when sempron moves to 90nm SOI process
Moderators: NeilBlanchard, Ralf Hutter, sthayashi, Lawrence Lee
when sempron moves to 90nm SOI process
currently, the semperon consists of both the athlon xp and athlon64 line.
semperon is replacing duron in amd's lineup.
eventually semperon will move to 90nm SOI process.
according to www.theregister.co.uk/2004/07/28/amd_roadmap/ -
that means that we will have a 90nm SOI athlon xp, renamed semperon (intel called its 100mhz .13 pentium 3's "celeron").
if athlon xp @ 130nm and 2ghz is about 70 watts what will moving to 90nm SOI reduce it for? (it's my understanding that current leakages increase, but maybe offset by SOI)
i would imagine that the semperon @ 90nm AND SOI would be the silent pc's cpu of choice for both performance and heat characteristics.
semperon is replacing duron in amd's lineup.
eventually semperon will move to 90nm SOI process.
according to www.theregister.co.uk/2004/07/28/amd_roadmap/ -
that means that we will have a 90nm SOI athlon xp, renamed semperon (intel called its 100mhz .13 pentium 3's "celeron").
if athlon xp @ 130nm and 2ghz is about 70 watts what will moving to 90nm SOI reduce it for? (it's my understanding that current leakages increase, but maybe offset by SOI)
i would imagine that the semperon @ 90nm AND SOI would be the silent pc's cpu of choice for both performance and heat characteristics.
Just don't expect the same performance from Sempron compared to Athlon XP (Barton). It got half the size of L2 cache (ok, maybe not a big difference) and they just don't overclock as good (I've read some reviews somewhere...). I'd never expect a 2500+@2442MHz at stock Vcore as I get from my stock Athlon XP.
Instead of Sempron, I would rather go for a mobile Athlon XP (45 W) and underclock/undervolt it. Lower power and the ability to do some serious overclocking in the future when I might need it.
I seriously don't think that we will se such a superb CPU again for that price. I think it's the last totally unlocked CPU for sure (expensive FXs and others excluded).
Right now I see no reason whatsoever for buying a Sempron, and there's a reason why AMD have ended the low end Athlon XPs. Believe me, the reason is not for your best...
And there's no point waiting for faster Socket A Semprons, because it wont happen! Socket A is on it's way out, and even if it takes years before it's gone, they will never make a Sempron faster than 2200MHz.
Stores are beginning to run out of Athlon XP, even here in Sweden!
Instead of Sempron, I would rather go for a mobile Athlon XP (45 W) and underclock/undervolt it. Lower power and the ability to do some serious overclocking in the future when I might need it.
I seriously don't think that we will se such a superb CPU again for that price. I think it's the last totally unlocked CPU for sure (expensive FXs and others excluded).
Right now I see no reason whatsoever for buying a Sempron, and there's a reason why AMD have ended the low end Athlon XPs. Believe me, the reason is not for your best...
And there's no point waiting for faster Socket A Semprons, because it wont happen! Socket A is on it's way out, and even if it takes years before it's gone, they will never make a Sempron faster than 2200MHz.
Stores are beginning to run out of Athlon XP, even here in Sweden!
Oh and by the way, du you see something wrong here?
AMD Athlon XP 2500+ Barton 11x333=1833 MHz, 512kbyte L2 cache
AMD Sempron 2600+ Thoroughbred B 11x333=1833, MHz 256kbyte L2 cache
How can I get better performance with half the L2 cache? I must have missed something here, AMD can't be that stupid...
Socket A Sempron really pisses me off!
AMD Athlon XP 2500+ Barton 11x333=1833 MHz, 512kbyte L2 cache
AMD Sempron 2600+ Thoroughbred B 11x333=1833, MHz 256kbyte L2 cache
How can I get better performance with half the L2 cache? I must have missed something here, AMD can't be that stupid...
Socket A Sempron really pisses me off!
-
- Posts: 1283
- Joined: Wed Sep 03, 2003 1:35 am
- Location: Sweden, Linkoping
The currently released version is 32 bit. The Sempron 3100+ will be based on the Athlon64 core but stipped down so it will use 32bit memory bus, but I believe it will still be a 64bit CPU. It will probably be crippled by to little cache and too slow memory bus, so performance may vary a lot in different applications. Memory hungry applications are likely to suffer a lot.Sledge:
The Sempron is a 32-bit CPU.
AMD did a real ugly marketing trick here.Mats:
Oh and by the way, du you see something wrong here?
AMD Athlon XP 2500+ Barton 11x333=1833 MHz, 512kbyte L2 cache
AMD Sempron 2600+ Thoroughbred B 11x333=1833, MHz 256kbyte L2 cache
How can I get better performance with half the L2 cache? I must have missed something here, AMD can't be that stupid...
Even though AMD does not specifically say this (for copyright reasons etc) the performance numbering is very clear.
Athlons performance numbers are comparable to P4 clock speeds.
Semprons performance numbers are comparable to Celeron clock speeds.
Geode performance numbers are comparable to Via C3 clock speeds.
So a Sempron 2600+ will be slower than a Barton 2500+.
Yes, you can. Look for other threads here in this forum. There's a huge one called something like "AMD Athlon XP-M, owners and wannabe owners" It's the best Socket A CPU you can get! Make sure you get a motherboard model that's working properly with it. There are a few who doesn't, I think.dan wrote:can you use mobile processors in desktop standard boards?
Not so. Atleast this is what AMD had to say according to Anandtech:silvervarg wrote: The currently released version is 32 bit. The Sempron 3100+ will be based on the Athlon64 core but stipped down so it will use 32bit memory bus, but I believe it will still be a 64bit CPU.
On a positive note, AMD has confirmed our suspicion that the NX bit and Cool'n'Quiet features of Athlon 64 will be available in K8 versions of the Sempron (3100+ and higher). The only A64 feature disabled in Sempron is x86-64.
So Sempron is a 32bit part on all platforms. Atleast AMD is being clear on this one. Having Sempron exist as both a 32bit and a 64bit part depending on platform would have added to confusion.
cool,
i would like to wait for amd to move to the 90nm SOI process for the athlon xp-m. i would imagine the combination of 90nm and SOI and low-k dialectic dopping would substantially reduce power consumption.
when i migrated from celeron 300 180nm to celeron 1100 130nm, i saw a 3x increase in performance and a decrease in power consumption at the same time.
i would like to wait for amd to move to the 90nm SOI process for the athlon xp-m. i would imagine the combination of 90nm and SOI and low-k dialectic dopping would substantially reduce power consumption.
when i migrated from celeron 300 180nm to celeron 1100 130nm, i saw a 3x increase in performance and a decrease in power consumption at the same time.
Just don't wait too long!dan wrote:cool,
i would like to wait for amd to move to the 90nm SOI process for the athlon xp-m.
Remember, Athlon XP under 3000+ has stopped in production , the others will follow. This means that if you want a socket A CPU better than Sempron, then you have to go for XP-M >> prices will go up.
Less amount of socket A CPUs made >> prices will go up.
Since Athlon XP is no more the best you can get, I really doubt that they will develop it further. Has it happened before that the older processor has become better after the entry of a new one? I don't think so, because AMD want us to buy the newest CPU. They don't wan't competition from an older product. That's partly why they're replacing the AthlonXP with Sempron.
The only CPU that have become better after retiring is in a way Pentium 3, it evolved into Pentium M. On the other hand, it evolved so much that it doesn't have much in common with P3 anymore. For instance, totally different sockets.
from the standpoint of silentpc computing though,
buying a 90nm SOI, all else being equal, offers higher performance at a given heat level (say 30 watts).
i am aware prescott is an exception, but it differs from northwood.
IF intel migrated northwood to 90nm strained silicon, i would imagine, from teh standpoint of silent pc computing, you would get reduced power consumption (despite current/transitor leakage) at a given speed, in comparison to its 130nm cousin.
buying a 90nm SOI, all else being equal, offers higher performance at a given heat level (say 30 watts).
i am aware prescott is an exception, but it differs from northwood.
IF intel migrated northwood to 90nm strained silicon, i would imagine, from teh standpoint of silent pc computing, you would get reduced power consumption (despite current/transitor leakage) at a given speed, in comparison to its 130nm cousin.
-
- *Lifetime Patron*
- Posts: 1809
- Joined: Sat Apr 24, 2004 1:45 am
- Location: At Home
A S754 @ 90 nm should be a great performer for the power output, provided that the process shift isn’t too traumatic.
You can run an Athlon 64 S754 @ 1 MHz @ 0.9 V for roughly 15W which is pretty damn fine.
Who knows what we’ll see with the process shrink, but the above figures are looking very good to me already.
The current Sempron S754 is based on the same core so should give at least comparable figures. Maybe a little lower power consumption, due to the smaller cache.
It’s a great performer also, as the on board memory controller means that cache size is much less important.
At the moment, the price differential between the lowest priced S754 Semprons and A64s is too marginal to make them good value. 90 nm should change all that.
If the process shrink goes well, we should hopefully see some very cheap and very low power chips. With multipliers unlocked below stock, Cool ‘N’ Quiet enabled, these babies sound the way to go. Forget socket A, this looks like the way forward.
You can run an Athlon 64 S754 @ 1 MHz @ 0.9 V for roughly 15W which is pretty damn fine.
Who knows what we’ll see with the process shrink, but the above figures are looking very good to me already.
The current Sempron S754 is based on the same core so should give at least comparable figures. Maybe a little lower power consumption, due to the smaller cache.
It’s a great performer also, as the on board memory controller means that cache size is much less important.
At the moment, the price differential between the lowest priced S754 Semprons and A64s is too marginal to make them good value. 90 nm should change all that.
If the process shrink goes well, we should hopefully see some very cheap and very low power chips. With multipliers unlocked below stock, Cool ‘N’ Quiet enabled, these babies sound the way to go. Forget socket A, this looks like the way forward.
-
- *Lifetime Patron*
- Posts: 1809
- Joined: Sat Apr 24, 2004 1:45 am
- Location: At Home
1 MHz would be rather challenging, even for Linuxdan wrote:1mhz? that reminds me of the days of my Apple IIe and the IBM PC XT.
though i suppose you could run DOS did you mean 1 Ghz?
All mobile Athlon XP-M and all Athlon 64 chips are unlocked below the stock multiplier value. Some older Athlon XP’s were also unlocked in the same way. The Athlon 64 FX series are completely unlocked.dan wrote:i know that intel locked the clock multiplier on celerons. i would like to lower my clock multiplier, not increase it.
are you saying amd lets you lower the clock multiplier?
that's awesome. i say that b/c my celeron tualatin has a 100mhz FSB. i wish i could lower the multiplier from 11 to 8.5 and increase the FSB to 133mhz.
so if you have a mobile athlon xp-m or athlon64, the best performance would be to find a target clock speed power consumption, say 20 watts @ 1.2 ghz, then lower the clock ratio to something like 5x and increase FSB to 250mhz DDR PC3700.
do you need an adapter to run xp-m in a standard socket A?
so if you have a mobile athlon xp-m or athlon64, the best performance would be to find a target clock speed power consumption, say 20 watts @ 1.2 ghz, then lower the clock ratio to something like 5x and increase FSB to 250mhz DDR PC3700.
do you need an adapter to run xp-m in a standard socket A?
-
- *Lifetime Patron*
- Posts: 1809
- Joined: Sat Apr 24, 2004 1:45 am
- Location: At Home
No adaptor is needed, but BIOS support is necessary for the XP-M to work. Most good over-clocking or under-clocking boards should support XP-Ms. Browse this forum or the forums at overclockers.com http://www.ocforums.com for advice on a suitable board.dan wrote: so if you have a mobile athlon xp-m or athlon64, the best performance would be to find a target clock speed power consumption, say 20 watts @ 1.2 ghz, then lower the clock ratio to something like 5x and increase FSB to 250mhz DDR PC3700.
do you need an adapter to run xp-m in a standard socket A?
You have the general idea for tweaking the settings, but make sure the BIOS supports the lowest multiplier that you think you’ll need.
This is also true for the A64, although BIOS compatibility isn’t an issue as you can use a desktop chip in the first place.
Aopen make a number of boards for Socket 754 A64 that support very low voltages (0.8V) and low multipliers (4) and Cool ‘n’ Quiet. They support Sempron S754 also.
They also have Silentek which automatically adjusts fan speed(s) depending on temperature. The AK86-L implementation of Silentek only supports the CPU fan, whereas the AK89 Max supports a secondary fan also. Both use passive cooling of the chipset also.
Watch out if you are going to under-clock an A64 by using a high (virtual) FSB, as you’ll potentially hit a problem in that not all A64 chipsets support an AGP/PCI lock. Without that your FSB overclock will be limited by the fact that your AGP and PCI buss speeds will be out of spec as you raise the FSB. The Nvidia Nforce 3 (all versions) and VIA K8T800 Pro do support this lock, whereas the VIA K8T800 doesn’t. It’s always best to check reviews of individual boards to see whether a manufacturer has successfully implemented a feature, as this lack of a working lock on early A64 boards was a real issue. Later BIOS versions may be needed on some boards to support the lock.
Of the two Aopen boards that I mentioned above, only the AK89 Max has the lock. Anandtech managed to overclock this board’s FSB to 347 MHz using clockgen. http://www.anandtech.com/mb/showdoc.aspx?i=2033&p=7
I think all current Nvidia and VIA Athlon XP chipsets have a working lock, although I can’t speak for SIS. Again it’s best to conform this with individual reviews.
Don’t forget that Clockgen http://www.cpuid.com/clockgen.php can be used to tweak various settings from within Windows. Check the site for compatibility, although it supports all Nvidia 2 and 3 chipsets. The A64 version is more flexible and allows you to change VCore and the Multiplier values.
If you want to use a mobile A64 then things get more complicated. There are issues with BIOS support with these chips. Also they aren’t so compelling, as the desktop chips offer all the features of the mobile parts, which isn’t the case with the Athlon XP.
AFAIK, the mobile A64s are simply hand picked A64s that are basically tested to work at lower voltage settings. You are basically paying a premium for a guaranteed low power consumption chip. A CG revision (Newcastle) desktop A64 should get close to that and with less hassle and cost.
'BIOS support is necessary for the XP-M to work'
if it will run the comparable desktop XP, it will run the XP-M, however, unless it has overclocking features (voltage, fsb, multi adjustments) then you will be stuck with 600mhz or something, unless u want to pin mod. but who would want an XP-M without under/overclocking features anyway?
if it will run the comparable desktop XP, it will run the XP-M, however, unless it has overclocking features (voltage, fsb, multi adjustments) then you will be stuck with 600mhz or something, unless u want to pin mod. but who would want an XP-M without under/overclocking features anyway?
according to theinquirer
http://www.theinquirer.net/?article=17901
blurb "The 90 nano chip is around half the size of the previous incarnation of the notebook Athlon 64, and Brookwood reckons that "bodes well" for manufacturing capacity and yields.
It gobbles 35 Watts compared to 62 Watts, which would still scorch your lap if you were senseless enough to use an AMD notebook in the nude"
for the comparison to be meaningful, it most likely, though it doesn't say expressly, is at the same mhz, although that mhz is not stated.
so going from 130nm to 90nm SOI, despite current leakage, almost halves power consumption.
while i don't have any hard numbers, i would imagine that an underclocked and undervolted athlon64 to, say, a power envelope of 20 watts, would outperform any solution from intel, such as a prescott undervolted and underclocked to 20 watts, or via at 20 watts.
http://www.theinquirer.net/?article=17901
blurb "The 90 nano chip is around half the size of the previous incarnation of the notebook Athlon 64, and Brookwood reckons that "bodes well" for manufacturing capacity and yields.
It gobbles 35 Watts compared to 62 Watts, which would still scorch your lap if you were senseless enough to use an AMD notebook in the nude"
for the comparison to be meaningful, it most likely, though it doesn't say expressly, is at the same mhz, although that mhz is not stated.
so going from 130nm to 90nm SOI, despite current leakage, almost halves power consumption.
while i don't have any hard numbers, i would imagine that an underclocked and undervolted athlon64 to, say, a power envelope of 20 watts, would outperform any solution from intel, such as a prescott undervolted and underclocked to 20 watts, or via at 20 watts.
-
- *Lifetime Patron*
- Posts: 1809
- Joined: Sat Apr 24, 2004 1:45 am
- Location: At Home
There is a 130 nm mobile Athlon 64 rated at 35 watts also (2700+, 1.6 GHz, 512KB L2 Cache). Not sure what that means for this 90 nm part? Its early days for 90 nm, so they may manage to reduce the power consumption down the line.dan wrote: so going from 130nm to 90nm SOI, despite current leakage, almost halves power consumption.
http://www.newegg.com/app/viewProductDe ... 449&depa=1
-
- *Lifetime Patron*
- Posts: 1809
- Joined: Sat Apr 24, 2004 1:45 am
- Location: At Home