AMD Athlon64 Socket 939 FX53 and 3800+ Review.

Introduction.

Athlon64 Socket 939 is finally here. Many of us have been watching the roadmaps and following AMD’s Athlon64 CPU progress from the start. Some of us put off buying new hardware waiting for faster chips, new revisions of chipsets and so on. Others (including myself) were blow away by the initial performance gains offered by these chips and their integrated memory controllers. Especially, relating to gaming benchmarks. However these performance gains go much further than just gaming and also offer excellent results in daily applications such as WinZip as you can see in the benchmarks below.

What is Socket 939 all about?

There are basically 3 of these processors available at the moment. The FX53, 3800+ and 3500+. I still see a lot of people still asking, but doesn’t the FX53 require registered memory? There are two FX53’s available. The one is a Socket 940 CPU (which does require registered memory) and the other is Socket 939 and does not require registered memory. This is also true of all the Socket 939 CPU’s. Both are Dual Channel CPU’s though. There will be a FX55 (2.6 GHz) released later in the year as well according to the roadmaps.

AMD speed ratings.

I have to talk about this a bit as I find it very interesting. In this review I am able to compare the current two fastest Socket 939 CPU’s to the fastest Socket 754 CPU. The interesting thing is they all operate at 2.4 GHz clock speed. The difference between them is as follows:

The Athlon64 Socket 754 3700+ has 1MB Level2 Cache and has a Single Channel Memory controller.



The Athlon64 Socket 939 FX53 has 1MB Level2 Cache and has a Dual Channel Memory controller.



The Athlon64 Socket 939 3800+ only has 512KB Level2 Cache, but is Dual Channel as well. As you can see CPU-Z is not able to read the stepping yet.

Now I for one am most curious to see how the 3800+ with less cache and Dual Channel is theoretically rated faster than the 3700+ with more cache and Single Channel. On paper though the FX53 should blow the others away in performance, but does it? We shall see.

Notes.

Some things I think you should keep in mind while looking at the benchmarks. Firstly, the motherboard I am using is basically a review board and not a retail one. Secondly, it was supplied with its own Corsair memory in Dual Channel setup. This memory is low latency, but cannot run RAS to CAS of 2, only 3. To my surprise, by replacing the Corsair with my own Mushkin 2-2-2 modules I saw no performance increases what so even! My only conclusion is that this setup has been matched to work together and I was impressed with the performance offered while running at 2-2-3 timings. Also there doesn’t seem to be a newer BIOS available and the current BIOS is still dated February 2004. I would be most curious to see how a retail board, running my memory at 2-2-2 would compare. I suspect it would be a little faster even.

All FX53 / 3800+ benchmarks were run with memory timings of:
CAS Latency – 2
RAS to CAS – 3
RAS Precharge – 2
TRAS – 8

FSB - HT 16-bit / 1GHz upstream and downstream.

MSI VIA K8T800Pro motherboard.



As mentioned above this is not a retail board and as such comes with no manual. It is not even on the MSI website yet. I will give some details quickly though. It has a black PCB and is similar to most of the new board’s i.e. SATA RAID, IDE RAID, two pairs of DIMM slots for Dual Channel setup, 4 onboard USB ports, LAN, fire wire, SPDIF out and what appears to be 5.1 sound. In fact it is a very well featured board. As for BIOS the max volts for memory is 2.85 and VCore is 1.7V. FSB goes up to 280 and no multipliers.

One thing to note though. Same as with my newer Gigabyte BIOS, there is a 2T function under the Advanced Chipset Features with regards to memory. Now by default this is Enabled. Leaving it enabled though produces a huge drop in Memory Bandwidth performance. To illustrate, with 2T enabled the Memory Bandwidth was 4890 in Sandra Int Buffered. But, with 2T disabled it produced Bandwidth of between 5970-6019. Needless to say I Disabled it.

Finally, the BIOS kept reporting the CPU temperature as 56’C while idle. Even after replacing the stock AMD HSF with the a monster Gigabyte cooler and some Artic Silver Ceramique the temps remained at 56’C. The same was true with the 3800+, thus I’m sure readings are incorrect.

This post has been edited by solo: Jun 20 2004, 11:08 PM

The setup.

AMD Athlon64 Socket 939 FX53 (2.4 GHz) 1MB Level2 Cache
AMD Athlon64 Socket 939 3800+ (2.4 GHz) 512KB Level2 Cache
MSI VIA K8T800Pro MS-6702E Motherboard
Corsair TwinX XMS3200LL DDR SDRAM at 400MHz (2x512MB DIMMS)
Mushkin Level 2 Black DDR433 2-2-2 (Used in 3400+ and 3700+ benchmarks)
Gigabyte 3D Cooler Pro PCU21-VG
Antec 400 Watt PSU
Powercolor ATI Radeon 9800 Pro

Software
WinXP Pro with SP1
VIA Hyperion 4in1 – 4.51 Drivers
Catalyst 3.10 Drivers
Catalyst 4.5 Drivers
DirectX 9.0B

The Benchmarks.

Sandra 2004 with SPI (Version Sandra2004.SP1-9104)

CPU Arithmetic Benchmark - Dhrystone ALU
Socket 754 3400+ = 9581
Socket 754 3700+ = 10491
Socket 939 3800+ = 10416
Socket 939 FX53 = 10416

CPU Multi-Media Benchmark – Floating Point iSSE2 it/s
Socket 754 3400+ = Only saved SSE result.
Socket 754 3700+ = 24785
Socket 939 3800+ = 24595
Socket 939 FX53 = 24601

Memory Bandwidth - Int Buffered iSSE2
Socket 754 3400+ = 3071
Socket 754 3700+ = 3104
Socket 939 3800+ = 6000
Socket 939 FX53 = 6019

These may be synthetic benchmarks, but take a look at the CPU Arithmetic scores. The 3800+ and FX53 genuinely score the same. I even re-ran them and they gave same results each time. Very interesting. The memory throughput here from the Dual Channel setup is almost mind blowing. It hovered between 5970-6019 and as you will see further on some benchmarks love the extra bandwidth. I have a suspicion that a newer patch might be required for Sandra to enable better Socket 939 measurements though.


3DMark2001SE (330)



In 3DMark2001 the Dual Channel FX53 is running about 500 points ahead of the Single Channel 3700+. The 3800+ and 3700+ are running neck and neck . 1MB L2 Cache versus Dual Channel. Something else to note, which I think naughty mentioned as well. The Catalyst 4.5 drivers are good for benchmarking.

3DMark03(340)



Here the processors are running fairly close to one another, with only a 140 point difference between the 3400+ (754) and FX53 (939). But, as well all know 3DMark03 is mostly about GPU performance and not system performance as in 3DMark2001. Again the Catalyst 4.5’s are shining.

Aquamark 3



These figures seem to confirm something that I said once before. Aquamark seems to love the extra memory bandwidth. I pointed this out once before in another thread. The Pentium 4’s tend to do well in this benchmark and I’m thinking that those high FSB’s and their massive bandwidth when overclocked are doing the trick. Just look at the increase in this benchmark between the 3700+ and 3800+. 1366 Points is a huge increase. I recon a Dual Channel 3800+ / FX53 paired with a Nvidia 6800 Ultra will make a killer combination.

Super PI (1 Million) in seconds.

Socket 754 3400+ = 40
Socket 754 3700+ = 36
Socket 939 3800+ = 35
Socket 939 FX53 = 35

I was surprised to see the FX53 and 3800+ both scoring same times here. The 3800+ did beat the 3700+ though.

Cinebench 2003 v1

Rendering (Single CPU)
Socket 754 3400+ = 312
Socket 754 3700+ = 341
Socket 939 3800+ = 339
Socket 939 FX53 = 338

Shading (CINEMA 4D)
Socket 754 3400+ = 370
Socket 754 3700+ = 402
Socket 939 3800+ = 404
Socket 939 FX53 = 407

Shading (OpenGL Software Lighting)
Socket 754 3400+ = 1687
Socket 754 3700+ = 1846
Socket 939 3800+ = 1876
Socket 939 FX53 = 1895

Shading (OpenGL Hardware Lighting)
Socket 754 3400+ = 3289
Socket 754 3700+ = 3495
Socket 939 3800+ = 3886
Socket 939 FX53 = 4013

ScienceMark 2.0

Memory Bandwidth
Socket 754 3400+ = 3007
Socket 754 3700+ = 3011
Socket 939 3800+ = 5735
Socket 939 FX53 = 5792

Molecular Dynamics
Socket 754 3400+ = 93.6 seconds
Socket 754 3700+ = 84.3 seconds
Socket 939 3800+ = 75.2 seconds
Socket 939 FX53 = 76.1 seconds

Primordia
Socket 754 3400+ = 389 seconds
Socket 754 3700+ = 356 seconds
Socket 939 3800+ = 347 seconds
Socket 939 FX53 = 346 seconds

Basically I have started to include the ScienceMark benchmarks as not everyone is just interested in gaming benchmarks. The Athlon64’s score very well when it comes to raw number crunching as well. Especially mathematical and scientific calculations.

Winzip 8.1 SR-1 (5266)
Compression = Default
Compress - 6549 Files, 136 Folders
Size - 482 MB
Source - i386 Folder (WinXP Pro SP1 CD)

Socket 754 3400+ = Compress (2 min 16 sec) Decompress (1 min 31 sec)
Socket 754 3700+ = Compress (2 min 06 sec) Decompress (1 min 23 sec)
Socket 939 3800+ = Compress (1 min 59 sec) Decompress (1 min 01 sec)
Socket 939 FX53 = Compress (1 min 46 sec) Decompress (1 min 07 sec)


WinRAR 3.30
Compression = Best
Compress - 6549 Files, 136 Folders
Size - 482 MB
Source - i386 Folder (WinXP Pro SP1 CD)

Socket 754 3400+ = Compress (5 min 25 sec) Decompress (1 min 52 sec)
Socket 754 3700+ = Compress (4 min 51 sec) Decompress (2 min 10 sec)
Socket 939 3800+ = Compress (4 min 20 sec) Decompress (1 min 38 sec)
Socket 939 FX53 = Compress (4 min 10 sec) Decompress (1 min 47 sec)

In both WinRAR and Winzip the extra memory bandwidth seems to be making a big difference with the FX53 shaving 41 seconds off the 3700+ in Winzip. These are the type of applications we use daily and as such very relevant benchmarks.

LAMEVersion 3.96
Command Line: -b 128 -m s -V 4 -B 320
Front End: RazorLame 1.1.5
Wave Source: U2 - The Joshue Tree

Socket 754 3400+ = 5 min 49 sec
Socket 754 3700+ = 5 min 17 sec
Socket 939 3800+ = 5 min 17 sec
Socket 939 FX53 = 5 min 18 sec

Conclusions.

Well I think the figures speak for themselves. Does the 3800+ deserve it’s “3800” rating over the “3700” rating of the 3700+ ? All the benchmarks above (except the synthetic Sandra ones) indicate that it does. The Dual Channel setup even with half the Level2 cache is balancing out the performance and even putting it slightly ahead.

If you were putting off buying the first generation of chips in favour of these then the wait might have been worth it. However, as you can also see the top Socket 754 processor still holds it’s own in many of the benchmarks. The deciding factor for many will be price. The FX53 is to the AMD’s what the Extreme Edition is to the Pentiums. Extreme performance usually comes at a price. But, with alternatives such as the 3800+ and 3500+ available there are other options.

If you currently have a Socket 754, it is worth upgrading to the new Socket 939? That is a good question and I would answer it like this. If you are looking for the that extra 3 – 10 % (depending on benchmark) boost in performance and can afford it, go for it. If you have never own an Athlon64 and are thing of getting one then a Socket 939 motherboard and 3500+ CPU will be an awesome upgrade, but so will a 3400+ or 3700+ and Socket 754 motherboard. And should cost substantially less now that the newer chips are out.

Socket 754 and Socket 939 should still be around for a good few months. From reports the 32 Bit 754 and 939 Sempron’s will start appearing later this year prolonging the life of the current boards.

Unfortunately these benchmarks were run with a Radeon 9800 Pro, not truly showing the full potential of these CPU’s in a high end system. Hopefully, I will be able to get my hands on a decent graphics card during the week and then report back with some really mind blowing figures. Hint, hint ….

Pro’s:

- Extreme performance.
- Big increase in performance in applications that rely on the extra memory bandwidth.
- 64 Bit Operating System compatible now and in future.

Con’s:

- Price (Top CPU currently retails for around $825 in the States.)

Rating.

9/10

This post has been edited by solo: Jun 21 2004, 12:44 AM

Will start reading now, is this your own review?

Wow, brilliant review.

I must say that I'm not really impressed with the FX53. Maybe I was expecting too much, considering its pricetag.

I'll stick with 754 for a bit longer :>

Yeah, the FX53 seems like nothng special, theres hardly any difference between itself and the 3800+. What are the prices for all these CPU's. Solo where did you get hold of this equipment?

awesome review solo

seems this chip is great, bar the price (which is the case far too often)
keep it up

nice byron - wish i had these chips and that mobo to play with

me is now jealous

but anyways as long as you keep reporting what you find im sure it can help the rest of us to decide what to buy so great work there

and yeah it bears out what ive been reading all over that dual channel is not doing amd very great favours - so socket 754 is still seeming good enough for most buyers

Thanks for the nice review, what are the prices on these chips anyway? or have I missed something?

Nice one Byron. Vey Nice indeed

Now I'm very confused. All these cpu's are very fast (much faster than any pc I've had on me. Overclocked or not)

Hopefully my GFX card will find a home sooner rather than later...

Well Nitin I can assure you if you were in Cape Town I would let you come "play" with these goodies as well ... Nitin packs his bags

Yes G-Force this is 100% my own review, except fro the picture of the motherboard. Graphs = Microsoft Excel. The same kit goes out to reviewers world wide hense some figures are identical, which I only noticed after I did some browsing after publishing this. Only thing is I have to return it in a weeks time.

I see TomsHardware also mentioned the 2T timing thing in their review. But, they actually got a new BIOS from MSI which disabled the 2T timing by default.

Thanks for the comments guys. It's really encouraging to know you guys take the time to read my stuff and encourages me to so more next time.

This post has been edited by solo: Jun 20 2004, 11:00 PM

Said that because tomshardware's graphs are very similair (I think) and the review looks so professional

&


MB! haha, i thought i missed a few hundred years when i read that! you must be thinking of RAM when you wrote that (habit i guess), great write-up solo, keep it coming!



oh please do keep writing, you're in a very fortunate position of being able to accses hardware like this (even if it is for a short time), and its important for you to at least even if it is sparsely to write somthing up and then post it somewhere for the rest of us who for the meantime can do nothing but window shop, AMD it seems is on an up and up (just like their pricing) ..

Oops, sorry about that hikizume. 512MB L2 Cache would be nice though

Damn Byron !!! You got busy after I left on Saterday. Keep up the good work.

i totally agree, its nice to see local reviews of high-end equipment conducted by someone with excellent technical expertise providing meaningful info instead of the usual marketing drivel

keep it up byron

This post has been edited by enigma: Jun 21 2004, 12:33 AM

Here is an update on the FX53.

There is a new version of Sandra available.
Sandra SP2 Ver. 2004.7.9.129

Old CPU score was 10416.
New CPU score is 11021.

Multi-Media score stays the same.

Memory Bandwidth is now constant at around 6040.

Overclocking

This non-retail board is messing me around to be honest. The temps are not right.
Had there been multipliers in BIOS I could have pushed my memory bandwidth some more. I have tried all versions of Clockgen on this motherboard, but none of them allow for changing the FSB only the multi-pliers. I put in my Mushkin and upped the FSB to 220 Mhz. I got a nice Sandra Memory Bandwidth score of 6553

Now this is one of the reasons that the FX53 commands such a premium price. The multi-pliers are totally unlocked. Unlike the normal Athlon64 where you can only decrease the multipliers, with this beauty you can increase them as well.

And the proof.

Normal multi-plier = 12. Here it is upped to 13x200 = 2.6 Ghz.
SuperPI 34 seconds (Memory only running at 200 Mhz).
No extra VCore !.

Wow, you are really clued on everything eh? This review is rather professional, would've mistaken for something you'd see at anandtech etc.

Maybe you should setup a website to start this type of thing! You can get a host of like-minded guys to send you reviews and you could upload it! A south african hardware review site, that would be cool! Or if ur too lazy to host a site maybe you could ask some guys here to do it for you!
Well done dude!

Hmmm Dual channel, and unlocked multipliers, Hmmm.

Now me needs a new PC, the FX53 is looking really inviting (Along with something like Dual 6800 Ultras - Come on lucky numbers, win me the lotto)

Right ! As I said before benchmarking this beauty with a Radeon 9800 Pro will hardly show off it's true power. Finally, with the help of Firestar's Leadtek Nvidia 6800GT at stock speeds and an FX53 at "only" 2.6 HGz ............



http://arc.aquamark3.com/arc/arc_view.php?run=212994878

CAN YOU FEEL THE POWER NOW !!!

This post has been edited by solo: Jul 14 2004, 11:46 PM

Its about time you got the supplier's moneys worth out of that thing

Nicely done.