AMD Opteron Model numbers revealed...

[ragiepew]

The inquirer has several stories up this morning regarding Opteron model numbers...

1
2
3

I've linked just a couple of them... hit The Inquirer for more...

AMD will use numbers such as the 842 (1.6GHz) with the first number, eight, representing scaleability, the second number four represent relative performance against other Opterons, and the third digit showing where the chips are in the overall scheme. µ

Anyway, is it just me or are these numbers more confusing than the AXP ratings? I mean I understand the first number... "Chip scalability" but what exactly is the difference between the last two...

 

[Nemesis77]

Does the number matter? I mean, Sun has names like Sune Fire 3800, Sun Enterprise 10000, Sun Fire 15K etc. etc. How about those At Radeons? 9000, 9100 (AKA 8500), 9500, 9600, 9700, 9800, (and the pro-models as well). Do the numbers make any sense? I doubt it. Well, Radeon-numbers are supposed to tell the "relative performance between different prodcuts" or something. Yet people not complain, even if they use arbitary numbers like that.

At least AMD has some reasoning behind their numbers. But in the end, it doesn't matter.

 

[paralazarguer]

Argh, the inquirer...shakes head...have you read their B.S. Story about the Barton 400mhz fsb...what a "publication"

 

[splice]

It maybe more confusing, but most of us are not going to be buying Opteron chips. I don't see it as been to difficult to unserstand.... Anyways, "most" chip mfg's use numeric names for their processors which have very little to do with clock speed (ie: MIPS R4x00, NEC V8xx, MOS6502, Hitachi SH-1/2/3/4, MOT 68k, Zilog Z80, ARM7, etc...)

As far as AMD's number scheme... it seems a little limited (0-9), unless they plan to use alpha chars too. This is how I understand it:

842

Digit 1 (2): Overall Scheme - I think this will be the model/revision of the chip (core changes/die-shrink/MEM-IO speed or type/etc)
Digit 2 (4): Relative Performance - Performance compared to other opertons in it's current scheme? (2) Only 10 degrees of freedom here if no alpha chars (0-9).
Digit 3 (8): Scaleability - for use in upto 8-way system.. probably always going to be on of the following (2,4,8)

 

[ragiepew]

Originally posted by: paralazarguer
Argh, the inquirer...shakes head...have you read their B.S. Story about the Barton 400mhz fsb...what a "publication"

While I agree somewhat the TheInq isnt an official publication they have been known to be right on several occasions.

Also, I do understand that most of us wont be buying an Opteron based system it just seems as if they are making a big change in naming conventions. While the AXP used 1700, 1800. etc... those were all frequency relative meaning that this chip(s) would perform about equal to a 1700 or 1800 mhz Pen... Thunderbird Athlon... even though it's only clocked at ~1400mhz. I'm just wondering whether or not AMD's customers will be thrown off by the naming conventions, especially when Intel is sticking to MHz. I know Intel isn't the only game in town, especially in the 64bit chip arena, however they do have quite a bit of influence.

 

[Woodchuck2000]

I think that dispensing with MHz/PR ratings is a step in the right direction. The 1.6GHz opteron will probably in the same league as a 2.6GHz Xeon - removing clock speed from the name forces people to look at benchmarks for comparison rather than bitching about the PR system.

 

[RaynorWolfcastle]

Originally posted by: Woodchuck2000
I think that dispensing with MHz/PR ratings is a step in the right direction. The 1.6GHz opteron will probably in the same league as a 2.6GHz Xeon - removing clock speed from the name forces people to look at benchmarks for comparison rather than bitching about the PR system.

You're kidding right? You're expecting people to actively seek out benchmarks? That's extremely naive to say the least.

So you think that selling Opterons 842, 846, and 858 is a good idea? I wanna know what frequency these processors run at without digging through f*&%ing spec sheets. At least with the current PR ratings they usually write "1700+ Quantispeed Architecture running at 1.47 GHz". How the hell am I supposed to know how much of an improvement there is between the 842 and the 852 without digging through benchmarks and white paper reports?

I really hope that they don't use this numbering scheme for the Athlon 64.

 

[Sunner]

I kinda doubt they'll use numbers like that even for the Opterons.
Maybe along with some kind of other "model number" scheme, but not by itself.

As for Sun's systems, that's different, those are entire systems you're looking at.
In a SunFire 6800, you can have 24 UltraSPARC-III CPU,s and...what...256 GB of main memory, but the CPU's are still called UltraSPARC-III xxx MHz, and you buy xxx MB DIMM's for them, no wierd model numbers there, just different names for their entire systems.

Besides, for the market Opteron is targeting, MHz, model numbers, etc isn't as important as it is for the consumer market.
I buy lots of stuff for the company I work at, and I don't just look at who has the highest MHz, I look at what we need, budget, etc, then gather some info about various available products, to try to make an informed decision.
Other business customers do the same.

"Avarge Joe" type Customers don't, which is why model numbers, MHz, etc are important.

 

[Nemesis77]

Originally posted by: RaynorWolfcastle
You're kidding right? You're expecting people to actively seek out benchmarks? That's extremely naive to say the least.

You think that people who buy these servers (equipped with SPARCs, Itaniums, Opterons, Xeons and the like) stare at the MHz? I think not.

 

[RaynorWolfcastle]

Originally posted by: Nemesis77

Originally posted by: RaynorWolfcastle
You're kidding right? You're expecting people to actively seek out benchmarks? That's extremely naive to say the least.

You think that people who buy these servers (equipped with SPARCs, Itaniums, Opterons, Xeons and the like) stare at the MHz? I think not.

actually I was refering to desktop processors (Athlon 64) , I guess I should have been clearer about that... But I still think that their numbering scheme is needlessly convoluted...

Boss: "So Bob, what kind of processor do you think our server needs?
Bob: "I'm thinking judging by these charts here *out come the benchmarks* that the 699 and the 747 are pretty good, but we might think of getting the 844 so that in the future we can think of adding another 844 or perhaps we can replace them with 856s. Or maybe we shoulf just buy the 699s now and wait for the 870s to come out, those should be significantly faster.
Boss: *blank stare* so the one we want is...?

My point is that you don't need to try to compress a comprehensive performance assessment based on 24 different benchmarks, the competiton's available processors, and the cost of tea in India in a 3 digit number. Clarity should not depend on whether or not it's aimed at corporate buyers. If a corporate buyer is going to be looking at benchmarks anyway, what's the point in trying to compress 20 pages of graphs into a 3 digit number? Why not just continue to use clockspeed to name different revs of the same product then?

 

[Nemesis77]

Originally posted by: RaynorWolfcastle
actually I was refering to desktop processors (Athlon 64) , I guess I should have been clearer about that...

Well, this is about Opteron model-numbers, not Athlon 64's (which is the desktop-version)

But I still think that their numbering scheme is needlessly convoluted...

How so? I mean, we have had such monstrocities as 68000, 68010, 68020, 8080, 8088, 80386, 80486, 21064, 21164, 21264...

Why not just continue to use clockspeed to name different revs of the same product then?

Because this scheme actually tells more about the CPU than plain-jane MHz-rating does.

 

[RaynorWolfcastle]

Originally posted by: Nemesis77

But I still think that their numbering scheme is needlessly convoluted...

How so? I mean, we have had such monstrocities as 68000, 68010, 68020, 8080, 8088, 80386, 80486, 21064, 21164, 21264...

Why not just continue to use clockspeed to name different revs of the same product then?

Because this scheme actually tells more about the CPU than plain-jane MHz-rating does.

I see you conveniently left out what I said between those 2 quotes

If a corporate buyer is going to be looking at benchmarks anyway, what's the point in trying to compress 20 pages of graphs into a 3 digit number?

Can you please answer that for me?

ummm... and these monstrosities are exactly not a problem since they denote a PRODUCT LINE. Hence, I can easily compare a 20 MHz 80386 to a 30 MHz 80386, furthermore, I can infer that the the 30 MHz version will be roughly 50% faster in CPU bound tasks. If you're comparing a an 80386 with a 21064 (hypothetically ) you'd have to look at benchmarks ANYWAY to tell me how they compare.

If I tell you that I'm hesitating between a model 842 and a model 852 Opteron can you easily tell me what kind of performance difference there is between the 2? Probably not, you'll have to break out the 20 pages of benchmarks before you can say something like "it's about 10% faster in most applications". Which begs the question, why the hell give up the the simplicity of frequency in exchange for a model number that no one will ever really use anyway?

Do you see the point I'm making?
AMD's idea of wanting to give the customer more information is misguided because it's too complicated for a quick and dirty evaluation of performance and it's too simplistic for a comprehensive evaluation of performance; hence it is needlessly convoluted

 

[Woodchuck2000]

Originally posted by: RaynorWolfcastle

Originally posted by: Woodchuck2000
I think that dispensing with MHz/PR ratings is a step in the right direction. The 1.6GHz opteron will probably in the same league as a 2.6GHz Xeon - removing clock speed from the name forces people to look at benchmarks for comparison rather than bitching about the PR system.

You're kidding right? You're expecting people to actively seek out benchmarks? That's extremely naive to say the least.

So you think that selling Opterons 842, 846, and 858 is a good idea? I wanna know what frequency these processors run at without digging through f*&%ing spec sheets. At least with the current PR ratings they usually write "1700+ Quantispeed Architecture running at 1.47 GHz". How the hell am I supposed to know how much of an improvement there is between the 842 and the 852 without digging through benchmarks and white paper reports?

I really hope that they don't use this numbering scheme for the Athlon 64.

Graphic card names tell you nothing about the specs of the card and I don't here you bitching that the Radeon 9000 is slower than a GF 4200... Why does the frequency of a CPU matter any more? It's not as if you can use it for comparing performance with other cpus.

 

[Nemesis77]

Originally posted by: RaynorWolfcastle
I see you conveniently left out what I said between those 2 quotes

If a corporate buyer is going to be looking at benchmarks anyway, what's the point in trying to compress 20 pages of graphs into a 3 digit number?

Can you please answer that for me?

Because that 3 digit numbers tells more about the CPU than the MHz-rating would.

If I tell you that I'm hesitating between a model 842 and a model 852 Opteron can you easily tell me what kind of performance difference there is between the 2?

Nope, since I just heard about this thing, and I haven't heard the details yet. Once the CPU's are released, we get real benchmarks, then that number would tell me more.

why the hell give up the the simplicity of frequency in exchange for a model number that no one will ever really use anyway?

Because it tells more about the CPU than plain MHz-rating would. It also tells about features that server-customers want to know about (scalability). Does it give you detailed information regarding the performance? Of course not. But it gives you the generic idea.

I would wait 'till we hear official confirmation to this scheme along with detailed explanation.

 

[RaynorWolfcastle]

Originally posted by: Woodchuck2000

Graphic card names tell you nothing about the specs of the card and I don't here you bitching that the Radeon 9000 is slower than a GF 4200... Why does the frequency of a CPU matter any more? It's not as if you can use it for comparing performance with other cpus.

CPUs and video cards are very different beasts. the difference between the Radeon 9000, 9100, 9500, and 9700 is more than just clockspeed, there are fundamental architectural differences. That is if you ask me to compare the 9000 to the 9100 to the 9500 based on clock speed alone I can't really tell you what the performance difference is (let alone their featureset, something that doesn't directly apply to CPUs, but I digress).
This means that there's no point in naming video cards by clockspeed.

I imagine that AMD will be selling Opterons will likely have the exact same architecture at 1.6 GHz, 1.8 GHz, 2.0 GHz, 2.2 GHz, etc... The difference between all of these is only their clockspeed not their architecture, so their freq. is a valid way of comparing these processors to each other.

My point is that you can use the freq. to compare various revs OF THE SAME ARCHITECTURE in a quantifiable way. Models like "842" tell you even less about the processor since no one knows off the top of their head how much faster the "843" is over the "842". It doesn't help you know how it compares to Xeons/Alphas/SPARCs either since Intel/Sun/Compaq doesn't release their chips with 3 digit performance indicators.

Bottom line: frequency gives you 1 piece of quick, quantifyable, information; a 3 digit model number gives you none. That's all I'm saying.

*edit* I'm done with this thread, there's no point in arguing over something this ridiculous any further. You have your opinion on the proposed model numers, I have mine; I can live with that.

 

[ragiepew]

why the hell give up the the simplicity of frequency in exchange for a model number that no one will ever really use anyway?

The reason for this is pretty obvious.. in that MHz only means something when compared to the same brand CPU... for example Athlon v. Athlon or P4 v. P4. The minute you compare a P4 and Athlon in terms of pure clockspeed it becomes a big problem. Sure the model numbers dont really say much about the processor itself but maybe its a better way of describing the processor. I do, however, hope that they include the frequency somewhere easy to find but that should not be the end basis on what processor is purchased... especially when it comes to server class CPU's.

 

[Pariah]

As far as AMD's number scheme... it seems a little limited (0-9), unless they plan to use alpha chars too. This is how I understand it:

Not really, you have to consider the market. How many Itanium and Itanium2 clock speed grades are there? 2 each. Sun has 4 or 5 for each of their lines. This isn't the desktop market where a new speed grade is released every 2 months.

AMD's idea of wanting to give the customer more information is misguided because it's too complicated for a quick and dirty evaluation of performance

When you're spending up to $25k or more on a system, there's no such thing as quick and dirty.

If I tell you that I'm hesitating between a model 842 and a model 852 Opteron can you easily tell me what kind of performance difference there is between the 2? Probably not, you'll have to break out the 20 pages of benchmarks before you can say something like "it's about 10% faster in most applications".

Unlikely that scenario would ever come up. These CPU's are not sold by themselves but as part of a system. It is up to the vendor to demonstrate the capabilities of the systems to the customer to get them what they need. No one is blindly buying in this market based on stat sheets or visiting THG for Quake III benchmarks.

It doesn't help you know how it compares to Xeons/Alphas/SPARCs either since Intel/Sun/Compaq doesn't release their chips with 3 digit performance indicators.

It doesn't matter what numbering scheme AMD chooses, it won't have any useful basis for comparison to multiple completely different architectures.

 

[RaynorWolfcastle]

OK, so I couldn't help myself .

Pariah, you're validating my point with your comments. Why bother making some complex formula for numbering your CPU if no one uses it anyway?
I'm pretty sure that vendors will be selling the same system with CPUs of varying speeds though, so freqency still has some value for comparing systems that are identical apart from differently clocked CPUs hence frequency is likely more practical than the 3 digit numbering scheme

 

[Pariah]

Pariah, you're validating my point with your comments. Why bother making some complex formula for numbering your CPU if no one uses it anyway?

No, I'm not. I don't know what you find so complicated about the number scheme. IMO, it is far better than the travesty that is the current XP rating system. The current scheme is the "complicated" formula method that needs to be constantly tweaked and changed to keep it somewhat accurate. Now if you buy an Athlon XP 2600+ you could get one of 3 different CPU's, and with the impending release of a 400MHz bus version, you could get one of 4. The MHz mean nothing because once again, there are 3 different revisions of Athlons floating which all perform differently at the same clock.

The new Opteron number system eliminates any confusion of what you are getting. 1st number gives scaleability, if you want a dual system you buy a 2, if you want 8way you buy an 8. Second number gives core revision which would tell you cache amount, bus speed, etc(Barton,TBred 266,TBred 333, for example), third number gives relative speed ranking within the revision. Seems logical and straightforward to me.

 

[ragiepew]

TextThe new Opteron number system eliminates any confusion of what you are getting. 1st number gives scaleability, if you want a dual system you buy a 2, if you want 8way you buy an 8. Second number gives core revision which would tell you cache amount, bus speed, etc(Barton,TBred 266,TBred 333, for example), third number gives relative speed ranking within the revision. Seems logical and straightforward to me.

Wow, that explanation makes a lot of sense, and the best one yet... i was a bit confused about what each number meant, especially the last two, but thanks for clearning that up for me.