Making an ordinary Athlon run as fast as a P4, per clock

GP2 · 2003-12-15, 21:37

Somewhere buried in the "M40 Again!!!!" thread in the Lounge, Dresdenboy stated that Athlons can do as much per clock as P4s can.

This would apply to the ordinary non-SSE2 Athlons, not just the new Athlon 64s.

I know nothing about assembler and don't own any Athlons, but I thought this deserves its own thread, rather than being buried in a long general thread.

Dresdenboy, can you try to provide more details? Any estimated implementation time frame or estimated speed increase?

flava · 2003-12-15, 22:45

Yes, good ideea! A speed increase for AMD cpu's would mean a lot. A very important percentage of us crunchers have Athlons and Durons.
It's been quite a few years since I wrote any assembler code ... but with some documentation about the Athlon maybe I could be of some help, who knows?

PageFault · 2003-12-16, 00:55

This old thread dealt with some issues concerning AMD performance. The content is more opteroon specific but still worth looking at.

Don't expect anything official for the tbred - it is a dead cpu and not worth any development time.

S00113 · 2003-12-16, 21:38

Quote:

Originally posted by PageFault
[Don't expect anything official for the tbred - it is a dead cpu and not worth any development time.

I wouldn't be so sure about that. Have a look at http://opteron.mersenneforum.org/png/machines.png

I don't think many AMD owners are running to the shops right now to buy Opterons, and definitively not Pentium IVs. In fact I bought myself a brand new XP 2800 Barton on a new mainboard with 1GB dual channel DDR RAM today. I don't need 64 bits yet, and the Thunderbird gives most bang for the buck. It will probably take years until we see more Athlon64s or P4's successors than the current Athlons.

PageFault · 2003-12-16, 22:02

What AMD owners do is irrelevant - AMD released the next gen already and within a year this will consist of the full lineup. Putting time on something at the end of its life cycle isn't a wise investment, indeed it never was a wise investment at all concerning AMD - not enough market share to make the result worthwhile. It seems that the code's current development reflects that already.

BTW the linked plot is irrelevant - it doesn't tell me anything about installed capacity, which is the number required to infer production by cpu type.

PrimeCruncher · 2003-12-16, 22:33

Quote:

Originally posted by PageFault
BTW the linked plot is irrelevant - it doesn't tell me anything about installed capacity, which is the number required to infer production by cpu type.

If you're talking about how many PCs PrimeNet is installed on, it most certainly does. I agree with GP2. Athlons as we know them may be at the end of their life cycle but will doubtless be a major GIMPS component for years to come. There are still 39 486s for crying out loud! Seeing that at the moment we have about 24,000 Athlons, comprising over 41% of the computers on PrimeNet, I think any optimizations are worth the effort.

S00113 · 2003-12-16, 22:59

Quote:

Originally posted by PageFault
What AMD owners do is irrelevant
[...]
BTW the linked plot is irrelevant - it doesn't tell me anything about installed capacity, which is the number required to infer production by cpu type.

I wouldn't say that what most GIMPSers do is irrelevant..

The chart shows very clearly that it takes a long time for a new CPU to become more common than it's predecessor. Why isn't that relevant? Of course the new platform is a bit faster, but an Opteron isn't that much faster. I still believe that old Athlons are going to do more work than Opterons for years to come.

I work at a very large university, and we have just go two Opteron machines in for testing. The mprime stress test fails with a round off error on one of them. The other one is working fine, but it's going to take at least a year or more until we start buying them in any quanta for servers. We are going to buy a lot of servers and workstations before that, and believe me: Computers grow old at universites.

Check where the top producers work if you don't think old computers are producing significant results!

PageFault · 2003-12-17, 01:26

My comment was towards AMD users.

Quote:

Originally posted by S00113
I wouldn't say that what most GIMPSers do is irrelevant..

The chart doesn't show anything - it is meaningless. What it shows is active accounts, not speed, not hours per day worked or results produced.

It shows many PIII's but without any indication of what they produced. Had it showed results, comparing those results to the most productive type would have showed the PIII's to be irrelevant. It showed my PIII untill I retired it - nine months for a 33M test was irrelevant compared to a machine doing the same in twenty days.

BTW the opteron is new - that is the reason to focus dev time on it - it will be relevant, as it is AMD's future for the coming years. What is relevant is having code ready for it once it is mass adopted. George could expend six months dev time on the obsolete cores, only to have them become irrelevant due to upgrades, and then have no code for the current crop. Not the way to manage a business.

I recall George posting somewhere that six to nine months were expended on the SSE2 code, back in 2001 when the willie was hardly relevant. That move paid off rather well, once the northwood proliferated, with the dividend being an expected one.

Quote:

The chart shows very clearly that it takes a long time for a new CPU to become more common than it's predecessor. Why isn't that relevant?

I work at a large university. I know how computers grow old, having recently disposed of the PI and PII machines. I didn't bother running prime95 on them as the expected production was irrelevant.

Quote:

Computers grow old at universites.

I can quickly check, as I happen to be on the top producing team. Our top guns tend to work in environments where P4's produce a large number of results. One reason for this is that new machines tend to be P4's in corporate environments. Our top home farmers tend to buy lots of P4's. curtisc recently ramped up massively and an analysis showed this was due to fast machines, i.e. P4's, he didn't add a bunch of 486's.

Quote:

Check where the top producers work if you don't think old computers are producing significant results!

QuintLeo · 2003-12-17, 01:42

Current 32-bit Athlon machines will be relevant for a few YEARS, at least. There are way too many of them already around, and they're going to BE around for a long time.

I am certain that, for the next year and likely the next 2-3 years that the *only* CPU class that will have more examples participating in GIMPS is the P4/P4-Xeon class.

There's still a lot of Thunderbird CPUs around, for that matter - which are the SAME as the Palomino and the Thoroughbred from a software standpoint - same archetecture, same cache, the *ONLY* differences that software can tell being the clock speed and a couple codes in certain registers on the CPU itself.

I'm not saying we should ignore the Opteron/Athlon64 - they are going to be very common in a year or two. But it's NOT wise to ignore the installed base of machines that are ALREADY producing for GIMPS - *especially* if the improvement for those machines can add 50% or more to their current performance. It is also goint to take at LEAST a year, more likely 2-3, before the Opteron/Athlon64 can match CURRENT 32-bit Athlon production.

"not the most current" does NOT equal "obsolete".

BTW - I worked for a largish company up to about a year ago (NCS, now the core part of Pearson Education). They were still buying P-III systems as of Febuary of 2002 (when the system I was using got upgraded from a P-II/266), and didn't start buying P4 systems 'till Febuary of 2003 (they buy on an annual cycle). They won't be buying ANY Athlon64 or Opteron systems in the forceeable future, as the company they buy systems from has no current plans to support those CPUs.

To put things further in perspective - there has been at least one *386* machine used in GIMPS in the last few months (I know, I did it) - and there are still quite a few K6 and PPGA P3/Celeron machines participating, more than 5 years after those CPUs were no longer being made. The company I used to work for still has a few 486 systems in daily use, and a LOT of P-II systems on desktops. Like most companies, they only buy new systems to replace an old system that dies, or when growth makes them add new systems from scratch for new employees.

> I know how computers grow old,
> having recently disposed of the PI
> and PII machines. I didn't bother
> running prime95 on them as the
> expected production was irrelevant.

Those systems would have been fine for factoring usage. It was a waste to NOT use them, for as long as they were working.

PrimeCruncher · 2003-12-17, 02:13

Quote:

Originally posted by PageFault
My comment was towards AMD users.

Quote:

Originally posted by S00113
I wouldn't say that what most GIMPSers do is irrelevant..

Newsflash: Athlons are most GIMPSers. 41% as I stated.

Quote:

The chart doesn't show anything - it is meaningless. What it shows is active accounts, not speed, not hours per day worked or results produced.

Then dump the chart and try logic. If 4 out of every 10 machines is an Athlon and we're doing about 9 teraflops of work, those PCs have to be contributing. Granted, a small percentage are no longer active and will expire, but the majority of them are still running and are not likely to be replaced in the near future since they aren't obsolete for most people yet.

Quote:

BTW the opteron is new - that is the reason to focus dev time on it - it will be relevant, as it is AMD's future for the coming years. What is relevant is having code ready for it once it is mass adopted. George could expend six months dev time on the obsolete cores, only to have them become irrelevant due to upgrades, and then have no code for the current crop. Not the way to manage a business.

True, but they are still very expensive, which is why adoption thus far has been slow. And George is already working on optimizing for Opterons.

Quote:

I recall George posting somewhere that six to nine months were expended on the SSE2 code, back in 2001 when the willie was hardly relevant. That move paid off rather well, once the northwood proliferated, with the dividend being an expected one.

The Athlon is already extremely relevent and, as QuintLeo said, will be for some time to come until the Athlon is completely phased out. Even after that you will still be able to buy them for a while and a lot of GIMPS enthusiasts will probably buy a few as cheap "close-out" products.

Quote:

I can quickly check, as I happen to be on the top producing team. Our top guns tend to work in environments where P4's produce a large number of results. One reason for this is that new machines tend to be P4's in corporate environments. Our top home farmers tend to buy lots of P4's. curtisc recently ramped up massively and an analysis showed this was due to fast machines, i.e. P4's, he didn't add a bunch of 486's.

Of course farmers tend to buy P4s - they have the most optimizations. But if we optomize the same for Athlons then I figure that:

Lower price for Athlon = farmers can afford more CPUs and buy Athlons = more work done

E_tron · 2003-12-17, 04:38

I think that we should make optimizations for the Barton core. The Barton core is the only Athlon XP currently manufactured by AMD. AMD disables some of its L2 cashe to make into lower end processors like the Thorton(256) and Applebred(128).

Barton core will or currently powers the:

Applebreds(duron processors with 128k L2 cashe enabled):
1.4, 1.6, and 1.8ghz

Thorton(the T-Bred imposter. AMD actually sold me one of these as a T-Bred

. It features 256 L2 cashe enabled):
2000+ @ 1.6ghz
2200+ @ 1.8ghz
2400+ @ 2ghz
2600+ @ 2.08ghz
2700+ @ 2.166ghz

Barton(the full featured version. It uses all of the 512K L2 cashe enabled)
2500+ @ 1.83ghz (my personal favorite)
2600+ @ 1.92ghz
2800+ @ 2.08ghz
3000+ @ 2.16ghz
3000+ 400FSB version @ 2.1ghz
3200+ @ 2.33ghz
3200+ 400FSB version @ 2.2ghz

2003-12-15, 21:37	#1
GP2 Sep 2003 5×11×47 Posts	Making an ordinary Athlon run as fast as a P4, per clock Somewhere buried in the "M40 Again!!!!" thread in the Lounge, Dresdenboy stated that Athlons can do as much per clock as P4s can. This would apply to the ordinary non-SSE2 Athlons, not just the new Athlon 64s. I know nothing about assembler and don't own any Athlons, but I thought this deserves its own thread, rather than being buried in a long general thread. Dresdenboy, can you try to provide more details? Any estimated implementation time frame or estimated speed increase?

2003-12-17, 01:42	#9
QuintLeo Oct 2002 Lost in the hills of Iowa 2⁶·7 Posts	Current 32-bit Athlon machines will be relevant for a few YEARS, at least. There are way too many of them already around, and they're going to BE around for a long time. I am certain that, for the next year and likely the next 2-3 years that the only CPU class that will have more examples participating in GIMPS is the P4/P4-Xeon class. There's still a lot of Thunderbird CPUs around, for that matter - which are the SAME as the Palomino and the Thoroughbred from a software standpoint - same archetecture, same cache, the ONLY differences that software can tell being the clock speed and a couple codes in certain registers on the CPU itself. I'm not saying we should ignore the Opteron/Athlon64 - they are going to be very common in a year or two. But it's NOT wise to ignore the installed base of machines that are ALREADY producing for GIMPS - especially if the improvement for those machines can add 50% or more to their current performance. It is also goint to take at LEAST a year, more likely 2-3, before the Opteron/Athlon64 can match CURRENT 32-bit Athlon production. "not the most current" does NOT equal "obsolete". BTW - I worked for a largish company up to about a year ago (NCS, now the core part of Pearson Education). They were still buying P-III systems as of Febuary of 2002 (when the system I was using got upgraded from a P-II/266), and didn't start buying P4 systems 'till Febuary of 2003 (they buy on an annual cycle). They won't be buying ANY Athlon64 or Opteron systems in the forceeable future, as the company they buy systems from has no current plans to support those CPUs. To put things further in perspective - there has been at least one 386 machine used in GIMPS in the last few months (I know, I did it) - and there are still quite a few K6 and PPGA P3/Celeron machines participating, more than 5 years after those CPUs were no longer being made. The company I used to work for still has a few 486 systems in daily use, and a LOT of P-II systems on desktops. Like most companies, they only buy new systems to replace an old system that dies, or when growth makes them add new systems from scratch for new employees. > I know how computers grow old, > having recently disposed of the PI > and PII machines. I didn't bother > running prime95 on them as the > expected production was irrelevant. Those systems would have been fine for factoring usage. It was a waste to NOT use them, for as long as they were working. Last fiddled with by QuintLeo on 2003-12-17 at 01:44

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Do normal adults give themselves an allowance? (...to fast or not to fast - there is no question!)	jasong	jasong	35	2016-12-11 00:57
benchmarks over-clock definition?	lfm	PrimeNet	4	2009-11-15 00:43
Clock Problems	R.D. Silverman	Puzzles	5	2006-12-13 00:29
The Clock Problem	davar55	Puzzles	9	2006-05-26 01:53
Alarm Clock	JuanTutors	Lounge	2	2004-06-21 09:39

2003-12-15, 22:45	#2
flava Feb 2003 2·59 Posts	Yes, good ideea! A speed increase for AMD cpu's would mean a lot. A very important percentage of us crunchers have Athlons and Durons. It's been quite a few years since I wrote any assembler code ... but with some documentation about the Athlon maybe I could be of some help, who knows?

2003-12-16, 00:55	#3
PageFault Aug 2002 Dawn of the Dead 5·47 Posts	This old thread dealt with some issues concerning AMD performance. The content is more opteroon specific but still worth looking at. Don't expect anything official for the tbred - it is a dead cpu and not worth any development time.

2003-12-16, 22:02	#5
PageFault Aug 2002 Dawn of the Dead 5×47 Posts	What AMD owners do is irrelevant - AMD released the next gen already and within a year this will consist of the full lineup. Putting time on something at the end of its life cycle isn't a wise investment, indeed it never was a wise investment at all concerning AMD - not enough market share to make the result worthwhile. It seems that the code's current development reflects that already. BTW the linked plot is irrelevant - it doesn't tell me anything about installed capacity, which is the number required to infer production by cpu type.

2003-12-17, 04:38	#11
E_tron Sep 2002 Austin, TX 3·11·17 Posts	I think that we should make optimizations for the Barton core. The Barton core is the only Athlon XP currently manufactured by AMD. AMD disables some of its L2 cashe to make into lower end processors like the Thorton(256) and Applebred(128). Barton core will or currently powers the: Applebreds(duron processors with 128k L2 cashe enabled): 1.4, 1.6, and 1.8ghz Thorton(the T-Bred imposter. AMD actually sold me one of these as a T-Bred . It features 256 L2 cashe enabled): 2000+ @ 1.6ghz 2200+ @ 1.8ghz 2400+ @ 2ghz 2600+ @ 2.08ghz 2700+ @ 2.166ghz Barton(the full featured version. It uses all of the 512K L2 cashe enabled) 2500+ @ 1.83ghz (my personal favorite) 2600+ @ 1.92ghz 2800+ @ 2.08ghz 3000+ @ 2.16ghz 3000+ 400FSB version @ 2.1ghz 3200+ @ 2.33ghz 3200+ 400FSB version @ 2.2ghz