-
We’re currently investigating an issue related to the forum theme and styling that is impacting page layout and visual formatting. The problem has been identified, and we are actively working on a resolution. There is no impact to user data or functionality, this is strictly a front-end display issue. We’ll post an update once the fix has been deployed. Thanks for your patience while we get this sorted.
AT / Gary Preview: 9850BE and new AOD
- Thread starter heyheybooboo
- Start date
What's the current technical theory on why Phenom overclocks are better with the new chipset? I'm rather ignorant of the cause of the existing limitations in overclocking Phenom's on existing mobos.
From experience on Intel systems I am keenly aware just how much the chipset can make the difference when the weakest link in the overclock is the chipset (versus the CPU). My unlocked QX6700 for example would hit 3.7GHz on my striker extreme (680i chipset) but only by increasing the CPU multiplier and Vcc. If I tried to do much of any overclocking by way of increasing the FSB the sucker was inherently unstable and would crash in a matter of hours to days.
But the same CPU on a newer X38 chipset mobo (asus P5E WS Pro) could hit 4GHz (higher net overclock) and on a 400MHz FSB (versus 266MHz on the 680i). I was astonished just how piss-poor the 680i chipset/mobo actually was in hindsight.
But if the AMD chipset has been hindering Phenom clocks in similiar fashion (is it the HT freq that must be overclocked as the analogue to FSB OC'ing on Intel?) then I don't understand why the BE's would be impacted in which the CPU multiplier is simply being increased (as was the case with my QX6700).
From experience on Intel systems I am keenly aware just how much the chipset can make the difference when the weakest link in the overclock is the chipset (versus the CPU). My unlocked QX6700 for example would hit 3.7GHz on my striker extreme (680i chipset) but only by increasing the CPU multiplier and Vcc. If I tried to do much of any overclocking by way of increasing the FSB the sucker was inherently unstable and would crash in a matter of hours to days.
But the same CPU on a newer X38 chipset mobo (asus P5E WS Pro) could hit 4GHz (higher net overclock) and on a 400MHz FSB (versus 266MHz on the 680i). I was astonished just how piss-poor the 680i chipset/mobo actually was in hindsight.
But if the AMD chipset has been hindering Phenom clocks in similiar fashion (is it the HT freq that must be overclocked as the analogue to FSB OC'ing on Intel?) then I don't understand why the BE's would be impacted in which the CPU multiplier is simply being increased (as was the case with my QX6700).
SlowSpyder
Lifer
I belive this is just an update to the southbridge, still a 790FX northbridge. I'm not really sure why the southbridge change helps with the overclocks, but apparantly it does.
the SB750 is also called the magic south bridge.
i have no idea as well how the SB helps.
Im guessing on AMD he must play a more vital role, however i always thought the NB was the most important.
And yes ALL NVIDIA chipsets for intel are a POS.
However this yorkie is really loving the 780i i have.
i have no idea as well how the SB helps.
Im guessing on AMD he must play a more vital role, however i always thought the NB was the most important.
And yes ALL NVIDIA chipsets for intel are a POS.
However this yorkie is really loving the 780i i have.
Yeah and there are two ways to look at this...you could say the new SB helps or you could say the old SB's hindered.
So what could possibly be the issue with the old SB's that they hindered Phenom's clockspeeds?
And here we are nearly a year after the first introductions of Barcelona and AMD is just now getting around to resolving the clockspeed problem imposed by these older SB's?
It seems wacky to me no matter how I look at it.
The old southbridge hindered. However TDP goes up scary fast when over 3.2 ghz. AMD really needs a refresh and in a hurry.
They are also in a tough spot as most of these phenoms are going past 3.0 ghz with the new board, however they can't certify a 3.0 ghz chip as the socket hasn't changed and risk consumers returning multiple products because of trying to use the old chips.
They are also in a tough spot as most of these phenoms are going past 3.0 ghz with the new board, however they can't certify a 3.0 ghz chip as the socket hasn't changed and risk consumers returning multiple products because of trying to use the old chips.
If you consider my posts to be of the anti-AMD nature then you, dear recently joined poster, are in for quite a shock at what you will be reading from other posters in threads around here. :laugh:
Getting back to the contents of my post...can you answer any of the questions I posed or shed light on why the questions themselves are misguided?
Believe it or not I tend to post questions out of legitimate curiosity aimed at extinguishing identified ignorance on my part...the majority of my posted questions on these forums are decidedly not flame bait. Feel free to openly disagree but I hope my posting record speaks to what I say here.
heyheybooboo
Diamond Member
I've got an MSI 790fx with sb600. I got a Phenom 9600BE from the Egg last week and it clocked on up to 2.7GHz using the multiplier like a champ. I could boot at 2.8GHz but couldn't run Prime. I'm not ready to jack up the voltage quite yet - 2.7 will run Premiere just fine - lol.
This is pretty much above my paygrade but here goes ...
The AMD base clock is 200MHz. With AM2+ you have the CPU, HT link and north bridge all working off the base clock. Even the southbridge 'link' works off the base 200MHz clock.
The 'northbridge' is not really a northbridge like the Intel FSB. The northbridge in the AM2+ platform is the L3 cache / integrated memory controller.
With the southbridge it gets interesting. You have PCIe, PCI, SATA, USB, etc., and the base clock speed has to be 'skewed' with a frequency divider (this is not a 'divider' to the extent it is adjustable). The 'dividers' are not really an issue with the base 200MHz clock but they do to a small extent seem to limit SB performance as compared to other platforms. The key is keeping all the clock signals in phase throughout the system.
So at the base 200MHz clock and SB clock timing 'dividers' everything stays in phase. The total system clocks are 'syncronous' in a fixed relationship to one another.
When you start to increase the base 200MHz clock those fixed relationships start to get a little goofy. You have to start bumping up voltages to keep everything in phase. More voltage - more heat - less stability. Simply increasing the base clock overclocks total system timings in every aspect - thats why using the CPU, NB and HT multipliers is preferable. With AM2+ the CPU, NB and HT link each have individual multipliers. In Gary's preview the AMD OC 'shorthand' would be 200 x 16 x 12 x 10. It's very interesting to note that he achieved greater stability (and 3.2GHz) by reducing voltage (and that the 9850BE reached 3GHz at stock voltage smiply by increasing the multiplier).
The sb750 is not 'magic'. It apparently does improve stability, performance and timing with the overall system PLL. In previous BIOSs with sb600 and sb700 it seemed to help when the link to the southbridge was locked at 200MHz (but few BIOSs had this adjustment).
In the case of AMD and overclocking I think you are seeing a fundamental shift in theory away from simply increasing the base clock timing for performance. The basis for this shift is related to AMD's new 'split' personality 😛. The NB and HT are independent (as is the SB) with their own power planes. The CPU cores are independent or 'split' and it is my understanding that the next step is for each core to have its own power plane. Increasing performance is gained not thru increasing the 200MHz base clock but by raising the core multiplier (and voltage).
In our multicore future this is a significant advancement. The system may wait in a low power state and jack up the multi (and the voltage) on an individual core (or more) as needed. Seems pretty slick. I hope they can make it work.
I also think this is why enabling C&Q is the way optimize your clocking with the Phenom. That's how I understand it, anyway. Some folks have mistakenly disabled cool and quiet because that's the way they've always OC'ed an AMD proc.
This is pretty much above my paygrade but here goes ...
The AMD base clock is 200MHz. With AM2+ you have the CPU, HT link and north bridge all working off the base clock. Even the southbridge 'link' works off the base 200MHz clock.
The 'northbridge' is not really a northbridge like the Intel FSB. The northbridge in the AM2+ platform is the L3 cache / integrated memory controller.
With the southbridge it gets interesting. You have PCIe, PCI, SATA, USB, etc., and the base clock speed has to be 'skewed' with a frequency divider (this is not a 'divider' to the extent it is adjustable). The 'dividers' are not really an issue with the base 200MHz clock but they do to a small extent seem to limit SB performance as compared to other platforms. The key is keeping all the clock signals in phase throughout the system.
So at the base 200MHz clock and SB clock timing 'dividers' everything stays in phase. The total system clocks are 'syncronous' in a fixed relationship to one another.
When you start to increase the base 200MHz clock those fixed relationships start to get a little goofy. You have to start bumping up voltages to keep everything in phase. More voltage - more heat - less stability. Simply increasing the base clock overclocks total system timings in every aspect - thats why using the CPU, NB and HT multipliers is preferable. With AM2+ the CPU, NB and HT link each have individual multipliers. In Gary's preview the AMD OC 'shorthand' would be 200 x 16 x 12 x 10. It's very interesting to note that he achieved greater stability (and 3.2GHz) by reducing voltage (and that the 9850BE reached 3GHz at stock voltage smiply by increasing the multiplier).
The sb750 is not 'magic'. It apparently does improve stability, performance and timing with the overall system PLL. In previous BIOSs with sb600 and sb700 it seemed to help when the link to the southbridge was locked at 200MHz (but few BIOSs had this adjustment).
In the case of AMD and overclocking I think you are seeing a fundamental shift in theory away from simply increasing the base clock timing for performance. The basis for this shift is related to AMD's new 'split' personality 😛. The NB and HT are independent (as is the SB) with their own power planes. The CPU cores are independent or 'split' and it is my understanding that the next step is for each core to have its own power plane. Increasing performance is gained not thru increasing the 200MHz base clock but by raising the core multiplier (and voltage).
In our multicore future this is a significant advancement. The system may wait in a low power state and jack up the multi (and the voltage) on an individual core (or more) as needed. Seems pretty slick. I hope they can make it work.
I also think this is why enabling C&Q is the way optimize your clocking with the Phenom. That's how I understand it, anyway. Some folks have mistakenly disabled cool and quiet because that's the way they've always OC'ed an AMD proc.
...
I just need a stupid X58 board to have neha up and running.
You think im gonna be playing with an AMD?
:X
Wanna name this one? I was thinkn Venus. or Aphrodite, since this will most likely be the most jealous build on AT at that time. 😛
looks like someone beat me to this post. Yes, I think AMD R&D is cooking up something again. well for lack of any news in the AMD CPU department I guess it's good to know they are working on something or rather. even some voodoo OC magic is nice.
Recently heard some rumors they are hard at work on the next core, who knows maybe by Xmas, they'll have another sleeper hit like the 4850/4870s. I'd love that and cut the quad price by half! $150 Q9xxx for Xmas anyone?
Recently heard some rumors they are hard at work on the next core, who knows maybe by Xmas, they'll have another sleeper hit like the 4850/4870s. I'd love that and cut the quad price by half! $150 Q9xxx for Xmas anyone?
Explain to me what is so exciting about it and I will "shoo."
I will be excited when AMD puts out something that makes Intel say "Uh-oh", or "oh-sh*t", like they just did to NV. Id love to see Intel have to drop Nehalem prices due to a competitive market.
As far as a 65nm quad-core than can maybe hit 3.2ghz on the right board....that rates right up there with a dentist visit on the excitement scale.
SlowSpyder
Lifer
So you came here just to bash something positive happening for the Phenom platform? The Phenom is not the same architecture as the Intel quads, they don't clock the same in general.
Awesome info heyheybooboo, thanks for taking the time to post and educate some us on the background and details.
So with AM2+ it could definitely be the case that the signal is less noisy above 200MHz and thus provides ever greater margin that could be expended on overclocking.
It is an interesting note as you said that lowering the voltage improved the peak stable overclock...definitely hints at thermal limitations.
I really wish I had the time right now to dabble with a Phenom system on my vapoLS rig. Hopefully Deneb is out when I get some freetime to get back to my hobby again.
Is Deneb slated for having this "per core voltage regulation" technique? AMD needs some fancy logic techniques up their sleeve to get 45nm w/o HK+MG into a reasonable clockspeed/TDP situation.
Cookie Monster
Diamond Member
How about nameing it "LEONIDAS"? 😀
Heres a Foxcon MB preview
http://www.hardwarezone.com/ar...view.php?cid=6&id=2618
and also a German site did a review of this MB
http://translate.google.com/tr...owthread.php?p=3457487
http://www.hardwarezone.com/ar...view.php?cid=6&id=2618
and also a German site did a review of this MB
http://translate.google.com/tr...owthread.php?p=3457487
magreen
Golden Member
The real reason is until now the chips were on strike against Ruiz being CEO. Now that he stepped down they're operating up to potential! 😛
... I'm really only kidding. I'm also reasonably excited about this latest development for AMD and am really interested in seeing the full review tomorrow. I'm not saying it'll put em in the ballpark with a penryn quad (let alone Neha) but it's looking a little brighter for them.
As far as I'm concerned, this is the first piece of PR put out by AMD's cpu division in the last 2 years that didn't turn out to be a complete and utter lie. So I'm happy about it, if that makes sense.
heyheybooboo
Diamond Member
I need ""Mr. Wizard"" to help me with this one but I think the technical term is 'jitter'. Maybe an 8th grade science class could explain it to us 😛
Think about the old black and white sci-fi movies. Every lab had an oscilloscope. As long as the starting and endpoints of those waves occur at the same points everything is in 'phase' regardless of frequency. Jitter is the term used to describe the variance of the starting/endpoints of those frequency 'waves' when they lose their fixed relationship.
As I understand OCing a Phenom the phase gets knocked around and the resulting jitter leads to instability. It's a 'catch-22' because to reduce jitter you can up the volts but that also lends to greater instability.
That's the key behind the sb750. The 'phase' is more stable at higher frequencies without adding additional voltages to the system. You can gain higher frequencies by simply adjusting multipliers (not the 200MHz base clock) and the phase is not adversely impacted.
I think the 'option' for HkMG exists in a second spin at 45nm - or at 32nm. I don't know if 'leakage' is an issue with the current level of transistors in Phenom (450 million?). The mobile Puma/Griffin platform has split 'cpu core' power planes and it rolled out in retail last week (with the same IGP as the 780g chipset as I understand it). I'm sure Viditor has the inside info on all that and as to whether the 45nm will introduce it to the desktop.
Dadofamunky
Platinum Member
Jupiter Optimus Maximus.
I'm on a Roman kick right now.
TRENDING THREADS
-
Discussion Zen 5 Speculation (EPYC Turin and Strix Point/Granite Ridge - Ryzen 9000)- Started by DisEnchantment
- Replies: 25K
-
Discussion Intel Meteor, Arrow, Lunar & Panther Lakes + WCL Discussion Threads
- Started by Tigerick
- Replies: 25K
-
Discussion Intel current and future Lakes & Rapids thread- Started by TheF34RChannel
- Replies: 24K
-
-
