AMD's Socket 939, Athlon 64 FX-54 amd 64 3800+

← Back to Stories (view on slashdot.org)

AMD's Socket 939, Athlon 64 FX-54 amd 64 3800+

Posted by CmdrTaco on Tuesday June 1, 2004 @12:53AM from the thats-a-lotta-digits dept.

BudKnight writes "It looks like AMD is launching four new desktop processors, a new core, and a new socket infrastructure today. HotHardware has tested AMD's two new flagship processors, the Athlon 64 FX-53 and the Athlon 64 3800+. The new FX-53 no longer needs registered memory to function and the 3800+ has only 512K of cache, but it gets an upgraded 128-bit memory controller. The usual suspects also have reviews posted as well - TechReport, Hard|OCP, Beyond3D - more are sure to follow."

42 of 160 comments (clear)

No 64bit scores by brejc8 · 2004-06-01 00:54 · Score: 4, Interesting

One silly thing about review sites comparing AMD64 to anything else is that they are still running them in 32bit mode. I found running in 64 bit mode gives you about 20% improvement in general code.
When running guile working on very long integer operations we got a _6_ times improvement. Our simulations dropped from taking an 66 minutes to just over 11 minutes.

--
Mouse powered Chips, Open source Processors and Lego
1. Re:No 64bit scores by MrRuslan · 2004-06-01 00:59 · Score: 5, Insightful
  
  Due to lack of commericial 64 bit code they run 32 bit code cause thats what evryone runs in the real world and it still kicks ass. AMD is not joking around anymore and comming out with good stuff like this.This is the first time I seen something "backwards compatible" thise good.
2. Re:No 64bit scores by Piranhaa · 2004-06-01 01:00 · Score: 4, Insightful
  
  That is true, however the average consumer WILL NOT be running 64-bit until Windows does get its 64-bit edition running perfectly and shipping with it. Although I do think they should have at least benchmarked it with some type of *nix for 64-bit...
3. Re:No 64bit scores by rice_web · 2004-06-01 01:02 · Score: 3, Interesting
  
  Hopefully you understand that not many programs are optimized, as of yet, for 64-bit processors. When the more software makes the transition, you can bet that AMD will inch further ahead of Intel in the CPU-speed game (barring an Intel processor with 64-bit extensions anytime soon).
  
  --
  The Political Programmer
4. Re:No 64bit scores by WegianWarrior · 2004-06-01 01:06 · Score: 4, Insightful
  
  Since just about anything else avilable today (at least to general consumers) are running at 32 bit (disregarding the fact that it's still possible to get hold of 16 bit and 8 bit chips off course) it make damn good sence to compare the new AMD64 to other chips while running it in 32 bit mode. Otherwise, you would be comparing apples and oranges... since no other CPU runs in 64 bit mode.
  
  Could be interesting to benchmark a AMD64 in 32 bit mode against a AMD64 in 64 bit mode thought... that would say a lot about how much there is to gain in going to 64 bit mode in the first place...
  
  --
  Everything in the world is controlled by a small, evil group to which, unfortunately, no one you know belongs.
5. Re:No 64bit scores by Anonymous Coward · 2004-06-01 01:08 · Score: 2, Insightful
  
  The average consumer don't even know how many bits its CPU has.
6. Re:No 64bit scores by brejc8 · 2004-06-01 01:17 · Score: 4, Interesting
  
  Incase people are intrested here are the scores of scibench2 for a load of machines in our office.
  
  Athlon64 3200 64: 523.70
  Athlon XP2700: 467.15
  Athlon64 3200 32: 449.07
  Athlon XP2600: 448.42
  Pentium4 3.0GHz: 387.57
  Athlon 1400: 305.26
  AMD Athlon 950: 209.51
  Sparc 500MHz: 52.21
  Sparc 440MHz: 51.89
  
  --
  Mouse powered Chips, Open source Processors and Lego
7. Re:No 64bit scores by Short+Circuit · 2004-06-01 01:33 · Score: 2, Informative
  
  SGI's Origin supercomputer's are essentially clusters of Itanium 2 boxes in a NUMA architecture...And they're still supporting them.
  
  --
  tasks(723) drafts(105) languages(484) examples(29106)
8. Re:No 64bit scores by lachlan76 · 2004-06-01 01:39 · Score: 2, Interesting
  
  I wouldn't bet on an Intel x86-64 processor in the near future. The Yamhill extensions have probably been scrapped along with the Tejas and the rest of the P4 R&D.
  Moving the extensions back to the P6 architecture will probably extend the time before any Pentium-64 processors are made by a few years at least.
9. Re:No 64bit scores by swillden · 2004-06-01 01:40 · Score: 5, Insightful
  
  Otherwise, you would be comparing apples and oranges... since no other CPU runs in 64 bit mode.
  By that logic, when a processor comes out with a new multimedia extension, or an increased L1 cache size, or a deeper pipeline, or a more efficient instruction scheduler, we should do comparisons with the new feature turned off, because no other CPU has it.
  The real reason these chips were tested in 32 bit mode is because the testers ran WinXP on them for the tests. This is reasonable in that it's what most potential purchasers of the processors would be running, not because it's a more valid comparison against other 32-bit chips. If the most common software were available in 64-bit versions, it would be unreasonable not to use that and let the AMD64 chips show their full capability. (Assuming the software would run faster in 64-bit mode, which isn't necessarily true).
  Users of more flexible software would find it interesting to see how their favorite tools run in 64 bit mode, of course, but that's a smaller audience, so those tests will come later.
  
  --
  Note to ACs: I usually delete AC replies without reading them. If you want to talk to me, log in.
10. Re:No 64bit scores by MrRuslan · 2004-06-01 01:48 · Score: 2
  
  I agree with you. I think they need to take sometime and benchmark suse 32 bit and 64 bit or any other distro for that matter and compare on the same computer to see the advantage of 64 bit code.
11. Re:No 64bit scores by Rich0 · 2004-06-01 01:59 · Score: 3, Informative
  
  Assuming the software would run faster in 64-bit mode, which isn't necessarily true.
  
  Most likely it will - though not necessarily through the use of 64-bit math. The AMD64 adds several additional registers, and that alone can make for a huge performance boost when a complier knows how to use them...
12. Re:No 64bit scores by Jarnis · 2004-06-01 02:09 · Score: 3, Informative
  
  What?
  
  Have you been hibernating past few months?
  
  Prescotts with Intel's version of x86-64 are coming out by early autumn. MS delaying the OS is partly because they don't want to piss Intel off. They also want to ensure the thing works perfectly on both versions, and while they are largerly compatible, there are couple of small differences.
13. Re:No 64bit scores by BlowChunx · 2004-06-01 02:17 · Score: 3, Insightful
  
  ...probably about 3 people have them.
  
  We have an 8 way SGI Altix 350 down the hall. I didn't know I would be in such refined company.
  
  And they can't run 32 bit code natively, it must be done in emulation.
  
  Intel's original purpose was to put these chips in servers and high end workstations, places where 32-bit compatibility was not its main purpose.
  
  For CFD on the codes that I run (full Navier-Stokes equations with LES), an Itanium2 at 1.5 Ghz is twice as fast as a 3.0Ghz P4. (According to SGI these are $10k/chip, the lowly 1.4 Ghz are $3k/chip)
14. Re:No 64bit scores by MrRuslan · 2004-06-01 02:23 · Score: 2, Insightful
  
  The 386 did not run at a slower slock speed and kick the 286's ass.the p4 is is 1 GHZ faster in the clock department and still can't beat the AMD64 that's what make's this more impresive.
15. Re:No 64bit scores by hackstraw · 2004-06-01 02:39 · Score: 2, Informative
  
  Be aware that this is a benchmark for jvm's, and has little to do with machine performance. According to the benchmarks results page. A PIV 3.0 GHz computer can score anywhere between 228 and 557. This benchark also seems biased towards 32bit machines. One thing that is essential for any test is for it to at least be internally reliable.
Too long. by Piranhaa · 2004-06-01 00:58 · Score: 5, Interesting

It's a really good idea AMD is finally making the transition to dual-channel non buffered memory. They really should have done this a LOT sooner, before consumers started getting adjusted to the other socket, so they wouldn't have to replace their board when upgrading to the newer chip.
1. Re:Too long. by swordboy · 2004-06-01 01:10 · Score: 5, Interesting
  
  so they wouldn't have to replace their board when upgrading to the newer chip.
  
  Who still does this?
  
  I've never upgraded without swapping both the processor and mobo. This isn't a troll but rather just curiosity - I'd like to know if there is any significant percentage of non-gamers that upgrade *only* the processor. I'dathunk that the "processor bottleneck" was just a myth.
  
  I actually underclocked my Athlon 1800+ because it ran much cooler/lower power without any noticeable decrease in average usage habbits. Now, moving from a 5400rpm hard drive to a 7200rpm unit was a huge upgrade. I can't wait for cheap, desktop-oriented 10k and 15krpm units.
  
  --
  
  Life is the leading cause of death in America.
2. Re:Too long. by sumdumass · 2004-06-01 01:16 · Score: 4, Insightful
  
  The AMD64 platform design was released several years late and amd had to honor obligation with manufacturers that had already invested time and effort in the r&d based on the specs it issued before the delays. Once most of these "proto-type" boards (or whatever you will call them) were reasonably sold out, they could go ahead with thier advancments they discovered along the line.
  
  It sounds like the consumer got the raw end, but in reality it allowed the motherboard and chipset manufacturers to have a sence of confidence and time to develope working designs for the newer socket. It is a win-win type situation except for those that bought with the intention of upgrading. But then again most people in that situation (like me) would have been reading about the new socket when researching whether or not to get one and made a decision to live with it. Others would be just getting it for a system that would last a while bfore buying another.
3. Re:Too long. by aelbric · 2004-06-01 01:24 · Score: 2, Funny
  
  You underclocked your Athlon 1800+?!
  
  Turn in your National Geek Association membership card at the door as you leave, sir! ;-)
  
  --
  nos laetus epulor qui would domito nos
4. Re:Too long. by pebs · 2004-06-01 01:36 · Score: 2, Insightful
  
  You underclocked your Athlon 1800+?!
  Turn in your National Geek Association membership card at the door as you leave, sir! ;-)
  
  Blah.. Underclocking is completely acceptable for geeks. Most overclockers on the other hand, are the rice boy equivalent for the computer world. Overclocking is hardly even considered geeky. Most geeks, especially professionals, will look down upon overclocking.
  
  --
  #!/
5. Re:Too long. by sumdumass · 2004-06-01 01:57 · Score: 2, Insightful
  
  what about the people building thier own computer and selected a smaller processor because it wass afordable and then plan on buying a better one when availible. I did this witha system i built for less then $300 The processor i wanted at the time cost almost that much. so i took an extra year or so to get one. With the socket A, i was able to get even a newer/faster processor then i originally wanted to.
6. Re:Too long. by JFitzsimmons · 2004-06-01 03:08 · Score: 2, Insightful
  
  However, unlike ricing a car, overclocking actually does make your computer faster... assuming you can stand the heat, and the increase in power usage, and quite possibly the instability (if it isn't done properly).
  
  --
  Beware he who would deny you access to information, for in his heart he dreams himself your master. -Anonymous
7. Re:Too long. by PhrostyMcByte · 2004-06-01 03:32 · Score: 2, Interesting
  
  I'dathunk that the "processor bottleneck" was just a myth.
  
  It doesn't really matter for non-gamers. My dad runs an XP2000+ and is able to browse the web/play mp3s without a problem. But for gamers it really matters: current games are almost completely CPU limited. Until some games come out that seriously stress the video hardware with shaders, all the new video cards can do for you is increase resolution and AA/AF. With a recent system you should get around the same FPS in UT2004 and FarCry until you hit high resolutions and AA/AF settings that stress fillrate. Physics, AI, and just generally pushing lots of geometry through AGP is a big frame limiter.
Ceramic vs Organic packaging by danormsby · 2004-06-01 01:00 · Score: 4, Funny

Anyone got a definition on what the "Organic" packaging on the new Socket 939 processors is and how it compares with the previous ceramic packaging?
I guess it isn't wicker based.

--
Omnis amans amens
Sockets again by tronicum · 2004-06-01 01:02 · Score: 4, Interesting

I wonder how often they will introduce new sockets for just a bunch of CPUs.
939 will not support dual CPUs, after all that "Slot A", Socket 7xx/9xx nonsens you cant just buy a board and hope to upgrade the CPU. They change the memory systems, introduce new bus systems (graphic : PCI->AGP->PCI-X/PCI-Express).
Anyway I like my Athlon64 and at least the TDP (Thermal Design Power) of the new CPUs does not rise....
1. Re:Sockets again by Rufus211 · 2004-06-01 01:56 · Score: 2, Insightful
  
  > after all that "Slot A", Socket 7xx/9xx nonsens you cant just buy a board
  
  Erm, what "Slot A nonsens"? Yeah, years ago AMD switched from Slot A to Socket A for very good reasons and has stuck with Socket A up until now. In fact my CPU (Barton 3200+) recently died and I threw in a Duron 1ghz to keep my board running until I got a replacement. Again, PCI->AGP made a hell of a lot of sense and was again years ago, as does the current PCI -> PCI-E. Now the only "nonesens" has been the 754>939 change for single to dual channel, but AMD has publicly stated this will happen the whole time, plus they'll still produce at least one more round for 754 and 940 sockets.
  
  So yeah, things change and your old stuff won't work any more, stop wining. At least it's not as bad as Intel (how many sockets have they had in the past few years anyway?).
2. Re:Sockets again by CTho9305 · 2004-06-01 02:01 · Score: 2, Informative
  
  As I understand it, you can't build a 4-layer motherboard with socket 940 (you have to use 6 layers). By changing the pin layout for socket 939, it is now possible to use 4 layers. This should reduce the cost of motherboards.
  
  --
  My server
Bah! by hendot · 2004-06-01 01:03 · Score: 5, Funny

Sounds like a ripoff to me. You pay $x for a new cpu and they don't even give you the full 940 pins :p
Tinfoil anyone? by Gothmolly · 2004-06-01 01:10 · Score: 4, Interesting

IANACD (I am not a CPU Designer), but I'd imagine that they're redesigning these things for a reason, NOT just to screw users and force an upgrade cycle. Intel did the same thing with their CPUs, and IBM/everyone did the same when they went from 30 pin to 72 pin SIMMS, then to DIMMS, then to DDR DIMMS. Was this all a vast Taiwanese component manufacturer conspiracy? I somehow doubt it. When it first came out, the PCI bus was limited to 3 slots due to physical 'ring' characteristics on the signal lines. Some propeller-heads at HP figured out a way to get 4 slots, and everyone ooh'd and aaah'd over it. Nowadays we have more slots due to bridge chips, are we going to complain that those pesky motherboard manufacturers keep updating their chipsets?
Are you also angry at the music industry cabal that forced everyone to upgrade from vinyl to 8-track to cassette to CD to DVD ?

Schernau's 2nd law: bolding part of your post actually detracts from your argument

--
I want to delete my account but Slashdot doesn't allow it.
1. Re:Tinfoil anyone? by Short+Circuit · 2004-06-01 01:45 · Score: 3, Insightful
  
  IBM/everyone did the same when they went from 30 pin to 72 pin SIMMS, then to DIMMS, then to DDR DIMMS.
  
  The switch from 30-pin to 72-pin SIMMS, and then on to DIMMS, occured as a result of the memory bus width increasing.
  
  When the normal width for a memory bank reached 64 bits, they went from 30-pin to 72-pin. (It was easier to insert one 72-pin SIMM than four 30-pin SIMMs.)
  
  When the normal width for a memory bank reached 128 bits, they switched from SIMMs to DIMMS. (Again, it is easier to insert one DIMM than two 72-pin SIMMs. Imagine what would happen if every RAM upgrade meant you had to insert eight 30-pin SIMMs.)
  
  However, the switch from SDRAM DIMMs to DDR SDRAM DIMMs was strictly for bandwidth, not for making memory insertion cheaper/easier. ("Cheaper" because one DIMM socket takes up much less board space than eight 30-pin SIMM sockets.)
  
  --
  tasks(723) drafts(105) languages(484) examples(29106)
2. Re:Tinfoil anyone? by Bedouin+X · 2004-06-01 03:42 · Score: 2, Informative
  
  740 was not released in the last year, it has been around forever as the Athlon XP socket.
  
  940 is for the high end. You can't put a Xeon in a P4 board. It's not going anywhere.
  
  754 was a stepping stone to 939 which is better and cheaper. Intel did the exact same thing when the P4 came out and they went from Williamette to Northwood (socket 423 to 478). It may be inconvenient to people who didn't do their research - socket 939 was known to be coming from day one - but it's not unreasonable or unprecedented. Also, Socket 754 will be supported for at least another year so it's not like they are totally abandoning it today. The 3700+ which is very comprable to the 3800+ is on 754.
  
  --
  Dissolve... Resolve... Evolve...
Anandtech review by Karamchand · 2004-06-01 01:10 · Score: 4, Informative

Review on Anandtech! (I like them for their print view:)
More review links by JaF893 · 2004-06-01 01:23 · Score: 5, Informative

Here are some more review links for those who are interested:
Tom's Hardware
Bit-Tech
Driver Heaven
AMD Zone
Hard Tecs 4U
PC Perspective
Ace's Hardware
Sudhian
Aceshardware Review by hattig · 2004-06-01 01:27 · Score: 5, Informative

AcesHardware found that disabling the 2T memory timing in the BIOS improved S939 performence by over 10%. The only limitation with this is one DIMM per memory channel.

A lot of reviews you read today will not be using this, and the results will therefore be significantly lower than what is possible.
List of Links to reviews / news by d3bruts1d · 2004-06-01 02:18 · Score: 2, Interesting

We've posted a list of links to reviews and news regarding the AMD64 939-pin processors. It's available here for anyone that wants to look at more information on the new CPUs. Currently 18 reviews, and the list is updated as they come in.
Compile performance! by IceFox · 2004-06-01 02:44 · Score: 4, Interesting

Something I personally look for and most of the time isn't included in any reviews I was presently surprised to find in anandtech's review a Quake compile test!
http://www.anandtech.com/cpu/showdoc.html?i=2065&p =12
It doesn't specify what compiler or platform was used, but at the bare minimum it gives a little glimpse of what you might be able to achieve. Now all you have to do is apply that to a price/performance graph to determine what and how many you want to buy.
-Benjamin Meyer

--
Do you changes clothes while making the "chee-chee-cha-cha-choh" transformation sound?
1. Re:Compile performance! by HarvardAce · 2004-06-01 03:05 · Score: 4, Informative
  
  From a previous guide, a brief description of their build process.
  As introduced with our Athlon 64 article, we are looking at compile times for the Quake III Arena source code. In running this test, we compiled the source for both Q3A and Q3TA in both debug and release modes using the Batch Build feature of Visual C++ 6.0. The build was run three consecutive times for each processor (in order to try to reduce file read time impact), and we used the third compile time.
  
  --
  Note to self: Stop putting jokes in my insightful comments so I can get something other than +1 Funny!
Question: High temp issues? by geohump · 2004-06-01 03:51 · Score: 2, Interesting

I heard that Intel CPU's used to automagically reduce their clock rate when their temperature got too high.

This feature caused some businesses I know to forgo choosing any AMD cpu, since it couldn't protect itself in the event of an unattended fan failure.

First question - does the clocking down feature really exist on Intel CPU's? and second question, Does AMD have this feature yet?
1. Re:Question: High temp issues? by geohump · 2004-06-01 06:23 · Score: 2, Interesting
  
  I'm using the AMD CPU on my personal system right now. If I were trying to generate FUD I would have just blasted away instead of asking for information.
  
  According to my limited stab, googling for info, it looks like AMD still doesn't have the down-clock protection on its CPU's. Thats too bad.
  
  I was hopinfg that someone would point me at info saying otherwise.
  
  I would really like to be able to use them in my client installations but I can't really recommend them for anyplace that doesn't have 7x24 staff within hearing distance of any audible thermal alarms.
  
  I did learn that there are some MB's which will try to save the AMD CPU from cooking itself and that is encouraging. Still - I won't be recommending AMD for Aunt Tillie or my small customers until the AMD CPU's or virtually every AMD capable MB out there has some way of protecting the AMD CPU's from self destruction.
  
  That way I will get fewer "middle of the night" emergency calls.
  
  Interesting that you posted anonymously.
Re:No 64bit scores (and CPU cost ...) by hobbs · 2004-06-01 05:07 · Score: 3, Informative

Out of curiosity, I tied these scores to CPU prices as listed at http://www.pricewatch.com/: CPU SCORE US$ Athlon64 3200 64: 523.70 $255 Athlon XP2700: 467.15 $ 80 Athlon64 3200 32: 449.07 $255 Athlon XP2600: 448.42 $ 71 Pentium4 3.0GHz: 387.57 $203 Athlon 1400: 305.26 $ 97 AMD Athlon 950: 209.51 $ 69 Sparc 500MHz: 52.21 ??? Sparc 440MHz: 51.89 ???
Re:Any concrete guides how to build a 64-ONLY bit. by Wesley+Felter · 2004-06-01 07:02 · Score: 2, Informative

Sure, I've done this.

1. Buy whichever AMD64 processor you like.
2. Buy a motherboard that supports that processor.
3. Buy all the other parts (RAM, hard disk, graphics card, etc.).
4. Put it together.
5. Install Fedora Core 2.

Personally I used an Athlon 64 FX-51, an ASUS SK8N, Corsair low-latency RAM, and a Radeon 9800 Pro (if you're not playing games, get a cheaper video card).