Real World Comparisons: AMD 3200 -vs- Intel 3.2. Your thoughts, experiences....

[gaffo]

Actually, if I remember my own history correctly, the K6-2 worked fine
without the CXT tweak but performance went up by a significant bit
with it.

one of the k-6's had a problem with windoze-98First
Edition.................something to due with the timing. The chip ran
too fast for windoze to work correctly. M$ offered a patch to fix it
within a few weeks. This was YEARS ago and I've forgotten all the details.


try google.




--



"I think in this case international law
stood in the way of doing the right thing (invading Iraq)."
- Richard Perle


"He (Saddam Hussein) has not developed any significant capability with
respect to weapons of mass destruction. He is unable to project
conventional power against his neighbours."
- Colin Powell February 24 2001


"We have been successful for the last ten years in keeping
him from developing those weapons and we will continue to be successful."

"He threatens not the United States."

"But I also thought that we had pretty
much removed his stings and frankly for ten years we really have."

'But what is interesting is that with the regime that has been in place
for the past ten years, I think a pretty good job has been done of
keeping him from breaking out and suddenly showing up one day and saying
"look what I got." He hasn't been able to do that.'
- Colin Powell February 26 2001

 

[RusH]

Not in my neck of the woods at least. The cheapest Athlon64 I can
find is about $350 CDN (~$250 US) for the 3000+ (from a reliable
vendor, some bargain-basement places sell it for a bit less).
Motherboards for the Athlon64 start at about $175 CDN (~$135 US).
For comparison, I can get a P4 2.8C GHz processor for $275 (~$200
US) and a motherboard for $120 (~$95 US).

damn, i can get P4 for 280$ :/ , A64 for 340$
motherboards are in the same price range (I'm not talking about
ECS/AsRock crap), 20% more for a significant quality leap

The performance for the
applications the original poster listed would be very similar
between these chips as they were applications that the P4 often
does quite well in (high-bandwidth use and lots of SSE2
optimizations).

true for photoshop, I'm not sure about the rest.

For anyone reading here, don't bother putting too much value in
peoples comments about hard drives. Go to www.storagereview.com,

yes if your planning to buy a server harddrive
I like to belive in HDTach myselfe :

http://www.zacisze.zgora.pl/~bart/index2.html

Again, go to Storage Review for the real-deal here. Fire up their
"Performance Database" and compare this Samsung drive (they tested
the 160GB model) with some drives from Maxtor, Seagate, WD and
Hitachi. The Samsung drive does fine, but it pretty much middle of
the pack. There's not a huge discrepancy between the various
companies.

SamSung: SpinPoint 120 GB SP1203N (7200) ATA/133 (8MB cache)
http://www.zacisze.zgora.pl/~bart/images/sam_hdtach.jpg

SeaGate: Barracuda 120 GB ST3120022A (7200) ATA/100 (2MB cache)
http://www.zacisze.zgora.pl/~bart/images/sea_hdtach.jpg

Maxtor DiamondMax +9 120 GB (8MB) 6Y120P0 ATA/133
http://www.zacisze.zgora.pl/~bart/images/max1_hdtach2.jpg

From the prices I see it's only an extra $10-$15 for SATA over
ATA133 drives

WD Caviar 120GB, 7200, ATA/100 (WD1200JB) 8MB cache 116$
WD Caviar 120GB, 7200, SerialATA (WD1200JD) 8MB cache 140$
Europe prices sucks I guess.

which is less than I would pay for round cables.

Agreed with round cable price. I sell computer cables/plugs/other
little useless computer objects myself and know how hilarious prices
can be (for example $40 for a 1 meter long shielded round ATA133, not
to mention SCSI and Cisco cables ..shrug)

The AthlonXP 3200 will be slower than an P4 2.8C for most of the
applications listed by the original poster. The Athlon64 3200+ is
a fair bit more expensive, probably an extra $200 - $300 US on the
whole system price. Considering the entire system would work out
to about $700 US for the P4, an extra $200-$300 is a lot.

hmmmm :
http://www20.tomshardware.com/cpu/20030513/athlon_xp-15.html
there is no P4 2.8C in the comparison, but it should place somewhere
between XP 2700 and 3000

Plus, as I mentioned, I'm not satisfied with the current crop of
motherboards, almost all are based on VIA chipsets
I've used too
many VIA chipsets in the past with their drivers that just never
quite work right, even when there are no obvious problems that can
be pinpointed. See that recent thread titled "Why does everyone
hate VIA", or words to that effect, for most info.

and I'm affraid of flights

Pozdrawiam.

 

[Nadeem]

George Macdonald wrote:
Oh and the Maxtor diags don't work

with nVidia chipsets -

Damn!!!!!!!!!! Thats why things weren't working on my brother's pc.!!

--

Nadeem M Nayeck [ m n n a y e c k @ i n t n e t . m u ]
______ ______ .
.:_\_ . \\_ . \_::.
. .::./ ./ // ./__/.:::. . Registered LU #290695
:_<_____/<____ >_:. - Where Tahoma looks like sprayed shit...
. \/ .

 

[The little lost angel]

one of the k-6's had a problem with windoze-98First
Edition.................something to due with the timing. The chip ran
too fast for windoze to work correctly. M$ offered a patch to fix it
within a few weeks. This was YEARS ago and I've forgotten all the details.

It wasn't one of the K-6. It was basically any Cyrix or K6 above
350Mhz. They ran certain instructions twice as fast as the equivalent
Pentium 2 and caused the timing loop M$ used to finish in effectively
0 seconds. So X/0 sec = error.

It happened with a P2/3 starting at around 700Mhz.

One of the method for getting around it was to downclock the
processor, apply the patch and clock back up.



--
L.Angel: I'm looking for web design work.
If you need basic to med complexity webpages at affordable rates, email me
Standard HTML, SHTML, MySQL + PHP or ASP, Javascript.
If you really want, FrontPage & DreamWeaver too.
But keep in mind you pay extra bandwidth for their bloated code

 

[Keith R. Williams]

It wasn't one of the K-6. It was basically any Cyrix or K6 above
350Mhz. They ran certain instructions twice as fast as the equivalent
Pentium 2 and caused the timing loop M$ used to finish in effectively
0 seconds. So X/0 sec = error.

Close. The Pentium had a two-cycle NOP instruction. AMD thought this
wasteful so cut both cycles out (no need to dispatch a NOP, right? ;-).
Oops, timing loops relying on NOP broke.

It happened with a P2/3 starting at around 700Mhz.

Yep, that's where the two-cycle NOP failed.

One of the method for getting around it was to downclock the
processor, apply the patch and clock back up.

True. It wasn't a huge deal.

 

[Robert Redelmeier]

Close. The Pentium had a two-cycle NOP instruction. AMD thought this
wasteful so cut both cycles out (no need to dispatch a NOP, right? ;-).
Oops, timing loops relying on NOP broke.

I thought it was a quicker `loop` instruction vs `dec ecx / jnz`

-- Robert

 

[The little lost angel]

Close. The Pentium had a two-cycle NOP instruction. AMD thought this
wasteful so cut both cycles out (no need to dispatch a NOP, right? ;-).
Oops, timing loops relying on NOP broke.

Hmm, why would anybody use NOP for a timing loop??
I mean, as a noob programmer back in high school (or the equivalent of
age 15), I used these things too. Except in my loops generally were a
couple of calculations typically used in the program. Largely because
an empty loop tend to finish in 0 time AFAIK pPp

--
L.Angel: I'm looking for web design work.
If you need basic to med complexity webpages at affordable rates, email me
Standard HTML, SHTML, MySQL + PHP or ASP, Javascript.
If you really want, FrontPage & DreamWeaver too.
But keep in mind you pay extra bandwidth for their bloated code

 

[Tony Hill]

yes if your planning to buy a server harddrive
I like to belive in HDTach myselfe :

http://www.zacisze.zgora.pl/~bart/index2.html

The guys are Storage Review REALLY know what the hell they are talking
about, MUCH more so than any other review I've seen. They have done
extensive testing of drives for desktop and server use, and they have
also done extensive testing of drive benchmarking utilities.

One of the things they've come up with (complete with extensive
evidence) is that sequential read speed (what HDTach measures) is NOT
the most important factor in application performance. While it does
factor it, there are MANY other factors that affect performance as
well. They do still list sequential read speed (using Winbench, which
produces very similar results to HDTach), but they also do more
extensive testing.

FWIW here is their explanation of what they've found to influence
performance of hard drives:

http://storagereview.com/guide2000/ref/hdd/perf/perf/index.html

WD Caviar 120GB, 7200, ATA/100 (WD1200JB) 8MB cache 116$
WD Caviar 120GB, 7200, SerialATA (WD1200JD) 8MB cache 140$
Europe prices sucks I guess.

Could be, I'm seeing the WD 120GB ATA100 8MB for $131 Canadian
(~$100US) and the WD1200JD 120GB SATA 8MB for $152 (~$115 US). For
the Seagate 120MB/8MB cache the difference is $17 CDN (~$13 US), while
for Maxtor the difference is $12 CDN (~$8 US).

hmmmm :
http://www20.tomshardware.com/cpu/20030513/athlon_xp-15.html
there is no P4 2.8C in the comparison, but it should place somewhere
between XP 2700 and 3000

Check the latest round of Prescott tests, most of them have a 2.8C GHz
P4 (Northwood) processor and AthlonXP 3200+ thrown into the mix
alongside the new Prescott P4s. In particular, look at the
applications that the original poster asked about, ie Photoshop,
Illustrator, video editing and encoding. These are all areas that the
P4 tends to do well in, and while the AthlonXP is a great chip, for
the specific applications mentioned by the original poster, the P4 is
a bit better.

 

[Tony Hill]

I thought it was a quicker `loop` instruction vs `dec ecx / jnz`

My memory (admittedly a bit fuzzy here) is that Cyrix had the really
fast NOP while AMD has the fast 'loop' instructions.

Either way, both ended up breaking some brain-dead timing loops. This
one really can't be blamed on AMD or Cyrix though, since I've seen
DOZENS of brain-dead timing loops break on all sorts of processors,
including Intel chips. Sufficient stupid software can cause problems
with even the best hardware.

 

[Tony Hill]

one of the k-6's had a problem with windoze-98First
Edition.................something to due with the timing. The chip ran
too fast for windoze to work correctly. M$ offered a patch to fix it
within a few weeks. This was YEARS ago and I've forgotten all the details.

It was Win95, and it affected K6-2 chips that were clocked to 333MHz
or more (overclocked K6 chips could have similar problems as well,
though not many K6s overclocked very well).

The problem was a simple brain-dead timing loop that eventually broke
on all processors, it just affected the K6-2 first because the loop
ran a lot faster on those processors than on Intel processors.

 

[Keith R. Williams]

Hmm, why would anybody use NOP for a timing loop??

We *are* talking M$ here.

I mean, as a noob programmer back in high school (or the equivalent of
age 15), I used these things too. Except in my loops generally were a
couple of calculations typically used in the program. Largely because
an empty loop tend to finish in 0 time AFAIK pPp

You're not so dumb! M$ was.

 

[George Macdonald]

George Macdonald wrote:
Oh and the Maxtor diags don't work

Damn!!!!!!!!!! Thats why things weren't working on my brother's pc.!!

Hmmm, you mean the diags didn't work - yes? IIRC the diag software just
goes to a blank screen and then back to a DOS prompt but the drives work OK
with the nVidia chipset under the usual OSs. I spent ~1/2hour scouring the
Maxtor Tech Supp site before I stumbled on a FAQ which confesses about the
diags problem. Oops, they finally moved the confession up front:
http://www.maxtor.com/en/support/downloads/powermax.htm - still says
"slated for release soon". Makes you wonder what they understand by
"soon"... ummm, 'we can't figure it out'?

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

 

[Nadeem]

Hmmm, you mean the diags didn't work - yes? IIRC the diag software just
goes to a blank screen and then back to a DOS prompt but the drives work OK
with the nVidia chipset under the usual OSs. I spent ~1/2hour scouring the
Maxtor Tech Supp site before I stumbled on a FAQ which confesses about the
diags problem. Oops, they finally moved the confession up front:
http://www.maxtor.com/en/support/downloads/powermax.htm - still says
"slated for release soon". Makes you wonder what they understand by
"soon"... ummm, 'we can't figure it out'?

Rgds, George Macdonald

"Just because they're paranoid doesn't mean you're not psychotic" - Who, me??

yeah the diag software just wouldnt work. All the time I thought my
disks or the floppy disk drive was bad...

--

Nadeem M Nayeck [ m n n a y e c k @ i n t n e t . m u ]
______ ______ .
.:_\_ . \\_ . \_::.
. .::./ ./ // ./__/.:::. . Registered LU #290695
:_<_____/<____ >_:. - Where Tahoma looks like sprayed shit...
. \/ .

 

[Ted Grevers]

All-

Thanks for the great discussions around this. I think I will move
forward with the Intel 3.2GHz setup. This topic took me beyond things
that I traditionally consider in buying or even building a system.
Disks, video compatibility and other BIOS limitations. I think what
finally convinced me was the application testing that someone posted.
When looking at this, the numbers showed the 3.0GHz chip's abilities.
I figure that even if they are minute increases, the Intel 3.2's
performance will still exceed the AMDs in the applications that I
currently need this new machine to execute.

Thanks so much to everyone who participated in the discussion, and to
those who emailed me directly.

-Ted

 

[chrisv]

1 x 1 Gig of brand mem, no point in getting dual channel for a 20%
added cost

Huh? No point in doubling your bandwidth?

 

[Nate Edel]

Huh? No point in doubling your bandwidth?

Moreover, 1x 1gb DIMM is still usually _more_ expensive than 2x 512mb,
unless someone is badly overpaying for overclocker 512mb DIMMs, or someone's
found an especially cheap vendor for the 1gb models.

Crucial is neither the best vendor or the cheapest, but they're usually one
of the better and cheaper vendors, and as of today:

PC3200 - 1GB $429.99 (non-registered, non-ECC)
PC3200 - 1GB $439.99 (registered, ECC)
PC3200 - 512MB $91.99 (non-registered, non-ECC)
PC3200 - 512MB $139.99 (registered, ECC)
PC2700 - 1GB $299.99 (non-registered, non-ECC)
PC2700 - 1GB $349.99 (registered, ECC)
PC2700 - 512MB $81.99 (non-registered, non-ECC)
PC2700 - 512MB $127.99 (registered, ECC)
PC2100 - 1GB $275.99 (non-registered, non-ECC)
PC2100 - 1GB $289.99 (registered, ECC)
PC2100 - 512MB $77.99 (non-registered, non-ECC)
PC2100 - 512MB $121.99 (registered, ECC)

Doesn't sound like a 20% added cost to me. I'm not going to go price
overpriced overclocker RAM, but my guess is even if you buy into the "must
buy factory matched pairs" religion, matched pairs of 512mb will at most run
about the same as a single 1gb stick in the same speed/ECC/registeredness
from the same vendor.