Thoroughbred 2600XP Quicker Than Barton 2600XP ?

[Dominic Shields]

According to the Sisoft Sandra 2004 CPU benchmark, my Thoroughbred
2600 on a Soltek SL-KT400C board with 512 MB of Twinmos PC 2700 RAM
gives a pretty identical result to their reference 2600XP :
7946 MIPS, 3276 MFLOPS.

I built a machine for a friend from parts ordered from Ebuyer using an
Asus A7V8X-X KT400 board and 512 MB of Ebuyer PC 2700 RAM, I chose the
Barton 2600 thinking that it would be faster across the board. However
I was surprised to find that the Barton gave a repeatable result about
10% slower on the Sandra benchmark.

Is it probable that this is an unrealistic result or have I
misconfigured the CPU? I didn't have time to run more comprehensive
tests.

 

[GSV Three Minds in a Can]

According to the Sisoft Sandra 2004 CPU benchmark, my Thoroughbred
2600 on a Soltek SL-KT400C board with 512 MB of Twinmos PC 2700 RAM
gives a pretty identical result to their reference 2600XP :
7946 MIPS, 3276 MFLOPS.

I built a machine for a friend from parts ordered from Ebuyer using an
Asus A7V8X-X KT400 board and 512 MB of Ebuyer PC 2700 RAM, I chose the
Barton 2600 thinking that it would be faster across the board. However
I was surprised to find that the Barton gave a repeatable result about
10% slower on the Sandra benchmark.

Is it probable that this is an unrealistic result or have I
misconfigured the CPU? I didn't have time to run more comprehensive
tests.

This is not impossible - a Barton has 2x the cache size, but a lower
clock frequency. Unless the benchmark makes good use of the extra L2
cache, then the frequency decrease will be what you see.

Bartons come as 2500+ (1833Mhz), 2600+ (1917Mhz), 2800+ (2083Mhz), and
3000+ (2167 Mhz). [in 166/333 Mhz FSB speeds].

The T'bred XP2600+ runs at 2083Mhz.

 

[Tony Hill]

According to the Sisoft Sandra 2004 CPU benchmark, my Thoroughbred
2600 on a Soltek SL-KT400C board with 512 MB of Twinmos PC 2700 RAM
gives a pretty identical result to their reference 2600XP :
7946 MIPS, 3276 MFLOPS.

I built a machine for a friend from parts ordered from Ebuyer using an
Asus A7V8X-X KT400 board and 512 MB of Ebuyer PC 2700 RAM, I chose the
Barton 2600 thinking that it would be faster across the board. However
I was surprised to find that the Barton gave a repeatable result about
10% slower on the Sandra benchmark.

Is it probable that this is an unrealistic result or have I
misconfigured the CPU? I didn't have time to run more comprehensive
tests.

This is to be expected. Sandra's CPU benchmark is a pretty much
useless test, so don't read anything from it. The 'Barton' AthlonXP
2600+ runs at a slower clock speed than the 'Thoroughbred' AthlonXP
2600+, but it usually makes up for the clock speed difference with a
larger L2 cache. However this Sandra test fits entirely into the L1
cache, so it completely ignores the difference in L2. Therefore the
slightly lower clock speed of the Barton results in a lower score.

As mentioned though, this is a totally useless test, really doesn't
tell you anything useful about CPU performance because it's really
quite rare for an application to fit entirely in the L1 cache. Those
few apps that do fit entirely into the L1 cache are always total
oddball apps anyway, so Sandra isn't really useful for comparing
those.

In any case, overall the 'Barton' AthlonXP 2600+ and the
'Thoroughbred' AthlonXP 2600+ should offer similar performance on most
real-world applications. One chip will be slightly faster in some
apps, the other will be slightly faster in other apps, but overall
they'll be pretty darn close, hence the reason why they have the same
model number.