Anyone got a T'Bred core AMD chip laying around after a Barton upgrade? That they'd be willing to sell, cheap?

Finally figured out why my 1.4ghz AMD 1600 Palomino is producing at 1/3 the rate of my 1.47 ghz AMD 1700 Thoroughbred. No SSE instructions.

It has 3DNow!, but no SSE. On the same WU (copied from one computer to the other for testing) the 1700 was getting ~15 minute frame times, the 1600 was getting ~45 minute frame times. Turned off SSE on the 1700 and got the same 45 minute frame times.

So, now the hunt begins for an inexpensive T'Bred for my main system.


Anyone?

Newegg.com has a 1900+ for $50 shipped. Pretty good deal.

Or perhaps this cheap 1700+
http://www.tufshop.com/store/product2487.html

Yeah, I'll probably go ahead and get something from Newegg. Wish they still had the 1700 T'Bred, last they did it was down to $42.

And if that 1700+ at Tufshop is like the picture, that's the Palomino core. The T'Bred core is smaller and rectangular compared to the larger, square Palomino core. Also, says it has "3DNow! with full SSE compatibility", not that it has SSE.

Have to see how the budget holds out this week.

How timely. I was just commenting in another thread about maybe putting another machine together out of parts laying around from past upgrades. The cpu I have says AX1800DMT3C. The core is fairly large and square. So it does not have SSE and will not make a good Folder? Maybe that's why my XPC is such an underachiever also. CPU-Z says it's a 2000+ Palomino. So for $50 (for a 1900+ T-bred) I could triple the production from that machine? That'd be sweet, especially considering that's it's sole purpose in life.

David

Load up SiSoft Sandra benchmark program. (or equivalent) The CPU information section will let you know what goodes your CPU has (or doesn't have).

If you're using -forceasm for the client, check your log and see whether it says "3DNow! enabled" or "SSE enabled".

As I mentioned elsewhere, my P3-566mhz backup server almost has the same PPW rating at this 1.4ghz Palomino chip.

I have -forceasm but the log says nothing about either 3DNow or SSE. Since I added the -forceasm switch it has been working on a Tinker, which maybe doesn't say anything about 3DNow or SSE because it doesn't specifically support them?

CPUID says both 3DNow and SSE are supported.

David

Well, over the weekend I discovered the V4 beta console and installed that. Probably the only new feature that caught my eye was a forceSSE switch. Beta consoles are available here.

While checking to see if I could (secretely) get the extra 100 points a day from it (to stay ahead of the hoard trying to steal my place ), I noticed:

1) No real speed difference
2) You need the latest 1.53 Gromacs core to benefit from all new features - just delete FAHCore*.exe from your directory & restart, the new core will be downloaded
3) I can't teach EM to find anything except FAH3Console.exe, so you have to rename the exe to that.
4) With no switches on my t'bred, it fires up in 3DNow mode. Took 21min44 for a frame.
5) With forceSSE set, picks up SSE and took 9min30 for a frame in same WU. Note, in forceSSE flag the SSE must be uppercase, allegedly
6) Got clever and swapped both my boxes to forceSSE flag only. Sure as day follows night 1 box finished it's Gromacs and collected a Tinker. Deleted the Tinker, set it off again and it collected another Tinker. Deleted that one, added back advmethods flag and got Gromacs again.

Bottom line with v4 console: you still need advmethods to improve chances of getting Gromacs. You should use forceSSE if your CPU has SSE.

I have a machine with a 1800+ Palomino. It does have SSE. When folding the same protein, the 1800 takes about 15 minutes/frame and my 2500+ Barton 12 minutes. So I don't know from which model of Palomino AMD started to include SSE support. You could try in this post. It does not tell anything about SSE support but maybe from there you could find.

It's clear than the most important factor is the ability of the chip to handle SSE instructions. Also you should consider than a TBred will disipate much less heat than the Palomino at the same speed.

My 1800+ Palomino finished its Tinker and got a Gromacs, and now the log says

[09:09:09] Extra 3DNow boost OK.

There is no way to force SSE on with the 3.24 client, so I think I will download the Beta 4 client and see what happens with -forceSSE.

David

Dukla2000, I've been thinking of installing the 4.0 beta, but not sure about if it will have errors that finally will make it worse in performance. Have you expierienced any errors so far?
So you said that when 3dNow enabled the new client behaves faster than the older? How much faster? Just with the new client or with the new Gromacs core?

Haysdb, this is really odd. I'm quite sure that my 1800+ is a Palomino one. I didn't know when I bought it but later I checked with the number in the chip.

It's doing a gromacs core right with the 3.25 client version and -advmethods -forceasm on and it says :

[10:22:58] Extra SSE boost OK.

So, I don't know how can it be. Maybe some of the Palominos have SSE and some don't have? Just to know, when you bought it? I bought it when it was one of the slower/cheaper processors of amd avalaible.

And there is a way to force sse with the old cliente, -forceasm, it's only that it wasn't to acomplish that function that the flag was maked.

OK, I am now running the beta 4.00 client on my 2000+ Palomino. As mentioned in a previous post, CPUID says this CPU has SSE. On a Gromacs core, my times per frame has dropped from 17:05 to 8:48.

Here is what I did:

• Downloaded the beta 4.00 client (console version), installing it into the same directory as the existing 3.24 client.
• Stopped the old client (which was running as a service).
• Created a shortcut for the new client and added "-local -advmethods -forceSSE" at the end of the "Target" field.
• Deleted the old cores to force download of new ones.
• Double-clicked the new shortcut to fire up the client and checked all the messages. Other than the dire warning about the use of -forceSSE, everything looks good.
  
  Warning:
  By using the -forceSSE flag, you are overriding program
  safeguards that monitor the stability of SSE
  instructions on your system. If you did not intend
  to do this, please restart the program without
  -forceSSE. If work units are not completing fully,
  then please discontinue use of the flag.
  
• Updated the .reg file I used originally to create the service. I could have avoided this by renaming the .exe, but I left it as FAH4.00-Console-Pre1.exe. Double-clicked the .reg file to add the updated settings to the registry.
• Since I didn't change the location of the files EMIII uses, it had no problem with the new name, but just for grins I added a new box, and was able to do so. I got the "Possible Path Error" dialogue, but I just clicked OK and EM added the new box.
• Of note, the log contains the following message. I don't know if this means I cannot go back to using the 3.24 client? On the other hand, unless I have trouble with the beta client, I won't switch back since I definitely like the faster processing times! To clarify, I assume the quicker processing time has everything to do with -forceSSE and little or nothing to do with the new client per se.
  
  [10:23:33] Old queue version detected... converting
  [10:23:33] Queue v3.24 detected... converting
  [10:23:33] Loaded queue successfully.
  
  [Edit] I found the following in another forum:
  Quote:
  It does change the format of queue.dat, so when you install it over the top of an existing installation, you can't run the V3.xx client again. A back up the installation directory solves that problem
  [/Edit]
  
• Having confirmed that the new client was happy, I closed the console and re-started the service.

David

Hi mormakil. Apologies for any confusion from my post but

1) So far (48 hours on 2 tbred systems) I have had no Beta problems/errors/defects. My folding production is unaffected (and no better or no worse).
2) I don't have any comparisons of 3DNow performance between old and Beta versions: my only 3DNow number was with the v4 beta and is significantly slower than SSE with the v4 beta (also unsurprisingly running 3DNow the CPU temp was 2 degrees lower).
3) I didn't do any comparisons with the v4 console and the old client: I forced the client upgrade on both boxen at the same time I installed the v4 console.

My summary related to the v4 beta console with 1.53 Gromacs client is no difference (speed or stability) to before (when I was running 3.25 beta console with 1.51 Gromacs client). Arguable beneficial feature is forceSSE may give better SSE control.

yeah I had the same on both my boxen. Good point though: it seems if you decide to go back then you would possibly have to abondon the current WU.

And that also explains your non-SSE previously: you needed the 3.25 beta to really get things going So coming back to the start of this thread: Zyzzyx you need a beta console to get that Palomino humming!

My EM problem is using EM to start Folding on the system I run EM. On the second system I just run a DOS window/console: EM on the first machine can monitor that 'FAH4.0 Console.exe' on the second system no problem as it is just watching the work directory.

[edit] corrected: my log is identical to haysdb even though I was definately 3.25 beta previously. Not sure that Zyzzyx can really get SSE to work on his Palomino if he already has a beta (3.25) console, or for that matter why haydb 3.25 beta console would not give SSE on his Palomino but v4 beta will [/edit]

dukla2000, so EM can monitor the folder (which I have confirmed) but is not able to start the client? That's something that can be worked around, by starting the F@H client separately from EM.

It's interesting that I have been running the 3.24 client. I would have downloaded it only 30 days ago. The 3.25 client added support for SSE on Athlon processors? So maybe I could have just updated to the 3.25 client and accomplished the same thing? Regardless, I am happy to have doubled the production from my one Athlon box simply by downloading a new client! Of course, time will tell whether it is "stable" running with SSE. As Mormakil has noted, it won't do any good for it to be twice as fast if some of the WU's get tossed because of instability.

David

OK, now wait a dadgum minute. Standford lists the latest "stable" client as 3.24, which is what I had. What's this about a 3.25 client?

[Edit] Nevermind. I see now that it was the "3.25 beta". That's why I didn't have it. [/Edit]

And how the heck was mormakil getting SSE with his 1800+ Palomino when I was getting only 3DNow with my 2000+ Palomino? I was using -forceasm.

[Edit2] I can't read. He was using 3.25, I was using 3.24.
[/Edit2]

David

Yup: if there is no 'FAH3Console.exe' in the directory EM starts itself but not the console. I just renamed the v4 exe to FAH3Console as a strange avoidance!
I may be wrong, so editted my previous post. Although I was definately running 3.25 beta I have '3.24' in the log after the upgrade as part of the queue conversion message. Not 100% sure of the SSE/Athlon/beta story, but IIRC 3.24 with -forceasm ONLY applies it to the first WU, all subsequent (in that console run) did not get any asm!
Damn - so that means you will be a threat to me before August 2004!

But I figure the SSE/Athlon stability problem is with folk who have overclocked (many Athlon users) and not genuinely verified stability. The tbreds in particular have a 'mega-overclock' reputation: well I have 2 with 'great' steppings, and fact is they overclock easily, but to get them stable while overclocked is complicated. If you are running stock speed, stock VCore I would doubt you will have pronlems. And like I say, so far I have seen none with the v4 beta.

That machine is running with stock settings. I may have tweaked the memory timings a bit, because the memory is Corsair low latency, but that's it. I was under the impression that Athlons were NOT good overclockers, so I never even thought about it.

I am a little bit depressed that I no longer have an excuse to upgrade to a 2400+ , but it's also true that it feels like I *did* upgrade, to a 2x faster cpu!

David

I'm quite sure about that and almost 100% sure of that being the cause haysdb was not getting SSE with his 1800 palomino and 3.24 client. Ok, this will help the team with his production. I hope Zyzzyx will be able of getting SSE with either the 3.25 or the 4.0 client and forceasm/forceSSE.

Another thing, all this processor upgrade fever , makes me think of giving my brother a new processor for Christmas . But he only have a 100 FSB Motherboard as far as I know. So my choices are limited to a 1300-1400 Duron. I've been looking and I think it has SSE, but anyone could confirm? Anyone know if it will be in general faster than an Athlon (non XP) 950? I couldn't find a direct comparison between them so far. Of course if it has SSE it will be faster folding

It will be a good gift for him and also for me, don't you think?

Clever boy mormakil.

Hmmm, who do I know that needs a computer for Christmas? That way they can pay for the electricity.

David

As far as I can see at AMD and in Duron model 8 Data Sheet the Durons do NOT have SSE. So will be 3DNow (same as 950) and only clock speed faster than the thunderbird Depending on the BIOS, you could try a low speed tbred (XP1600, XP1700) at 12.5*100 and maybe get the SSE working?

I get different data everywhere I look. Celeron vs Duron review at Tom's Hardware
And also it says it's 266 FSB and I see them in shops and this reviews as 200Mhz. I'm getting really upset with all this Athlon mess.

More interesting is what you said about putting an 1800 XP in a 100 Mhz FSB Mobo, is that possible? I don't know sincerly, I just supossed it was not possible.

I'll like to know that, but it's not important now, i've just look in the BIOS and it does allows 133 FSB, in fact memory is fixed at 133Mhz, so it makes me think which processor it has. Does the old 900 Athlons (non XP) have 133Mhz bus? I guess I'll have to ask DFI about this Mobo (it's a kt133) supporting an XP1800 Tbred

EDIT- On the other side DFI CPU's support list of this Mobo tells it only suppports 100Mhz processors, with 1.2 Ghz Duron maximun. I just don't understand, I'll have to ask them.

Depends on mobo/BIOS. It is an experiment/lottery! My FIC AZ11EA had no tbred support, but would run my XP2100 (spec 13*133) at 20*100 when fsb & multiplier set to Auto. I couldn't get it to even boot if I set the multiplier to something more sensible!

IIRC the thunderbird 900 was only available as 9*100 but it may overclock to something like 7*133 or more.

Then again, maybe you just have to get your brother the t-shirt for Christmas instead.

ps - I hate to say but wouldn't be the first time Tom's got something wrong. OK - happens to anyone, but their history is they don't correct mistakes even when they are told about them.

What about this thread It seems that the 1300 Duron Morgan have SSE and will be a quite good improvement, also it's really cheap 36€ (~42$)

About the T-Shirt I'm not sure it will help me increase my production

That thread is curious - if I read it correctly then only Calavaro (post on 12 May) claims to have a Duron running with SSE. Now I would figure the AMD links (my earlier post) should be the real truth and they don't mention SSE.

I have not followed the Duron evolution, but wonder if there is any link between the model 8 Duron and model 8 Athlon (tbred): are the Duron's actually the same core except failed quality control to be an Athlon? So some may have working SSE?

But I also remember when I got my first SSE from AMD reading you had to re-instal Win to get SSE enabled. Maybe that is why some Duron folk dont have SSE running?

Also methinks AMD have to pay Intel royalties for SSE: maybe that could explain why they don't mention it although some downgraded Athlons labelled as Duron have it?

You didn't read it exactly right. The thread it's about the changes made in the Gromacs core avoiding the use of SSE with Athlons/Durons because of instability problem. Next they enabled the use of SSE with Athlons with -forceasm flag, but not with Durons even if supported as it's deduced from Calavaro, ChelseaOilman and bruce posts.



And then as reply from a Pande Group user:



Then it seems that they will do this in few days an allow the use of SSE with a flag in the new Gromacs Core. But they don't reply about that. Calavaro posting the last post in this thread confirming he's using SSE with a 1300 Morgan indicate that maybe they just fix it and didn't say anything as long as people will know it was fixed when they tried.

Could you tell me where in the tech specs of Athlons, AMD mentions the inclusion of SSE?

This from the AMD site:

SSE Support in AMD Processors


SSE (Streaming SIMD Extensions) support is included in AMD's 3Dnow!™ Professional technology, which is supported in all Model 8 and Model 6 AMD Athlon™ processors (AMD Athlon XP, AMD Athlon MP, mobile AMD Athlon 4) and Model 7 AMD Duron™ processors.

The original Model 1 and Model 2 AMD Athlon processors and Model 3 AMD Duron processors do not support SSE as a whole, but do support certain SSE instructions, including software prefetch instructions and streaming store instructions (MOVNTQ). These instructions are included in Enhanced 3DNow! technology, which is supported by these processors.

To determine whether an installed processor supports SSE, use a utility such as AMD CPUID, which can be downloaded from the AMD Processor Utilities and Updates page. CPUID feature flags should be used to determine instruction set support, rather than deriving the processor's capabilities from vendor specifiers combined with CPUID model numbers. Further qualification of SSE should be done by checking for OS support. Even if SSE support is present in the processor, it may not be usable if the OS lacks support for the additional architected registers.

For additional information, refer to the technical document titled, "AMD Processor Recognition Application Note", which can be downloaded from the AMD Athlon™ XP Processor Tech Docs page. In particular, note:

Table 4 - "Summary of Standard and Extended Feature Bits"
Tables 6-11 - "Standard Feature Flag Descriptions"
Subsection titled, "Determining Instruction Set Support", under the chapter, "Identifying Supported Features".

Didn't look at the tables, etc. mentioned. Gotta shower and off to work.

I just figure all the other folk in that thread are assuming/repeating that Durons have SSE, rather than explicit knowledge. Maybe that is patronising, but I am not sure the Pande folk would claim to be experts on all the cpu flavours etc. Some of the other posters are unclear about what they have experienced to what they think they have experienced!

If you check any of the Data Sheets here then in the Overview (last para before heading 1.1) you can find the Athlon model 6 (palomino), Model 8 (tbred) and Model 10 (Barton) all talk about SSE. The Athlon model 4 (Thunderbird - I have an old copy) and Duron (model 3, model 7 & model 8 ) data sheets make no such mention.

[edit] and the quote mas92264 found on the AMD site is opposite of what I interpret from the Data sheets - OK - so even AMD is not sure [/edit]

Ok... I only skimmed all that, but did see my plight mentioned a few times.

Looks like I might have to change my tale a bit. A closer look at what SiSoft Sandra reports says I have a Thunderbird core, not a Palomino. I hadn't even heard of the Thunderbird core before. I'd thought that if it wasn't the newer T'Bred, it was a Palomino. Learn something every day. Perhaps that's why it does not have SSE, as reported by Sandra.


Regarding the F@H client: I have been running the 3.25beta with -forceasm, also tried the 4.00beta the other night with -forceSSE. Nothing.

Looks like I'm purty much screwed on this until/unless I grab a new cpu.

I admit I haven't read the whole thread yet so I might be missing something but:
1) I have a 2000+ Palomino and it DOES have SSE.
2) I have to force SSE with -forceasm option otherwise my frame times suck comared to p3 1 GHz
3) Every time I get a new WU I need to restart F@H because somehow it "forgets" about the -forceasm switch and does not use SSE on subsequent WUs

Hope that helps.

<OT>My passion to folding forced me to further overclock my Celeron 2.4. Initially I got it to 2.7 and run like that for 2 weeks. Yesterday I got it to 2.9 GHz Getting about 90 PPD now</OT>