism
I know the Phenom's not the best but I still think it's a good bang for the buck - got mine with motherboard for $220 - recently reduced to $200 at Fry's. Intel Quad is still $270. Overclocked dual core may be an alternative but it's kind of silly to criticize the Phenom for a rare bug when the alternative is an overclocked chip.

I don't think this is good logic, unless you're running a system that relies on four cores for best performance. What are you doing with your system primarily?

Overclocking is a non sequitur if your motherboard supports your FSB speed easily and you don't have to raise voltages. It's no different from buying a product at stock, frankly, as long as you have decent cooling. If a person can buy an Intel dual core and easily get better performance for less money, then the Phenom becomes irrelevant.

Needless to say, there is no reason to go multiple core unless the applications you are running take advantage of them. In my case, the Phenom box is mostly used for video editing and general office usage - the only games I play are older titles with my grade school son. For this usage, I?m extremely happy with the quad-core processor; it absolutely rips through the encoding tasks even though the cores themselves are running relatively slowly.

It seems somewhat ironic to me that many people bashing Phenom for an unlikely to ever occur errata issue, while these same people are overclocking the crap out of their Intel rigs thus opening themselves up to instability issues.

So the real question to me is how does the value stack up, comparing Phenom to Intel. Comparisons of Phenom 9500 to quad core Intel are irreverent right now due to the cost disparity ? about $80 difference. So that leaves Phenom 9500 against E8400 or E6750. If one is not overclocking, the Phenom is a good choice for users like myself where the extra cores come into play. If you are overclocking, Intel seems like an equally good choice (or even better), hardly a slam dunk though. One caveat: if you live near a Fry?s store where they sell the 9500 w/motherboard combo for $200, the value is virtually unbeatable right now.

BTW: my other rig has an overclocked E2200 with Abit IP35-E motherboard. Love this system but it cost me almost as much as the Phenom rig. I?m not a fanboy either way just a practically minded hobbyist.

 

[batmang]

Originally posted by: hennethannun
Yup, by all accounts the TLB errata only shows up in a very few circumstances such as (but not exclusively) virtualization(This and this are slightly more descriptive). So it's stuff that doesn't happen very often at all for us non-server types. but that's not really the point. Phenom is a late addition to the cpu scene. It was delayed several times, launched at speeds 30% lower than was initially planned and can only barely match the clock for clock performance of it's main competitor. Given all that information, PLUS the fact that is has a serious, universal design flaw (even one that doesn't come up too often), why bother? there is already a mature, non-glitched alternative product available. That is what makes this problem such a disaster for AMD. when you show up at the party late AND sloppily dressed then you can't really expect anyone to pay much attention.

idontcare: That tomshardware article doesn't have particularly helpful data (or at least, the data is presented in a rather convoluted way), but if you look closely you will see that phenom does, more or less, compete with the Q6600 clock for the clock (obviously it would depend on what you want to do, but over all phenom is within 5% of kentsfield clock for clock). the problem is that kentsfield is now Intel's second line product AND the C2Qs have way more overhead. most 6600s can pull 3ghz without a minimum of difficulty, while the 9600 black struggles to reach 2.9ghz. And on top of that phenom uses more power and runs hotter too.

Also, I don't think we can really comment on the performance of the B3s yet, since we haven't seen any of them. It's possible that they will fix the TLB without affecting performance, but it's equally possible that there is a serious design flaw that somehow slipped through the early stages of development. In which case it could be possible (though by no means probable) that there is no easy way to fix the problem without doing something similar to the current, performance killing, workaround on a hardware level. You're right that the B3s *should* offer un-patched B2 performance (possibly even a bit better) clock for clock without any TLB related issues, but we just don't know right now.

Word. Well said. I still love my Spider rig though.