The state of overclocking

[Mercutio]

Is anyone out there still overclocking? P4s or XPs? Is it all overclocking based on FSB or are there tricks to unlocking the newer chips?

At one level, I have to wonder "why overclock?", but on another, there certainly is fun to be had trying to build a stable, cool, quiet overclocked PC...

 

[honold]

i'm overclocking damn near everything

p4 1.8a -> 2.4ghz
4x256mb samsung original pc800 rdram -> pc1066
abit siluro geforce4 ti4200 64mb -> ti4600 (300/600)

100% stable for like a year now

the cpu is a 100->133 fsb overclock (none of the agp/pci stuff is affected), the memory is a change from 3x to 4x in the rdram timing, and the geforce4 is done in the display settings after enabling coolbits in the registry

 

[blakerwry]

I haven't been overclocking lately... mainly because I want as stable and cool/quiet machines as possible.

As far as I know, there are ways to unlock the Durons (model 7) and AthlonXP's (models 6 and 8) with hardware modifications... and the Epox boards seem to be able to soft unlock thoroughbred core CPU's... But for most people a simple 10%-20% FSB increase will suffice.



The popular Intel trick seems to be just running your FSB at 133(533) instead of 100mhz (effectively 400).

 

[Mercutio]

$50 P4 Celerons sure do look like they'll overlock nicely. $50 Celerons + $50 'boards might be a nice combo for a budget box. At least, they will if RAM comes down a bit (speaking of, when the hell are new Durons gonna come out? Socket A boards are dipping below $40 in some cases but I can't put a $40 1.3GHz chip in a new machine and still respect myself. ). I might be able to do a reasonably nice P4 Celeron at a $300-ish price point (see "fleet computer challenge" if you can't understand my interest in this).

 

[Buck]

Thoroughbred XP1800+ CPUs are priced very good at the moment. Combine that with a Gigabyte GA-7DXE and 256MB of RAM and you have a good system at a good price.

 

[Mercutio]

Yes, but XP1600s are even cheaper and really very nearly as fast.

... and my momentary infatuation with P4 Celerons as overclockers OR budget machines lasted until about mid-way through the article on the front of the news page.

Sigh.

 

[Buck]

AMD seems to be in a hurry to dump all CPUs previous to Thoroughbred - good for us.

 

[blakerwry]

there are t-bred 1600's too....


I have a question... im gonna look it up and do a bit of research that will never pay off.. but would be nice to know... I want to know what speed the T-bred b starts at and how it differes from the t-bred A...

I think it starts at 2600+, and is slightly larger... making heat transfer a little better, so temps should be slightly better even though it produces similar amounts of heat to the tbred A... do they offer both the t-bred A and t-bred B in 166mHz FSB?

 

[CityK]

there are t-bred 1600's too

No...T-Bred start at 1700+

I want to know what speed the T-bred b starts at and how it differes from the t-bred A...

Read the articles listed below for details...things might have changed a bit since then though

http://www.xbitlabs.com/cpu/athlonxp-2800/
http://www.xbitlabs.com/cpu/athlon-xp-2600/

 

[CougTek]

I overclock my Athlon XP 1500+ to ~1700+ ½, strickly to boost my F@h numbers. Nothing else here is overclocked. My two other local CPUs aren't very overclockable (Athlon 500MHz and Thunderbird 800MHz) and I have little to gain by overclocking my graphic card since I'm not a gamer.

Regarding your 300$ budget box, The best CPU you could get for it is, IMO, a Thoroughbred 1700+. AMD is also supposed to move to rev.B Thoroughbred even for the 1700+ and 1800+ models eventually, though I don't think you'll find anything else then rev.A Thoroughbred at less than 2400+ rating currently. Still, a rev.A Thoroughbred should be able to run at roughly 1.8GHz, which is a 2200+ rating. Not bad for a 50-55$ CPU.

What are your other requirements for these cheap boxes? CPU/heatsink, motherboard, mem, graphic card, integrate audio, HDD, CD-ROM, keyboard/mouse, LAN and enclose/PSU? Need modem, speakers, floppy (hope not)?

 

[blakerwry]

Regarding your 300$ budget box

I have a $300 budget box?


I was just asking out of curiosity... plus I have a friend who will be putting together a computer in 3 or so months and just wanted an estimate of what would be available and best when he does put his computer together.... we were looking at the Nforce2 based A7n8x Deluxe, whatever Athlon CPU is at the price break (currently 2100+), and probably 2 512MB PC2700/3200 sticks of RAM, capacity isn't much needed so hopefully the 800JB's will be using 60GB platters by then..

Of course, the hammer will be on the horizon(although he probably wont be able to afford it when it comes out) so I'm not sure what he'll decide.



Just on a side note, I can get my leadtek geforce3 Ti 200 (originally clocked at 175/400 to stabily overclock to the maximum possible via the display settings(220/500). I only ran this way for benchmark purposes because of the awesome heat production. Suprisingly, rivaTuner allows users to go higher and many have reported success at higher clock speeds.


My RAM.. or maybe the controller doesn't overclock well at all.. I can't really go above 145mhz with my pc2100 (133mHz) RAM... i don't want to up my memory voltage either so I'll just stick with stock speeds.

 

[blakerwry]

another question...now that t-breds support 333mHz operation.. what's the difference between them and the barton (barring the L2 cache size)?

In other words, why should I get a Barton? and when will I know if a motherboard supports the barton?

 

[CougTek]

Regarding your 300$ budget box

I have a $300 budget box?

I was talking to Mercutio.

I might be able to do a reasonably nice P4 Celeron at a $300-ish price point (see "fleet computer challenge" if you can't understand my interest in this).

But if Blakerwry is also interested into a 300$ (please, allow at least 350$), I can provide suggestions too.

You can have a lot more for 300$-350$ in the US than you could get here. I will buy a little house someday in the States, just so I can have an address to receive cheap computer parts.

 

[CougTek]

Barton is supposed to support 400MHz FSB, or so did I read. It will also be launched at lower frequencies than the elder Thoroughbred, so its rating won't be that much higher. I wouldn't wait for it if I were you. Toroughbred rev.B is the hot CPU to get.

 

[blakerwry]

what is the max slated barton speed? I have seen as high as 3200+. If the Barton runs cooler with similar performance to the t-bred core and has a higher possible speed then I'm interested.

 

[Mercutio]

I have an almost permanent interest in low-cost but reasonably high quality PCs. Probably more than others here, because of the area where I live and the markets that I serve. Oh, and the fact that my hobby is basically paid out of pocket.
The last time I could really put together a great $350 PC, FIC 503+ and AMD K6-2/350s were priced below $90 as a pair and memory prices were at a low point. A $350 PC that doesn't make many sacrifices in quality or performace is a great thing to pull off.

Anyway, I have a budget of my own now, so cheap, decent PCs are a good thing, in my opinion.

None of this has anything to do with overclocking, of course.

I've been thinking about getting something and overclocking it. But the 2.4GHz P4 I have leaves me limp, performance-wise (yes, the $200-if-I-get-all-my-rebates Dell machine) and I have no idea how fast a P4 has to be before it encodes video more than two tenths of a frame/sec faster than my XP2100.

 

[blakerwry]

They have a new p4 optimized version of Virtual dub so we can pit our computers against each other.. my sb51g w/ 2.0 Celeron & 512 pc2700 and your 2100+

 

[Clocker]

I'm overclocking my two Athlon XP 1700+'s to 2000+ speed using ~150Mhz on my ECS K7S5A's at the stock voltage. I'm too lazy to unlock the processors and I don't think the board has multiplier adjustment, anyway.

I'm overclocking my 2400+ to 2600+ speed (2.1Ghz rather than 2.0) at the stock voltage. I'm overclocking the FSB & RAM from 133Mhz (266DDR) to 200Mhz (400DDR) on the Epox 8RDA+ (nForce2) board it is in.

The Epox 8RDA+ unlocks the multiplier for you for any AthlonXP 2400+ or faster.

C

 

[Groltz]

I will buy a little house someday in the States, just so I can have an address to receive cheap computer parts.

OT: How far are you from the Canada/US border anyway, Coug? Is there no way you can set up a mail-drop on the US side?

 

[honold]

The last time I could really put together a great $350 PC, FIC 503+ and AMD K6-2/350s were priced below $90 as a pair and memory prices were at a low point. A $350 PC that doesn't make many sacrifices in quality or performace is a great thing to pull off.

the 503+ is one of the worst motherboards i've ever dealt with, heh. only had two of them with k6-2/300 cpus, the agp stuff was a nightmare. i wish i could remember the details, but i may have repressed the memories.

of course, if the 503+ is second worst that would make the 'bad motherboard list' look about like this:

1) abit bp6
2) abit bp6
3) abit bp6
...
10,647) abit bp6
10,648) abit bp6
10,649) fic 503+

 

[Mercutio]

It was a wonderfully stable board. They were fantastic when combined with Permedia2 or Voodoo3 cards. I'd rate the VA-503+ as by-far the best of all socket 7 boards (this after endless exposure to PC Chips, Shuttle, DFI and Asus SS7 boards).

Excluding PC Chips, I think my worst experiences have been with Shuttle (pre AK32L), Octek and Abit boards. Fortunately I avoided contact with the BP6.

 

[honold]

i believe i was using a creative labs voodoo banshee on both systems

 

[CityK]

You can have a lot more for 300$-350$ in the US than you could get here. I will buy a little house someday in the States, just so I can have an address to receive cheap computer parts.

I agree. But that's only half the story; the other side is product availability. I need a little State side house of my own, just so I can get some of the products that never seem to make it to Toronto, or at least sure take their sweet time getting here.

CK

 

[LiamC]

Overclocking experiences

Currently my system is T'bred A 1900+ (ES - unlocked), EPoX 8KHA+, 512MB Crucial PC2100, TripleXXX GF4 4200 64MB, WD 600JB, Volcano 7+ (1/2 kilo coppa whoppa)

System runs as 2200+ via 150MHz bus (300DDR) and 12X multiplier. Memory latency times as low as I can squeeze then. I do development work on this with no problem (except for my own stupid coding mistakes). I back up regularly.

My server

AXIA T'bird 1GHZ @ 1.33GHz @1.8V - Volcano 6 Cu, IBM 75GXP (boot - I know, I know!), 1000JB (Data), EPoX 8K7A+, 512MB (256 Crucial, 256 Kingston ValueRAM . PCI, AGP within spec. Memory timings as above.
I used to overclock this like I do to the 8KHA+ now, but the on-board HighPoint controller won't tolerate out of spec PCI speeds. Disable that and it overclocks like a champion.


My wife's
I don't touch it any more on fear of death.

Previous server

AMD 550@650 SLOT A FIC SD-11 (thanks Tony), GlobalWin FKK32, NinjaMicro O/C device, 512MB KingMAX PC-150



I do most of my o/cing by FSB as a faster bus leads to better speed than a higher multiplier, so long as you don't go too crazy and you know the importance of tRAS.

OC systems rated as far as ease/stability goes

Celeron 300A/Abit BH6
EPoX8K7A+
EPoX8KHA+


FIC SD-11

 

[Mercutio]

i believe i was using a creative labs voodoo banshee on both systems

I'd say "No wonder" but you probably wouldn't believe me. Pretty much all the Super7 boards had a lousy AGP implementation to begin with, but the Voodoo3 at least basically treated an AGP slot like a faster PCI slot, instead of using all the other crap that AGP is supposed to do. And I think that's one of the reasons that the Voodoo3s were such good, stable cards.

The Banshee was probably the biggest bastard of its day. Forebear to the V3, I know, but closer in reality to the Voodoo Rush, the thrice-damned first effort on the part of 3dfx to make a 2d/3d card. 3dfx learned from a lot of their mistakes on both the rush and the banshee. I had a couple of run-ins with that one as well.

 

[honold]

i know, and i know

strange story, i actually soldered some stuff on the pre-production banshee like 6 or 7 years ago when cl usa was writing drivers for it. they had sent them a 64bit card which had to have some TINY stuff reworked on it to make it work on a 32bit pci slot. for some reason, they came to compusa for that service (which we didn't do).

i did it for free with my own soldering iron because we were dead, and they said they'd send me some 'early stuff'.

a creative labs poster

 

[CougTek]

OT: How far are you from the Canada/US border anyway, Coug? Is there no way you can set up a mail-drop on the US side?

Many web stores won't deliver to P.O. boxes. Anyway, I figure that it must be a hassle to setup one of those for a non-US resident. Since I sometimes wear a beard, they could think I'm an evil arab terrorist and refuse my request ;-) I'm at least at an hour and a half of the lines. I think the closest State of here is Maine (biker's heaven), but I would have to check to be sure.

Anyway, I'm not about to buy any property in the U.S.A. in the short and middle term future. Wallet forbids me.

 

[Newtun]

FSB Limitations

How often, if ever, is an Athlon "FSB limited" when it isn't "MHz limited"?

For instance, an unlocked Tbird that works fine at 100 MHz FSB × 12 multiplier, but doesn't work at 133 MHz × 9, or that works at 100 × 12.5 but not at 166 × 7½?

Or an unlocked XP Tbred that works at 133 MHz × 12½ but not at 166 × 10?

 

[blakerwry]

for the most part, CPU's have a max speed.. whether you attain it through multiplier adjustments or through FSB adjustments doesn't matter much to the CPU.

However, your motherboard might not run stabily as you get to higher FSB clock speeds(especially faster than the processors it was designed to run) Your Northbridge will probably also get hotter as you increase FSB, requiring a decent heatsink and maybe a fan.


Also, adjusting the FSB has other implications, all busses on the system are obtained through the use of dividers reletive to the FSB(RAM, Hyper transport, AGP, PCI, ISA) If your motherboard is running these things out of spec then they mail fail to work correctly.... this is 1 reason why multiplier adjustments are considered to be a more reliable way of overclocking(less possible point of failure)

 

[honold]

most decent overclocker's boards have an ability to 'fix' the agp/pci speed regardless of cpu bus speed

 

[blakerwry]

the newer ones, yes... some are always fixed, some have a PCI lock and some have an AGP lock and some have both.

 

[blakerwry]

oh, I forgot to mention that I have read all t-bred B CPU's are unlocked... but you'll need a mobo that can adjust multipliers to take advantage of this...


Aparently Soyo has a beta BIOS(released yesterday) that will now use the t-bred and can adjust it's multipliers... yeah....


Epox is also known to be able to do this on their boards.

 

[CityK]

Apparently, from what I've seen (repeatedly) in user forums, the 1700+ & 1800+ T-bred "A" chips are also factory unlocked.

CK

 

[blakerwry]

that's what I've heard to, but haven't cofirmed it....

I guess maybe they did this so that you could run the 166mHz chips at 133 if your board didn't support 166....

but then, you have to think that if you buy a 133 FSB t-bred, you are probably going to be able to run it at 166 w/ a lower multiplier if you wish... making the factory 166mhz ones less apealing.

 

[Cliptin]

I want to know what speed the T-bred b starts at and how it differes from the t-bred A...

AMD is also supposed to move to rev.B Thoroughbred even for the 1700+ and 1800+ models eventually, though I don't think you'll find anything else then rev.A Thoroughbred at less than 2400+ rating currently.

The clearest identifier for me has always been the micron size for the die. Observing the descriptions on Pricewatch, the 2200 are .13 micron and the 2100 are .18 micron. Is this a valid identifier? What happened to .15, I recall this being the next step after .18?

 

[blakerwry]

.25 is the original athlon classics(slot A) .18 micron is palomino, T-bird, all Durons and model 2 slot athlons

.13 include the T-breds (both A and B)


I'm not sure about the mobile athlons... They look good for SFF PC's becasue of their low power/low heat specs so I would assume they are t-bred.

 

[LiamC]

Original Slot A (up to 650Mhz) was 0.25um
700 Slot A was hybrid 0.22um
750 Slot A + (512KB cache) & all Thunderbird Slot A's (256KB cache) were 0.18um Al

Most Thunderbird Athlon and Spitfire Duron Socket A below 950MHz were 0.18um Al

All Thunderbird Athlon > 1GHz were 0.18um Cu

I _think _ that all Morgan core Durons are 0.18 Al

All Palomino's are 0.18um Cu
All Throughbred are 0.13um Cu

There is no Thoroughbred A faster than 2200+

_My understanding_ is that most Thoroughbred's 2200+ or slower are "A" variants.

All Thoroughbreds faster than 2200+ "B" variants.

There may be the odd exception to these rules, but they will be very rare.

As limited production of "B" variants did not get into market until Nov/Dec, it will be six months (or more) the transition to the "B" stepping photomasks has been completed and the channel flushed of old inventory.

The "B" stepping has an added metal layer so it will cost more to produce (more to produce, another couple of chances to fail or perform poorly), so AMD won't be in a super hurry to transistion. Plus they have Barton and Athlon64 to worry about. I'd hate to be an AMD process tech/engineer right now.

 

[LiamC]

Cliptin,

you can make cells any size you want. 0.15um had some support because 0.18um lithography tools could be re-used with some re-jigging to "scribe" the smaller photomasks (gross simplification).

0.13um was too big a step as the light wavelengths used were too large. Every semi-con company I know has at some stage tried to re-use there lithography tools (or most of them) for the next generation to save costs - because these tools cost tens and hundreds of millions.

Intel did it at 0.25 to 0.18 and invested in new tools for 0.13 - and plans to use their 0.13 tools for 0.09 - bugger that 190nm.

AMD had new tools for 0.18um (partly becasue of the change to Cu) and I think showed how big a lead they had.

AMD re-used there 180nm tools for 130nm and it showed - the AMD 130nm process isn't as good as Intels. It will probably change back to AMD's favour at 90nm. The 90nm Hammers should be something special.

But back to your question, there was a talk of 150nm stuff, but the chip race was in full swing and while yields bin/splits and costs would certainly have been better for AMD if they hadn't been so ambitious, it's the top speed chips that garner revenue so there hand was forced.

Debugging a new chip fabrication process is costly and time consuming, so they probably bit the bullet.

NVIDIA and ATi are in this up to there necks as well. ATi banked on a stable and mature 150nm process, NVIDIA took a gamble. NVIDIA lost. if and when we see GeForceFX, it should clock insanely, but by the time NVIDIA (or more correctly TSMC) can get enough of them into the hands of consumers, ATi will most likely have a 130nm part AND a new revision of the Radeon, so when things matter to profits - ATi will be a contender.

But what would I know...