The 90nm SOI Sempron s754 processor

[Avalon]

Originally posted by: blckgrffn
I have a 3400+ @ 2.58, DDR400 @215 1T, 6800 NU @ 385/875, DFI Lanparty.

I hate steam too :| but I love the games, so I guess that is where are all stuck...

Later,
Nat

Nat, did you say if your 6800NU was unlocked or not? I can't remember, and can't seem to locate it in this post, although I was pretty sure you had said it somewhere. Anywho, I ran those lower AA tests for you. I ran the game at 2xAA and 8xAF forced in my control panel, and I dropped my image quality slider from high quality to plain jane quality. I scored 50.3fps in Doom 3 at 16x12, which is quite impressive, and the game looked just as fantastic to boot. Are you using the in-game 8xAF that is enabled when playing on the in-game high quality setting, or are you forcing 8xAF in your CP?

47.3fps if I turn my CP quality settings back to high quality, but leave my AA at 2x.
It seems like the lack of an extra 128MB memory is what makes the huge difference between 2x AA and 4x AA. Even softmodding my card only allows me to gain a few extra frames over you, as at that point it's mostly memory bandwidth determining the performance of the game. I don't think you'd really need 4x AA in Doom 3 anyway. The game is dark, and you're spending most of your time wetting your pants, so I doubt you'd notice any jaggie lines at 2x.

I also just ran some FarCry benchies. The scores are almost exactly the same as my 2.3 barton. FarCry seems GPU dependant almost right off the bat in my situation.

 

[sangyup81]

I got the Sempron 2600+ and the DFI Lanparty NF3. Waiting for OCing results so I can share with you guys. :-D

 

[Avalon]

I wish I still had my 6800GT with me, so I could see how much better I'd score with it in my new system over my barton system. My high res comparisons don't really offer much insight, as I'm utterly GPU limited. I can't remember who made the thread that said the only good upgrade for games is a video card, but they were right. I won't be benefitting very much from my new CPU except for newer games that I'd be playing 10x7 resolutions at, or my older games like NS which were CPU dependant because the video card had no problems smoking the Quake 3 engine.

Hope your 2600+ OC's well. Let us know.

 

[Toro 45]

335HTT - dropped mymem ratio to 120. Still going. I'm getting scared. Hold me.

LOL, I here you I took my dfi over 300 and it does make one a bit nervous

So where are you settling in at on this chip and at what voltage, 325 x 8 = 2.6ghz?

 

[Avalon]

Originally posted by: Toro 45

335HTT - dropped mymem ratio to 120. Still going. I'm getting scared. Hold me.

LOL, I here you I took my dfi over 300 and it does make one a bit nervous

So where are you settling in at on this chip and at what voltage, 325 x 8 = 2.6ghz?

325x8 with 1.5v, my LDT multiplier at 2.5x for an effective HT rate of 815mhz (a little over spec, but it seems my system can handle it) and 133mhz mem divider, effectively having my memory run at 216mhz with timings of 2-3-3-10 1T. My memory seems to like voltage, so I may try juicing it with the 3.1v that this board can provide and see if I can't go 2-2-2. One of my sticks can do 2-2-2 up to 215mhz at 2.85v, while the other one can barely do it at 200.

Actually...I just got my tax return for $200, and I've been eyeing those TwinMOS speed premium modules on Newegg said to be untested Winbond UTT chips used in OCZ VX. Cheap and tempting...

 

[Toro 45]

1ghz overclock on a $90.00 cpu, I'm very impressed. I was eyeballing those but wasn't to sure about running such a high fsb. So I snagged a A64 2800 for an extra $20.00 but I doubt it will come close to the 2.6 your sempy is running.

I'm glad to see yours hangin at such a nice overclock. The DFI UT board is pretty slick are you concerned about the heat on the chipset mine gets really hot at 300fsb?

 

[blckgrffn]

Thanks for checking!

My 6800 is locked, artifacts happen all over when I unlock it. That sounds about right that you get 50fps, and I agree that 256meg and more bandwidth is needed to really play the latest games at high res+AA. I am using the in game settings for everything in Doom3. When that game came out, everybody said that it looked awesome no matter what the resolution, but I don't think that those reviewers saw it at 16*12 w/AA, had they? The thing is that games look so much better when you can play at 1600*1200 with at least 2xAA (which I think makes a big difference in Doom3, cleans up the few jaggies that seem to appear on straight edges). For that reason, I hope to be upgrading to a Gainward 6800GT GS this week

Congrats on the great OC and the comprhensive thread, am I sure that this will be used for a reference for quite some time to come I think that I will be using the A64 2800+ for my linux box afterall for the 64bit coolness factor, but for $30-$40 less this seems mighty tempting...

Nat

 

[sangyup81]

Blah

My results are better than Budman's but nowhere near Avalon's. I'm trying to get to my prime95 stable speed and I've eliminated 310*8=2480 which lasted 3 minutes in prime. Dropped it to 308*8=2464. Hopefully I can get that to go over an hour.

BTW, I'm at 1.55v. At stock voltage (1.4), the chip couldn't even last a minute in prime95 at 300*8=2400.

My 3000+ Winchester did way better. I have that one at 285*9=2567 1.55v prim95 stable (which for me is over 12 hours)

 

[Budman]

Originally posted by: sangyup81
Blah

My results are better than Budman's but nowhere near Avalon's. I'm trying to get to my prime95 stable speed and I've eliminated 310*8=2480 which lasted 3 minutes in prime. Dropped it to 308*8=2464. Hopefully I can get that to go over an hour.

BTW, I'm at 1.55v. At stock voltage (1.4), the chip couldn't even last a minute in prime95 at 300*8=2400.

My 3000+ Winchester did way better. I have that one at 285*9=2567 1.55v prim95 stable (which for me is over 12 hours)

at least I'm not the only one getting lower than expected speeds on these semprons.

 

[Toro 45]

sangyup81, I know your board is very capable but have you ever run your Winchester at 310 like your trying with the Sempron? Also did you try raising the chipset voltage? Just trying to help you squeeze a bit more out of it.

 

[sangyup81]

I've booted into Windows at 333htt but it doesn't do me any good if it's not stable. I did raise the chipset voltage, first to 1.7 and then 1.8.

But if it were a chipset problem, I would expect much worse then prime95 stability such as not being able to boot into Windows for one or not being able to boot up period.

Still, the Sempron 2600+ is $80 while the Winnie is $140 so I'm definitely not unhappy with being able to get 2.4 with the Sempron.

BTW, I'm using a 550W Mad Dog Power Supply. I read in another threat that those are made by Topower, the same guys who make the OCZs, so I don't think it's the PSU. I'm getting over 11.7 on the 12v rail which is within 2.5%. Most people say 5% is acceptable right? If only the power would stop going out at this house, I can have the prime results!

 

[Avalon]

Originally posted by: Budman

Originally posted by: sangyup81
Blah

My results are better than Budman's but nowhere near Avalon's. I'm trying to get to my prime95 stable speed and I've eliminated 310*8=2480 which lasted 3 minutes in prime. Dropped it to 308*8=2464. Hopefully I can get that to go over an hour.

BTW, I'm at 1.55v. At stock voltage (1.4), the chip couldn't even last a minute in prime95 at 300*8=2400.

My 3000+ Winchester did way better. I have that one at 285*9=2567 1.55v prim95 stable (which for me is over 12 hours)

at least I'm not the only one getting lower than expected speeds on these semprons.

You really have to pull out all the stops to find something that will allow you to reach higher speeds. I hit several walls before I made it to where I'm sitting at now. Also, you said earlier that your PSU is probably limiting you. I'm sure if you pick up something beefier, you'll be able to reach my point.

Sangyup: I don't trust your PSU. Mine has been proven solid at OC'ing, so I know I can drop any chip into my setup and be guaranteed that the peak I reach truly is the maximum of the chip. Just because a PSU is made from a well known manufacturer, does not mean the PSU will be good. Even good companies can and will put out some very cheap stuff.

Also, don't forget all the tricks along the way...keep your memory within reasonable specifications with dividers, make sure that your HT frequency never goes above 800mhz, make sure cpu spread spectrum is disabled, set your memory TRAS to 10, etc.

Toro: To be honest, I have not felt my chipset temp for myself. I'm standing on carpet with socks on, so I don't think I need to be touching metal in my system right now
Putting a finger close to the heatsink, though, does feel kind of warm. I am not worried though. If DFI designed their motherboard to do such extreme speeds and put faith in a passive aluminum sink to do it, I'll trust their judgement. If it fries, I'll get a new one.

 

[sangyup81]

Originally posted by: Avalon
Sangyup: I don't trust your PSU. Mine has been proven solid at OC'ing, so I know I can drop any chip into my setup and be guaranteed that the peak I reach truly is the maximum of the chip. Just because a PSU is made from a well known manufacturer, does not mean the PSU will be good. Even good companies can and will put out some very cheap stuff.

Also, don't forget all the tricks along the way...keep your memory within reasonable specifications with dividers, make sure that your HT frequency never goes above 800mhz, make sure cpu spread spectrum is disabled, set your memory TRAS to 10, etc.

Avalon, I've overclocked Bartons with this PSU before. The 754 setups is much less taxing on my rails than the Socket A setup was. +11.7v is my rail under load. I get +11.9v when I go idle. I mean just how good does a PSU have to be? And I don't think this can explain a 280+ mhz discrepency. I've had different power supplies and although this doesn't have rails like the OCZ Powerstream, it's way better than the Thermaltake I used to use and even that one was praised by forum people.

Spread spectrum is disabled and I do have the TRAS at 10 and I even set the divider to 100 to take the memory out of the equation. Now here's something odd. When I use 2.5x htt multiplier, the board doesn't boot. It works fine at 300x3=900 but not for 300x2.5=750. 333x3=1000 boots but 333x2.5=833 doesn't.

Perhaps the quality of the L2 cache is lower on the 2600+ than the 2800+? Do you know if the 2600+ just has 128k cache or it is a 256k cache chip with half disabled? I believe Budman's 3000+ is a 128k cache one? And yours is a 256k cache one correct?

If more people would try these out, we could find out!! =T

 

[DrMrLordX]

The 2600+ and 3000+ 754 Semprons are indeed 128k l2 cache models. The 2800+ has 256k l2 cache. I don't know if the 128k chips are "failed" 256k chips, or if they're both "failed" 512k chips, or what.

 

[Avalon]

Originally posted by: sangyup81

Originally posted by: Avalon
Sangyup: I don't trust your PSU. Mine has been proven solid at OC'ing, so I know I can drop any chip into my setup and be guaranteed that the peak I reach truly is the maximum of the chip. Just because a PSU is made from a well known manufacturer, does not mean the PSU will be good. Even good companies can and will put out some very cheap stuff.

Also, don't forget all the tricks along the way...keep your memory within reasonable specifications with dividers, make sure that your HT frequency never goes above 800mhz, make sure cpu spread spectrum is disabled, set your memory TRAS to 10, etc.

Avalon, I've overclocked Bartons with this PSU before. The 754 setups is much less taxing on my rails than the Socket A setup was. +11.7v is my rail under load. I get +11.9v when I go idle. I mean just how good does a PSU have to be? And I don't think this can explain a 280+ mhz discrepency. I've had different power supplies and although this doesn't have rails like the OCZ Powerstream, it's way better than the Thermaltake I used to use and even that one was praised by forum people.

Spread spectrum is disabled and I do have the TRAS at 10 and I even set the divider to 100 to take the memory out of the equation. Now here's something odd. When I use 2.5x htt multiplier, the board doesn't boot. It works fine at 300x3=900 but not for 300x2.5=750. 333x3=1000 boots but 333x2.5=833 doesn't.

Perhaps the quality of the L2 cache is lower on the 2600+ than the 2800+? Do you know if the 2600+ just has 128k cache or it is a 256k cache chip with half disabled? I believe Budman's 3000+ is a 128k cache one? And yours is a 256k cache one correct?

If more people would try these out, we could find out!! =T

A s754 setup should require more juice going this high than a barton setup. Even if your s754 shows better rails, you'll have to use a voltimeter to get a true reading. On my old mobile barton setups with my Epox 8RDA3i, my old Enlight PSU used to sit around 11.95v on the 12v rail with my 9700 pro, but when I added a 6800NU, it dropped down to 11.7, which is within variable spec, but when you're overclocking, you're talking about the utmost sensitivity to the slightest change or weakness in voltage. I had to back down 150mhz on my overclock. My old PSU was an enlight 420w with 18A on the 12v. Now I'm using a 460w Enermax with 33A on the 12v rail. What specs does your Mad Dog PSU have? I'd suspect that even if the specs looked good on paper, they may be overexaggerated.

You really do need a solid, trusted PSU for this kind of work. I would highly recommend mine for $80 if you don't want to fork up for another Powerstream. As far as the LDT multiplier...you're using the DFI, right? Mine has that issue when I use 2x LDT, except I get past boot screen, but get to this new screen that tells me it's waiting to initialize the nvidia raid array. My 2.5x and 1.5x settings work fine, though.

As far as the chips...the 2600+ and 3000+ are 128KB L2 cache chips, while the 3100+ and 2800+ have 256KB. I do have a slight feeling that the 128KB chips might be ones with disabled defective cache. If the cache is disabled, who knows what else on that silicon is faulty or ineffective?

 

[Toro 45]

The reviews said the chipset h/s got very hot on testing and when I touch mine it is close to burning my finger hot, if you don't have one already you may want to have an intake fan blowing on it.

Could you do me a favor when you get a chance, I noticed you got 37s on Super PI 1m. I can barely break 40 seconds with the wind at my back on a down hill slope, on my 2800@2.4ghz. ram speed same as yours(216fsb 2-3-3-7)

So if you could test your Sempy at 300x8=2400mhz / ram on 140 Divider=210fsb 2-3-3-7(ram) I'd really like to see how it compares to an A64. Mine usually does 40 seconds and once I got 39 seconds. I can get 40 seconds with my mobile xp 2600 set up. No hurry, thanks

 

[Avalon]

Originally posted by: Toro 45
The reviews said the chipset h/s got very hot on testing and when I touch mine it is close to burning my finger hot, if you don't have one already you may want to have an intake fan blowing on it.

Could you do me a favor when you get a chance, I noticed you got 37s on Super PI 1m. I can barely break 40 seconds with the wind at my back on a down hill slope, on my 2800@2.4ghz. ram speed same as yours(216fsb 2-3-3-7)

So if you could test your Sempy at 300x8=2400mhz / ram on 140 Divider=210fsb 2-3-3-7(ram) I'd really like to see how it compares to an A64. Mine usually does 40 seconds and once I got 39 seconds. I can get 40 seconds with my mobile xp 2600 set up. No hurry, thanks

Will do. I'll be right back and let you know.

41s.

 

[Toro 45]

Great, thanks for checking that. Looks like I'll have to hit 2.5ghz or better to see 38-39seconds consistantly.

 

[sangyup81]

Originally posted by: Avalon
A s754 setup should require more juice going this high than a barton setup. Even if your s754 shows better rails, you'll have to use a voltimeter to get a true reading. On my old mobile barton setups with my Epox 8RDA3i, my old Enlight PSU used to sit around 11.95v on the 12v rail with my 9700 pro, but when I added a 6800NU, it dropped down to 11.7, which is within variable spec, but when you're overclocking, you're talking about the utmost sensitivity to the slightest change or weakness in voltage. I had to back down 150mhz on my overclock. My old PSU was an enlight 420w with 18A on the 12v. Now I'm using a 460w Enermax with 33A on the 12v rail. What specs does your Mad Dog PSU have? I'd suspect that even if the specs looked good on paper, they may be overexaggerated.

You really do need a solid, trusted PSU for this kind of work. I would highly recommend mine for $80 if you don't want to fork up for another Powerstream. As far as the LDT multiplier...you're using the DFI, right? Mine has that issue when I use 2x LDT, except I get past boot screen, but get to this new screen that tells me it's waiting to initialize the nvidia raid array. My 2.5x and 1.5x settings work fine, though.

As far as the chips...the 2600+ and 3000+ are 128KB L2 cache chips, while the 3100+ and 2800+ have 256KB. I do have a slight feeling that the 128KB chips might be ones with disabled defective cache. If the cache is disabled, who knows what else on that silicon is faulty or ineffective?

30 amps on the +12v. I've been using the 9800pro for both setups. I would think that since you pump more voltage to the Socket A, it would pull more juice. I will play with the LDT settings again today. We're using the same board right? Power supply shouldn't affect LDT stability should it?

Man, I am temped to fork out for a 2800+ Sempron and see if my theory is right. It's roughly the same $$ as getting your PSU. ;-)

BTW, I do have a 420W OCZ Powerstream. I'm gonna play around with that one and see if I get a better overclock.

 

[Avalon]

Originally posted by: sangyup81

Originally posted by: Avalon
A s754 setup should require more juice going this high than a barton setup. Even if your s754 shows better rails, you'll have to use a voltimeter to get a true reading. On my old mobile barton setups with my Epox 8RDA3i, my old Enlight PSU used to sit around 11.95v on the 12v rail with my 9700 pro, but when I added a 6800NU, it dropped down to 11.7, which is within variable spec, but when you're overclocking, you're talking about the utmost sensitivity to the slightest change or weakness in voltage. I had to back down 150mhz on my overclock. My old PSU was an enlight 420w with 18A on the 12v. Now I'm using a 460w Enermax with 33A on the 12v rail. What specs does your Mad Dog PSU have? I'd suspect that even if the specs looked good on paper, they may be overexaggerated.

You really do need a solid, trusted PSU for this kind of work. I would highly recommend mine for $80 if you don't want to fork up for another Powerstream. As far as the LDT multiplier...you're using the DFI, right? Mine has that issue when I use 2x LDT, except I get past boot screen, but get to this new screen that tells me it's waiting to initialize the nvidia raid array. My 2.5x and 1.5x settings work fine, though.

As far as the chips...the 2600+ and 3000+ are 128KB L2 cache chips, while the 3100+ and 2800+ have 256KB. I do have a slight feeling that the 128KB chips might be ones with disabled defective cache. If the cache is disabled, who knows what else on that silicon is faulty or ineffective?

30 amps on the +12v. I've been using the 9800pro for both setups. I would think that since you pump more voltage to the Socket A, it would pull more juice. I will play with the LDT settings again today. We're using the same board right? Power supply shouldn't affect LDT stability should it?

Man, I am temped to fork out for a 2800+ Sempron and see if my theory is right. It's roughly the same $$ as getting your PSU. ;-)

BTW, I do have a 420W OCZ Powerstream. I'm gonna play around with that one and see if I get a better overclock.

If you're using the DFI Lanparty UT, yes, we have the same board. I'm not sure how a PSU would affect LDT stability. I've never heard of it being connected. It just seems that some of these DFI's are very quirky with their LDT settings. I'd say try your 420w powerstream with your setup, and see if things improve. I'm not familiar with the specs of a 420w powerstream, so I'll have to look it up when I get back from work. I still don't trust your Mad Dog, though. For PSUs, I always go quality over quantity, and for me, that means only brands that are highly regarded. I firmly believe your PSU is your most important piece of hardware.

 

[sangyup81]

I use the Mad Dog because it looks better haha. But I guess that's no longer the point.

Avalon, I'm going to disable my L2 Cache and see what it does for overclockability and SuperPi scores.

 

[sangyup81]

8*303 1.5v Prime95 stable for 17 hours and then I stopped it. I don't have that full suite of tests you did but just these ought to give a rough idea if the L2 cache matters much.

With 128k cache enabled:
SuperPi 1M: 40s
CS:Source Stress Test: 109.45fps

With 128k cache disabled:
SuperPi 1M: 99s
CS: Source Stress Test: 73.15fps

Alright, that's quite a hit. I don't think I will even bother to OC with the L2 Cache disabled now.

 

[Avalon]

Originally posted by: sangyup81
8*303 1.5v Prime95 stable for 17 hours and then I stopped it. I don't have that full suite of tests you did but just these ought to give a rough idea if the L2 cache matters much.

With 128k cache enabled:
SuperPi 1M: 40s
CS:Source Stress Test: 109.45fps

With 128k cache disabled:
SuperPi 1M: 99s
CS: Source Stress Test: 73.15fps

Alright, that's quite a hit. I don't think I will even bother to OC with the L2 Cache disabled now.

I didn't even know you could disable the L2 cache...lol.
Let me know how it goes when you hook up your powerstream. I checked the specs on it, and it seems good, although slightly overpriced. Using that, you will know if it's your mad dog or the cpu limiting you.

 

[sangyup81]

You can disable the L2 Cache by choosing Disable for External Cache in your Advanced Bios menu (or something like that)

Well if the Power Stream does have one thing on other power supplies, it's that you can adjust the rails.

And I do think I can get a better OC with the Power Stream but I don't think it can make up for the discrepency of our results. Ah well, I'll have results tomorrow.

 

[DrMrLordX]

I got my Sempron 2800+ and other parts in on Saturday. Everything's up and running . . . mostly. The DIMM that Mwave shipped to me was bad, so I'm RMAing it for a refund and getting two of these

http://www.newegg.com/app/ViewProductDesc.asp?description=20-146-541&depa=1

They should do better than the 1 gig Kingston value RAM I got from Mwave. Cheaper, too.

For the time being, I'm using my old PC2100 from my old machine. Blah, I say.