PDA

View Full Version : What's the diff between Athlon FX-55 and 64 4000+?


singularity2006
11-01-2004, 05:45 AM
Between these specs:

Model: AMD Athlon 64 FX-55
Core: ClawHammer
Operating Frequency: 2.6GHz
FSB: Integrated into Chip
Cache: L1/64K+64K; L2/1MB
Voltage: 1.5V
Process: 0.13Micron
Socket: Socket 939
Multimedia Instruction: MMX, SSE, SSE2, 3DNow!, 3DNow!+

and

Model: Athlon 64
Core: Clawhammer
Operating Frequency: 2.4GHz
FSB: Integrated into Chip
Cache: L1/64K+64K; L2/ 1MB
Voltage: 1.5V
Process: 0.13Micron
Socket: Socket 939
Multimedia Instruction: MMX, SSE, SSE2, 3DNOW!, 3DNOW!+

Is the higher clockspeed on the FX the only difference between the two?

lots
11-01-2004, 06:15 AM
yes..

The Athlon64 4000+ is a relabled Athlon FX-53. The only difference between the FX53 and the 55 is clock speed.

Sieb
11-01-2004, 07:21 PM
Clockspeed and target audience. AMD is blurring the line between them with the 754 pin A64s being phased out with the desktop 940pins in favor of all 939s. Later on, the FX will transition to DualCore whereas the A64 won't.

Thalaxis
11-01-2004, 07:45 PM
AMD has disclosed plans to move the Athlon64 line to dual-core, but not until 65nm. They're planning to release dual-core Opteron and Athlon64FX parts next year on 90nm.

I think it will be at minimum another year before they have dual-core mainstream processors on the market, but that's not even a prediction, it's just a relatively wild guess :)

As for why, just look at Sieb's answer, so that I don't have to retype it ;)

lots
11-01-2004, 07:58 PM
Indeed, 2006 is looking to be the year of dualcore for the masses.

Thalaxis
11-01-2004, 08:19 PM
That may be the understatement of the hour ;)

I'm hoping to have a dual-core Pentium-M laptop, and a dual-core Athlon64FX :)

lots
11-01-2004, 08:26 PM
Yeah, I've been thinking about dual core. But I was planning on purchasing in Dec, not in the summer. So, I might just buy 2 low class opterons and then use them for network render nodes when I grab a pair of Dual cored Opterons in the summer ;).

Or just buy high end now and wait a few years and get the next best thing out there :P

Thalaxis
11-01-2004, 08:30 PM
That's the sort of reason that I ended up with a P4 in the first place... I really wanted to get an Athlon64, but they weren't available when I was ready to make my purchase.

This time though, I'm not going to be buying a new machine for a while, because I'm working on house shopping AND getting married. :)

lots
11-01-2004, 09:25 PM
I think I may just wait for 90nm Opterons to start appearing on the market, and hopefully will be connected to a price drop.

Thalaxis
11-01-2004, 09:35 PM
That shouldn't require too long a wait. The only catch is that AMD might hold them in reserve and build up a bit of a stockpile, and use them to ambush Intel when they make their next P4/Xeon related product launch :)

Seriously, it seems that AMD's transition to 90nm is going pretty well, so it probably won't be long. If they're planning to launch any new speed grades before the holidays, they'll probably announce it quite soon, because most companies don't like to launch products that close to Christmas, because by the time it makes it through distribution, they end up missing the holiday buying sprees.

Also, after Christmas prices tend to drop. :)

lots
11-01-2004, 09:54 PM
Yeah, hence my reasoning for a Dec/Jan buy time frame, hopefully lower prices. Of cource I'm willing to give it a few months if that means I can have my "uber system" for considerbly less :). From what I can see of AMD's 90nm transition, they took it far more cautiously than Intel did and ONLY migrated what had already been in circulation to the 90nm process. Unlike Intel who added loads of features to an already complicated transition. Of cource the SOI process AMD impliments probably helps here too ;) From what I understand, AMD is using strained silicon for implimenting thier 90nm CPUs, which apparently helps with the transistor leakage, too. Though I'm not too informed of the specific techniques used to create modern CPUs :)

Also AMD is building another fab (right? I think I saw this somewhere). That hopefully means larger volume and lower prices from AMD in the future as well. Though they're still using 200mm wafers, time for an upgrade eh? :P

Thalaxis
11-01-2004, 10:14 PM
Yeah, hence my reasoning for a Dec/Jan buy time frame, hopefully lower prices. Of cource I'm willing to give it a few months if that means I can have my "uber system" for considerbly less :).

That's usually a good thing :)


From what I can see of AMD's 90nm transition, they took it far more cautiously than Intel did and ONLY migrated what had already been in circulation to the 90nm process. Unlike Intel who added loads of features to an already complicated transition. Of cource the SOI process AMD impliments probably helps here too ;) From what I understand, AMD is using strained silicon for implimenting thier 90nm CPUs, which apparently helps with the transistor leakage, too. Though I'm not too informed of the specific techniques used to create modern CPUs :)

Based on what info AMD's divulged, that seems about right. Unlike Intel, they don't have a fab that they can transition without interrupting their production, so they had to be more cautious, regardless of any other issues.

Also, I think they want to break into the laptop market some more, so they're putting extra effort into keeping the power consumption down.


Also AMD is building another fab (right? I think I saw this somewhere). That hopefully means larger volume and lower prices from AMD in the future as well. Though they're still using 200mm wafers, time for an upgrade eh? :P
Yes, they are. It's in the same location, and it's supposed to use 300 mm wafers, start with 90 nm process technology, and transition to 65nm at full volume. IIRC they plan for something like 20,000 wafer starts a week, which puts it at something close to 2x the capacity of IBM's Fishkill megafab.

AMD is looking to gain some market share, and right now I think they're pretty much capped out at their volume limit, so they can't grow any further without that new fab.

And now they have a couple of extra products to fab, in addition to flash memory; they have a MIPS-based network processor, and the new Geode, both of which are potential cash cows if they get a couple of design wins.

singularity2006
11-02-2004, 04:21 PM
Wow what a great discussion!! That's some great info. Yeah, I was thinking of upgrading to an Athlon 64 4000+ sometime next summer before I transfer schools..... that's a lot of good material to read through. Just skimmed some of it, but it's been pretty good so far!

Sieb
11-02-2004, 05:13 PM
http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_9485_9487,00.html

http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_9485_9488,00.html

http://www.amd.com/us-en/Processors/SellAMDProducts/0,,30_177_3532_1387%5E608,00.html

Lord Banshee
11-02-2004, 06:14 PM
http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_9485_9487,00.html

http://www.amd.com/us-en/Processors/ProductInformation/0,,30_118_9485_9488,00.html

http://www.amd.com/us-en/Processors/SellAMDProducts/0,,30_177_3532_1387%5E608,00.html
That road map was very helpfull

I am sad though i was really hopeing to upgrade to a dual core amd64 before 2006. But it doesn't seem to do that.

So i guess i will either get a dual core opteron 100 series or a AthlonFX dual core.... I can see the money going down the drain already.

Dexter | DTG
11-03-2004, 11:03 AM
The FX-55 has an unlocked multipler. Where as the 4000+ only can be changed below its default.

This makes bugger all difference unless you are overclocker/bencher. Also the FX-55 uses strained silicon in its production so it overclocks better.

singularity2006
11-11-2004, 04:34 AM
Clockspeed and target audience. AMD is blurring the line between them with the 754 pin A64s being phased out with the desktop 940pins in favor of all 939s. Later on, the FX will transition to DualCore whereas the A64 won't.
okay, so 754's are going out there door but i wasn't clear about what u meant between 940 and 939. So are u saying 940's will be replacing 939's too or will 939's be replacin' the 940's?

lots
11-11-2004, 06:12 AM
Neither.

Socket 940 is for the workstation (read: single, dual, and up opterons)
Socket 939 is for the consumer market (read: single chip Athlon64/AthlonFX, apparently dual core Athlon64/FX will use the same socket as well, or at least everything I've read indicates this)

Really, there's not much of a difference between the two sockets. Socket 940, probably just has the one extra pin for ECC Registered ram. Though I cant remember off the top of my head what the extra pin is for. Could be for electrical reasons and nothing really to do with how the chip processes data.

Socket 940 was actually introduced first (with the first Opterons in april of last year), then came socket 754 (the first Athlon64s) and then finally socket 939 (the latest Athlon64s).

I dont think there will be many socket changes for a while now. AMD has also made it very clear that socket 940 isnt going anywhere. Most of the customers of Socket 940 chips will dispise if AMD they try to phase it out. Socket life time is much longer in the server/workstation market than on the consumer end.

singularity2006
11-11-2004, 07:26 AM
ah, gotcha. Thanks for clearing that up....

Thalaxis
11-12-2004, 10:38 PM
Really, there's not much of a difference between the two sockets. Socket 940, probably just has the one extra pin for ECC Registered ram. Though I cant remember off the top of my head what the extra pin is for. Could be for electrical reasons and nothing really to do with how the chip processes data.

I think it also disables SMP support, so that you can't have a dual socket 939 motherboard, even though the Opteron and Athlon64FX are nigh identical. Since the difference is implemented in the physical package, it's far harder to hack than if it was just a CPUID thing.

lots
11-13-2004, 07:57 AM
Well here's how the Opteron is designed.. All Opterons, be it 1xx 2xx or 8xx series, have 3 HyperTranspot connections. The main difference between the 3 series is the number of coherent and non-coherent HyperTransport links. Coherent links are used to communicate cache-coherency info between multiple processors in a multi cpu setup. Non-coherent HyperTransport links are used for communication to IO devices, such as the chipset. Coherent links can also act as non-coherent links, for communication to the chipset. Now, since the Opteron 1xx are meant for single chip operation, they have no coherent links. Thus are only capable of uniprocessor usage. The AthlonFX, is essentially an Opteron 1xx that includes only a single, active non-coherent HyperTransport link (the other two are disabled), versus 3 HT links in the Opteron, though these days it is being packaged in the socket 939, and has the ability to use unbuffered non ecc ram (hence why I think the extra pin in Socket 940 is for ECC Registered ram) The Opteron 2xx series chips support 1 coherent link, meaning they can communicate with one other CPU. Makes sense right? dual CPUs being able to communicate directly with eachother. And the 8xx sereies Opterons have 3 coherent links, for 4-way and 8-way operation.

Thalaxis
11-13-2004, 04:08 PM
That's about right. Physically, the three HT links are identical. The 8-way SMP implementation uses a hypercube configuration, so that at least one of the processors is
connected to the I/O hubs (two can be, actually). The rest use all three HT connections to
communicate with other processors. That way the Opteron 8xx series can function in up to 8
way SMP configurations without any external glue logic. They don't scale as well from four to
eight as they do up to four, but not needing external glue logic makes it more cost effective.

Anand got a few leaks from the Via guys, and they hinted at more sophisticated on-die
switching fabrics. That means more than three HT links :D

lots
11-13-2004, 10:59 PM
Mmm tasty ;) Too bad systems like that are out of reach of my wallet. Though it would be nice to render on..

Thalaxis
11-14-2004, 05:34 AM
They might not be out of reach for all that long, though :)

CGTalk Moderation
01-19-2006, 07:00 PM
This thread has been automatically closed as it remained inactive for 12 months. If you wish to continue the discussion, please create a new thread in the appropriate forum.