AMD Nomenclature?

[Adjacent]

OK, I can't find it on AMD's website, but someone has to know. I understand
that the higher numbers i.e. 2800, 3000, 3200, etc. are increasingly better,
but what does it mean. I see that most of these have the same clockspeeds,
but the higher numbers (2800,3200...) cost more, so they must be better,
right? Why? What makes them better if not clock speed?

Thanks,
Ad

 

[Lachoneus]

OK, I can't find it on AMD's website, but someone has to know. I understand

that the higher numbers i.e. 2800, 3000, 3200, etc. are increasingly better,
but what does it mean. I see that most of these have the same clockspeeds,
but the higher numbers (2800,3200...) cost more, so they must be better,
right? Why? What makes them better if not clock speed?

In general, higher numbers mean faster CPUs, but the numbers are not
influenced by clock speed alone--FSB speed, L2 cache size, memory
bandwidth (single vs. dual channel), and marketing expedience are also
taken into account.

For example, the Clawhammer 3200+ runs at 2.0GHz and has a 1MB L2 cache.
The Newcastle 3000+ runs at the same 2.0GHz, but has just 512KB L2
cache. Both of the above chips have a single memory channel (socket
754). The Winchester 3200+ runs at 2.0GHz and has 512KB L2 cache, but
has a dual channel memory interface (socket 939), earning it an extra
200 computron bonus over the Newcastle.

Bottom line? It's hard to quantify a processor's performance with just
one number. AMD would have been sensible to just abandon the
performance number altogether just like they did for the Opteron. Or
maybe not.

 

[Adjacent]

In general, higher numbers mean faster CPUs, but the numbers are not
influenced by clock speed alone--FSB speed, L2 cache size, memory
bandwidth (single vs. dual channel), and marketing expedience are also
taken into account.

For example, the Clawhammer 3200+ runs at 2.0GHz and has a 1MB L2 cache.
The Newcastle 3000+ runs at the same 2.0GHz, but has just 512KB L2 cache.
Both of the above chips have a single memory channel (socket 754). The
Winchester 3200+ runs at 2.0GHz and has 512KB L2 cache, but has a dual
channel memory interface (socket 939), earning it an extra 200 computron
bonus over the Newcastle.

Bottom line? It's hard to quantify a processor's performance with just
one number. AMD would have been sensible to just abandon the performance
number altogether just like they did for the Opteron. Or maybe not.

Thanks for the clarification, inasmuch as that is possible .

The machine I intend put it in is my "workhorse" system. I mostly use it to
record and process video and audio, burn DVDs & CDs and other time and
somewhat processor intensive jobs. According to your explanation and since I
want to stick with 754 (& 3400+), Obviously the right choice is the
Clawhammer w/the 1 mb L2 cache. Any recommendation on motherboards? I've
been checking out Gigabyte GA-K8NS PRO.

Thanks again,
Ad

 

[General Schvantzkoph]

Thanks for the clarification, inasmuch as that is possible .

The machine I intend put it in is my "workhorse" system. I mostly use it to
record and process video and audio, burn DVDs & CDs and other time and
somewhat processor intensive jobs. According to your explanation and since I
want to stick with 754 (& 3400+), Obviously the right choice is the
Clawhammer w/the 1 mb L2 cache. Any recommendation on motherboards? I've
been checking out Gigabyte GA-K8NS PRO.

Thanks again,
Ad

Why do you want to stick with 754? The 939s support more memory than the
754s because they have two separate memory buses. I'd go with a 3500+ 939
rather than a 3400+ 754.

 

[Cuzman]

" I understand that the higher numbers i.e. 2800, 3000, 3200, etc. are
increasingly better, but what does it mean. I see that most of these have
the same clockspeeds, but the higher numbers (2800,3200...) cost more, so
they must be better, right? Why? What makes them better if not clock speed?
"


AMD's early Athlon, based on the Thunderbird core, didn't have a PR Rating
such as the 2800+, 3000+ or 3200+ that you refer to. The Thunderbird's
marketing relied on its clock speed, and buyers tended to compare it to an
Intel processor of similar clock speed.

When AMD moved to the Athlon XP, firstly with the Palomino core, the new
architecture had significantly begun to out-perform Intel per clock cycle.
If they were to continue marketing their processors by clock speed alone,
they would only be selling themselves short. Therefore, they introduced the
PR Rating, indicating the relative performance to their early Athlon
Thunderbird. However, these PR ratings generally bear more relevance to the
competition from Intel than to the Thunderbird.

AMD processors with the same clock speed may have differing PR ratings
because of a number of factors. Core architechture, L2 cache, FSB and
socket type are the major factors in their performance.

http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_9485_9487^10248,00.html

 

[Adjacent]

Why do you want to stick with 754? The 939s support more memory than the
754s because they have two separate memory buses. I'd go with a 3500+ 939
rather than a 3400+ 754.

Cost concerns, but after reading your post, I see I may be able to swing
3500+ for about $50 more. However, I'd like to know more regarding the
differences between the cores at that level (i.e., the 3500+ cpus on Newegg
have the Newcastle and the Winchester cores, with the Winchester costing
slightly more). I checked at amd.com, but a search of their website turns up
unrelated items. Do you know of a site that can fill me in on the details?

Thanks,
Ad

 

[Wes Newell]

Cost concerns, but after reading your post, I see I may be able to swing
3500+ for about $50 more. However, I'd like to know more regarding the
differences between the cores at that level (i.e., the 3500+ cpus on Newegg
have the Newcastle and the Winchester cores, with the Winchester costing
slightly more). I checked at amd.com, but a search of their website turns up
unrelated items. Do you know of a site that can fill me in on the details?

Don't take the model numbers of the A64 for much. They aren't a very good
indicator of real performance. I'd suggest you do a little more reasearch
before speding your money. For instance, a socket 754 Sempron 3100+
overclocks easily over the speed of the stock speed of the 4000+, and gets
very close to it in performance. Now consider a MB and 4000+ CPU will
cost you about $850 while the Sempron system would cost less than $200.
And the 3400+ 754 beats the 3500+ 939 in a lot of benchmarks. Don't be too
concerned abpout cache size and dual channel. It's the clock speed that
produces the real performance increase in most apps. Here's one reference
link.

http://www.xbitlabs.com/articles/cpu/print/sempron-3100-oc.html

 

[General Schvantzkoph]

Don't take the model numbers of the A64 for much. They aren't a very good
indicator of real performance. I'd suggest you do a little more reasearch
before speding your money. For instance, a socket 754 Sempron 3100+
overclocks easily over the speed of the stock speed of the 4000+, and gets
very close to it in performance. Now consider a MB and 4000+ CPU will
cost you about $850 while the Sempron system would cost less than $200.
And the 3400+ 754 beats the 3500+ 939 in a lot of benchmarks. Don't be too
concerned abpout cache size and dual channel. It's the clock speed that
produces the real performance increase in most apps. Here's one reference
link.

http://www.xbitlabs.com/articles/cpu/print/sempron-3100-oc.html

I beg to differ, it's total RAM size that matters most. The clock speed
differences are trivial, generally about 10% per speed step. Human beings
don't notice tiny differences like 10 or 20%. However having enough RAM so
that your applications all stay in memory rather than having to be paged
in from disk is huge, that's the sort of thing that you can notice.

As for overclocking, don't do it unless your hobby is overclocking.
Reliablity is far more important than wringing out a little more
performance.

Finally there is one more thing to take in to consideration and that's
future upgrades. There will be dual core 939 pin parts next year, the 754
pin package is only being used for bottom of the line processors so it
won't have a dual core option available.

 

[Adjacent]

Don't take the model numbers of the A64 for much. They aren't a very good
indicator of real performance. I'd suggest you do a little more reasearch
before speding your money. For instance, a socket 754 Sempron 3100+
overclocks easily over the speed of the stock speed of the 4000+, and gets
very close to it in performance. Now consider a MB and 4000+ CPU will
cost you about $850 while the Sempron system would cost less than $200.
And the 3400+ 754 beats the 3500+ 939 in a lot of benchmarks. Don't be too
concerned abpout cache size and dual channel. It's the clock speed that
produces the real performance increase in most apps. Here's one reference
link.

http://www.xbitlabs.com/articles/cpu/print/sempron-3100-oc.html

Not a bad consideration, however, circumstances dictate that I go with
64-bit, and I'd still be interested in any info regarding differences
between the two cores at the 3500+ level.

Thanks,
Ad

 

[Woodrow Phiefer]

" I understand that the higher numbers i.e. 2800, 3000, 3200, etc. are
increasingly better, but what does it mean. I see that most of these have
the same clockspeeds, but the higher numbers (2800,3200...) cost more, so
they must be better, right? Why? What makes them better if not clock
speed?
"


AMD's early Athlon, based on the Thunderbird core, didn't have a PR Rating
such as the 2800+, 3000+ or 3200+ that you refer to. The Thunderbird's
marketing relied on its clock speed, and buyers tended to compare it to an
Intel processor of similar clock speed.

When AMD moved to the Athlon XP, firstly with the Palomino core, the new
architecture had significantly begun to out-perform Intel per clock cycle.
If they were to continue marketing their processors by clock speed alone,
they would only be selling themselves short. Therefore, they introduced
the
PR Rating, indicating the relative performance to their early Athlon
Thunderbird. However, these PR ratings generally bear more relevance to
the
competition from Intel than to the Thunderbird.

AMD processors with the same clock speed may have differing PR ratings
because of a number of factors. Core architechture, L2 cache, FSB and
socket type are the major factors in their performance.

http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_9485_9487^10248,00.html

Thanks, Cuzman, especially for the link, as that lays it all out. However, I
still can't find any info regarding the cores.

Thanks!

 

[MCheu]

OK, I can't find it on AMD's website, but someone has to know. I understand
that the higher numbers i.e. 2800, 3000, 3200, etc. are increasingly better,
but what does it mean. I see that most of these have the same clockspeeds,
but the higher numbers (2800,3200...) cost more, so they must be better,
right? Why? What makes them better if not clock speed?

Thanks,
Ad

The numbers represent the performance of the CPU relative to the
original Athlon K7, which, not surprisingly is equivalent in
performance to a Pentium II. A 3200+ is supposed to perform as well
as an Athlon K7 or Pentium II clocked at 3200MHz, and assuming a
linear correlation in clock speed vs performance. Of course, a plot
of speed vs performance isn't totally linear, especially near the
upper end of a chip family's clock speeds. Using clock speeds as an
indicator of raw performance really goes out the window when you start
throwing different chip designs (even minor differences and
situations). Even intel realizes this, and has been changing their
naming systems.

 

[General Schvantzkoph]

Not a bad consideration, however, circumstances dictate that I go with
64-bit, and I'd still be interested in any info regarding differences
between the two cores at the 3500+ level.

Thanks,
Ad

There are virtually no differences between the two cores, it was just a
die shrink. They might have rolled up some bug fixes when the did the
shrink so if you want to make sure that you have the very latest version
get the 90nm part.

 

[Woodrow Phiefer]

There are virtually no differences between the two cores, it was just a
die shrink. They might have rolled up some bug fixes when the did the
shrink so if you want to make sure that you have the very latest version
get the 90nm part.

Thanks, General. Does the smaller core present greater heat issues that
haven't been addressed? And are you aware of any other advantages with the
later core? If not, of course, I would prefer to go with the cheaper of the
two.

Thanks,
Wood

 

[Noel]

There are virtually no differences between the two cores, it was just a
die shrink. They might have rolled up some bug fixes when the did the
shrink so if you want to make sure that you have the very latest version
get the 90nm part.

There are some small performance tweaks in the 90nm part, and it runs
quite a bit cooler than its 130nm equivalent.

 

[Wes Newell]

I beg to differ, it's total RAM size that matters most.

When comparing cpu's one would assume you compared them with the same
amount of ram. But OTOH, once you have enough ram, adding more could
actually slow the system down, and will in a lot of cases. i've never had
more ram than I currently have (512M) and I'm not using all of it now,
running about 70 processes. And I' don't recall ever using any swap space.

The clock speed differences are trivial, generally about 10% per speed
step.

True, but doubling the cache size, and using dual channel ram only
produces about half that performance wise. All AMD did with the 4000+ was
double the cache from 512K to 1m of the 3800+ and then labeled it a 4000+.
The extra cache size would give the cpu about 3% overall performance
increase. That's only about 100 in PR, not 200. Don't believe me, check
the benchmarks and you'll see that its the clock speed increase that
produces the major performance gains, not cache, and not dual channel.

Human beings don't notice tiny differences like 10 or 20%. However
having enough RAM so that your applications all stay in memory rather
than having to be paged in from disk is huge, that's the sort of thing
that you can notice.

Agreed, but that wasn't the focus of this. CPU performance was.

As for overclocking, don't do it unless your hobby is overclocking.
Reliablity is far more important than wringing out a little more
performance.

AMD builds a new core. This core is capable of xxGHz, but they will never
release all the cores at it's top speed. Same for Intel and any other cpu
maker. They release the core in at a speed that will sell the most and
leave them room to up the core speed once the consumer wants more. Anyone
that's followed cpu releases for a few years knows this. As for
overclocking. It's considered overclocking to clock the cpu faster than
the rated cpu speed, but consider that the 2500+ is a Barton core and the
3200+ is also the same barton core. So is it really overclocking to run
the 2500+ Barton core at a 3200+ speed? yeah, the cpu is overclocked, but
the core isn't. Overclcoking within reason is not a problem. And if you
don't exceed max vcore, it's within specs of the cpu. personnally I don't
like to go over .1v vcore increase of the max that AMD sets vcore to on a
core. that would be 1.6v for A64's I think. Although I admit having mine
up to 1.7v without any problems other than about a 8C increase in temp.

Finally there is one more thing to take in to consideration and that's
future upgrades. There will be dual core 939 pin parts next year, the
754 pin package is only being used for bottom of the line processors so
it won't have a dual core option available.

Trust me on this. Anyone that buys a dual core A64 will definately have
enough money for a new MB.

 

[Wes Newell]

The numbers represent the performance of the CPU relative to the
original Athlon K7, which, not surprisingly is equivalent in
performance to a Pentium II. A 3200+ is supposed to perform as well
as an Athlon K7 or Pentium II clocked at 3200MHz, and assuming a
linear correlation in clock speed vs performance.

The model numbers compare to the P4, not the P2 or even P3. they even lay
out the P4 systems they used for comparisons in this doc.

Benchmarking_Methodology2_v2.5.pdf

AMD Athlon" XP Processor Benchmarking and Model Numbering Methodology

 

[General Schvantzkoph]

Thanks, General. Does the smaller core present greater heat issues that
haven't been addressed? And are you aware of any other advantages with the
later core? If not, of course, I would prefer to go with the cheaper of the
two.

Thanks,
Wood

The smaller core was supposed to run cooler but I've seen at least one
review site which says that it doesn't run. The die size is smaller which
means that it has less area to transfer the heat however I wouldn't worry
about it. AMD packages a cooler with the chip which they've determined to
be adequate, I'm sure they throughly tested it. There are better coolers
available, I'll leave the recomendations for those to others.

 

[General Schvantzkoph]

AMD builds a new core. This core is capable of xxGHz, but they will never
release all the cores at it's top speed. Same for Intel and any other cpu
maker. They release the core in at a speed that will sell the most and
leave them room to up the core speed once the consumer wants more. Anyone
that's followed cpu releases for a few years knows this. As for
overclocking. It's considered overclocking to clock the cpu faster than
the rated cpu speed, but consider that the 2500+ is a Barton core and the
3200+ is also the same barton core. So is it really overclocking to run
the 2500+ Barton core at a 3200+ speed? yeah, the cpu is overclocked, but
the core isn't. Overclcoking within reason is not a problem. And if you
don't exceed max vcore, it's within specs of the cpu. personnally I don't
like to go over .1v vcore increase of the max that AMD sets vcore to on a
core. that would be 1.6v for A64's I think. Although I admit having mine
up to 1.7v without any problems other than about a 8C increase in temp.

Not all chips are created equal even on the same wafer. In fact one corner
of a chip can be faster than another, when you design a chip you have to
take these variations into account when you do the timing calcuations. As
part of the manufacturing process the chips are tested and separated into
speed grades. Only a small precentage of the chips fall into the highest
speed grades, the bulk will be in the center. Chip manufacturers
continually tweek their processes, primarily to improve yields and
secondarily to improve performance. If the center of the distribution
curve moves right they introduce new higher speed grades and drop off some
of the lower speed grades. When a manufacturer puts a speed grade on a
chip they are saying that they guarantee that it will run at a specified
clock rate at a specified maximum die temperature. It is possible that any
particular chip will run at a faster clock rate but it's not guaranteed.
The manufacturer would be leaving a lot of money on the table if the bulk
of the chips that they sold were underrated so it isn't likely that many
are. Overclocking is dependent on keeping the die temperature
substantially below the maximum temperature that the chip is specified
for, because CMOS gets faster as it gets cooler. Puting on a better heat
sink is a relatively cheap means of reducing the die temperature but it
isn't going to allow you substantilly over clock the CPU, you might be
able to run it a little faster but as I've said before it's not enugh so
that you'll notice it. There are exotic cooling solutions like water
cooling which might allow you to overclock the CPU by several speed grades
but those solutions are so expensive that you could build a dual processor
system for the same money.

 

[Wes Newell]

The manufacturer would be leaving a lot of money on the table if the bulk
of the chips that they sold were underrated so it isn't likely that many
are.

And they would leave more money on the table if they didn't compete with
the competition so I'll commit to saying that every single cpu they make
except the top of the line is under rated. Why? Simple. I, being a PC
builder, want 50K parts of a low end CPU in 90 days and if you won't
commit to providing them I'll go elsewhere. So AMD builds 50K parts, but
they all speedbin at a higher rating. Do you actually think they will not
mark them at the lower speed and sell them? Get real, I've been in this
business for over 30 years. All IC's built are on real or projected
demand, not what they will speed test at. It's all marketing and has very
little to do with what speed they test at.

This should be obvious. Just look at the Sempron line. The Sempron 3100+
(top of the Sempron line) has a default clock of 1.8GHz. So the first ones
to hit the market clock over 2.5GHz on a reduced default voltage of 1.4v.
(Normal for K8 cores is 1.5v). So please tell me why they sell these
clocked at 1.8GHz when they will clock 40% higher? So, basically, your
arguement doesn't hold water.

 

[John Smithe]

OK, I can't find it on AMD's website, but someone has to know. I
understand that the higher numbers i.e. 2800, 3000, 3200, etc. are
increasingly better, but what does it mean. I see that most of these
have the same clockspeeds, but the higher numbers (2800,3200...) cost
more, so they must be better, right? Why? What makes them better if
not clock speed?

Thanks,
Ad

high-performance/faster: higher clock speed, bigger cache, dual channel
memory, faster HyperTransport bus
low-performance/slower: lower clock speed, smaller cache, single channel
memory, slower HyperTransport bus

socket 939 implies dual channel memory and faster HyperTransport bus
socket 754 implies single channel memory and slower HyperTransport bus

Checkout these sites. The first site has many 'tech guides' you may find
helpful. The second site is an AMD site listing Athlon64 features vs PR
rating.

http://forums.amd.com/index.php?s=b6d3363e29a49650c968e49c67c958ab&act=idx
http://www.amd.com/us-
en/Processors/ProductInformation/0,,30_118_9485_9487%5E10248,00.html
http://www.amdzone.com/modules.php?op=modload&name=Sections&file=index&req=
viewarticle&artid=63&page=1
http://www.amdboard.com/amdid.html

The machine I intend put it in is my "workhorse" system. I mostly use it
to record and process video and audio, burn DVDs & CDs and other time and
somewhat processor intensive jobs.

If you are working with large files, as implied by video and audio
processing, more memory will be vital to good performance. The socket 939
motherboards will support more memory than the socket 754 motherboards. At
a FSB speed of 400MHz/PC3200 a socket 754 cpu, and therefore motherboard,
is rated for only two memory sticks while a socket 939 cpu is rated for 4
memory sticks. Read the motherboard manuals and specification sheets of
your candidate motherboards carefully to verify possible memory constraints
on socket 754 motherboards.

Any recommendation on motherboards? I've
been checking out Gigabyte GA-K8NS PRO.

Also checkout the MSI K8N Neo2 Platinum (939) or MSI K8N Neo Platinum (754)
and the DFI LANPARTY UT nF3 250Gb (754). I looked for a DFI 939 but
couldn't find one. I believe these, as well as the GA-K8NS PRO, have all
received good reviews.

http://www.msicomputer.com/product/p_list.asp?class=mb&cpu=3
http://www.dfi.com.tw/Product/xx_product_spec_details_r_us.jsp?PRODUCT_ID=2
840&CATEGORY_TYPE=MB&SITE=NA

Good Luck