AMD Releases Barton: Athlon 3000+

Harle writes "Today AMD has introduced a new version of the Athlon, codenamed "Barton," that features twice as much L2 cache as previous chips. Along with the increase in L2 cache comes an increase in the Athlon's performance rating -- specifically the new 2.17 GHz chip is rated at 3000+. The clockrate is actually slighly lower than the Athlon XP 2800+'s 2.25 GHz speed, so the question becomes "Does the cache improve performance enough to counter the loss in clockspeed?" For the most part, the answer seems to be "yes," however, it doesn't unilaterally stand up to the 3.06 GHz Pentium 4. With the recent delay of the Athlon 64 to September, this is AMD's top desktop chip for some time to come. The reviews are starting to pop up at Ace's Hardware and Extremetech." There's also reviews on The Tech Report, SimHQ, HotHardware, EarthV, in Norwegian on Hardware.no, and last but not least AMD's press release. I'm sure there's many many more links, but I'm tired of pasting them all in here, so post 'em below. *grin*

Comments

Anandtech link

[Anonymous Coward]

so post 'em below..

OK Then... Anandtech link

Re:Anandtech link

[zerocool^]

I can't wait until

Tom's hardware comes out with

A review so that I can swing

By that website and check out all

the new ads they have and see

How many pages they've broken the

review up into. I bet it's like

12.

640 kb on die cache

[fobef]

"This exclusive 512 KB L2-cache works together with the 128 KB L1-cache (64 KB data, 64 KB instruction) to form one impressive 640 KB on-die cache."

Am I the only one who can see Bill Gates drooling over this?

Aussie Athlon?

[CountZero007]

Yay, a processor code-named after our first Prime Minister :)

I'm never buying a 'Keating' or 'Howard' though...

Comparing to the P4 3.06...

[Junior J. Junior III]

however, it doesn't unilaterally stand up to the 3.06 GHz Pentium 4.

Well, of course not... if it did, they'd be calling it the Athlon XP 3060, wouldn't they? ;)

Out of curiosity

[Timesprout]

From the article on Ace Harware

This exclusive 512 KB L2-cache works together with the 128 KB L1-cache (64 KB data, 64 KB instruction) to form one impressive 640 KB on-die cache. According to AMD, the extra 256 KB cache boosts, an 2170 MHz Athlon XP from a 2700+ level to a 3000+ one.

If this is the case why do AMD, and Intel for that matter not put ever larger amounts of cache on their existing chips to achieve better performance ? Does the cost implications completely prohibit this or do the performance benefits tail off too quickly. SUN seem to able to achieve impressive performance with lower far lower Mhz (I know its different architecture) but I get the impression the large amounts of cache (2-4 MB) they use contributes significantly to performance.

Re:Out of curiosity

[AeternitasXIII]

Does the cost implications completely prohibit this or do the performance benefits tail off too quickly. SUN seem to able to achieve impressive performance with lower far lower Mhz (I know its different architecture) but I get the impression the large amounts of cache (2-4 MB) they use contributes significantly to performance.

Yes, cache is terribly expensive to place on chips in large amounts. It tends to be much harder to shrink than the rest of the transistors on the chip, and the design work necessary to scale the cache to meet the ever shrinking die size is complex and harsh. Overall, with consumer chips that need to be under a certain price threshold to be purchased, Intel and AMD have both found its far cheaper to keep increasing clock speed while decreasing die size than it is to increase cache.

My guess is that this latest move by AMD is an update to that mentality. It proceeds along their realization that they might be unable to compete solely on the grounds of clockspeed. However, with the decreasing performance returns for clockspeed increases, this is less of an issue for AMD than one would think. This new core seems to indicate its becoming cheaper for their engineers to spend more time on chip design as well as use the limited die space for cache rather than other components.

Re:Out of curiosity

[foobar3149]

The yield drops roughly quadratically with area of the chip. So, if the L2 cache occupies 50% of the chip and we want to double the cache size, the yield would drop to roughly 44% of the original yield. To make up for this the price would have to more than double and I do not believe that most people buying desktops are willing to accept that.

Re:Hurry Up!

[FyRE666]

I don't see how not selling a ridiculously overpriced CPU (the fastest P4) is going to damage them at all.

The vast majority of people don't need the P4 3GHZ, and anyone with any sense would find the current "sweet spot" in the CPU vs price tables. AMD have Intel beat all the way up to the AMD max speed, so how is Intel "kicking there ars"?

You could also say that Ferrari are kicking Ford's ass, since they make a faster car - it just doesn't work like that...

Benchmarking...

[ponos]

Well it is quite clear that the Athlon
architecture is at the end of its useful life.
However,the performance difference is somewhat
"exaggerated" in favour of the P4. Most of the
"content creation" applications and games
are SSE2 enhanced while, on the other hand
3dnow is propably less popular. The hard fact,
of course, is that P4 needs less time in these
applications so it is faster (whatever the
reason).

However, for general purpose usage, I firmly
believe that the Athlon is faster, mostly
because everyday applications do not need
huge memory bandwidth and cannot be made to
run with SSE.

I'm thinking that someone should start doing
some "Open Source" benchmarks where the source
is available. A good idea would be to run
a set of:
a) Kernel compile (or gcc compile or something
like that) and perhaps even "make check"
gdb or gcc or some other application (libc!).
b) MP3 compression with lame
c) Video compression with xvid or ffmpeg
d) Linpack/POVray for fpu
e) Ecasound/LADSPA for sound processing
f) Maybe a perl/high-level bechmark for some
standard system tasks.
g) Cachegrind some of the above (have a look
at valgrind/cachegrind!!)

Anyway, if someone has anything above XP 2600+
let's gather some results.

P.

Re:Mixing the cards... no wait: cores.

[Luminous Coward]

Just remember to do your research, and you'll be fine.

One can read the AMD Processor Recognition document which explains how to extract the information from the Ordering Part Number (OPN).

AMD Processor Ordering Part Number (OPN) Breakdown

AXDA 2700 D K V 3 D
^^^^ ^^^^ ^ ^ ^ ^ ^
-1-- -2-- 3 4 5 6 7

(1) Processor Core Architecture/Brand Name
(2) Model Number
(3) Package Type
(4) Operating Voltage(Nominal Core Voltage)
(5) Maximum Die Temperature
(6) Level 2 Cache Size
(7) Maximum System-Bus (Front-Side-Bus) Speed

(1) Processor Core Architecture/Brand Name

(only Thoroughbred and Barton cores are 0.13 m)

AXDA ----- AMD Athlon XP -- 0.13 m
AX ------- AMD Athlon XP -- 0.18 m
AMSN ----- AMD Athlon MP -- 0.13 m
AMP/AHX -- AMD Athlon MP -- 0.18 m
K7/A ----- AMD Athlon ----- 0.18 m

(6) Level 2 Cache Size

1 -- 64 KB
2 -- 128 KB
3 -- 256 KB
4 -- 512 KB (only Barton cores have a 512 KB L2 cache)

(7) Maximum System-Bus Speed

B -- 200 MHz
C -- 266 MHz
D -- 333 MHz