• Welcome to Overclockers Forums! Join us to reply in threads, receive reduced ads, and to customize your site experience!

Athlon64 3200+ DTR Initial Results

Overclockers is supported by our readers. When you click a link to make a purchase, we may earn a commission. Learn More.

Gautam

Senior Benchmark Addict
Joined
Feb 4, 2003
Location
SF Bay Area
Looking through the AMD sections on several forums, we see hundreds of threads daily regarding mobile Bartons, 1700+/1800+ DLT3C's, DFI Lanparties, Abit NF7's; anything and everything to do with the Athlon XP. However, it seems that information and even interest on AMD's latest and greatest, the Athlon74, is completely nonexistant. When I was planning my system, just about everyone's sentiment was to wait, and buy a mobile Barton in the meantime. What seems to be the public opinion is that A64's are overpriced, difficult to overclock, and have flaky hardware and software support at best. Hopefully I can put an end to these myths.

The processor I picked was the 3200+ DTR, as, according to the description at Newegg, it would be a CG revision, unlike the desktop versions, which to date appear to all be of C0. A newer stepping generally means more overclockability, and considering that this is the same revision as the 2.4GHz FX-53's, one would think it should fare pretty well. DTR stands for Desktop Replacement. It is not a true mobile processor. Furthermore, it is not unlocked beyond a 10x multiplier, and no Athlon64, not even a true mobile, is. The only real difference other than the revision is that the DTR lacks a heatspreader, which supposedly helps temps by a couple of degrees.

The motherboard I chose is the Gigabyte GA-K8N Pro, which uses the nVidia nForce3 150 chipset. Arguably, the most popular board choice is the Shuttle AN50-R, however, I opted against this for the reason that it seemed to be responsible for plenty of A64 deaths, even without excessive vcore or vdimm. Users that had it reported it to be difficult to deal with, and that it required many BIOS swaps, etc. Before I invoke the wrath of Maxvla, I'll say that for every failure of this board there seems to be a sucess story. However, Shuttle was simply a brand I didn't trust. My top two choices for motherboards are always Abit and Asus, but unfortunately, they do not have any motherboards out based on the nForce3 chipset, each utilizing the VIA K8T800 chipset, which lacks an AGP, which is crucial for my Radeon 9800. The only solid brand(IMHO) that had a board based on the nForce3 250 was Gigabyte, and the price, at $120, and loaded to the brim with goodies, was right.

I used the Thermalright SLK948U heatsink, as watercooling was beginning to become too LAN-unfriendly for me. The heatpipe-equipped SP-98 is around the corner, supposedly, but with the ridiculously high pricetag generally placed on the SP line, I most likely would have opted against it, even had it been out at the time. The heatpipe does not make a significant difference in cooling ability, considering the design of Thermalright's heatsinks. The processor, being a DTR, lacks a heatspreader, thus requires the heatsink to be mounted slightly lower than a desktop. I added washers while mounting the heatsink as a precaution, but I think that this was unnecessary with the nature of the SLK948's mounting mechanism.

The only two PCI devices that I am running are a Promise FastTrack S150 TX2Plus Serial ATA controller, and a Soundblaster Audigy2. In order to avoid hard drive corruption that results from high PCI buses, a controller card supporting 66MHz PCI buses is required. This card fits the bill perfectly, and allows very aggressive overclocking without issue. The Soundblaster also can handle PCI buses in excess of 50MHz.

Unfortunately, I'm still using my old PC3200 memory, which tops out at about 228MHz.

So, how does all of this hardware actually do?

After messing around with it a bit:

2508.JPG


Core voltage at 1.7v in BIOS, reported to be 1.74.

The memory bus is running asyncronously to the HTT bus, at only 210MHz. I'm keeping it conservative for now, but I'll be getting memory that should be able to handle much higher speeds this week. So far, it seems 100% stable. The only thing that I can not attest to ATM is Prime95; I'll do a thorough run in a few hours. However, it's handled several hours of benchmarking, gaming, etc, at this speed.

arith.JPG


Effortlessly trounces all the reference results, even though Sandra really isn't its forte.

3D performance is where it really shines. So far, I've gotten 22796 3DMarks in 3DMark2001SE. The asyncronous memory is really holding it back. Should be able to do over 1000 marks higher 1:1. With everything at stock(9800 Pro stock speeds) it got quite a respectable 19506. At stock speeds, it outdid my T-Bred at 2.6GHz by close to 1000 points.

I know, not too far from what you've expected. After all, we've all seen the benchies. We all know that it tops most processors in synthetic performance, but what does this mean in reality? I can honestly say that it was worth every last penny. If the A64 looks great on paper, it feels spectacular in real-world performance. Everything runs smoother than ever; games look breathtakingly quicker. And this is at stock speeds. I've got a couple of friends who upgraded from P4C's to A64's. When they said that it loaded things quicker, and "felt" faster, I thought they were just full of it. However, not even the benchmarks can show how much nicer it really feels. An AthlonXP, at just about any speed, cannot touch an A64. You'd need to be at about 3.1GHz with a mobile Barton to compete with a 2.5GHz in synthetic benchmarks, and I can guarantee that even at those speeds, it will still lack the smoothness of an A64.

The fact of the matter is, you can take a retail A64, load it with generic CAS2.5 memory, run it at stock, and still have it trounce even aggressively overclocked mobile Bartons, and even P4C's in synthetic performance, and by even more in real-world performance. The revisions due in the future (dual-channel support, PCI locks, etc.) most likely will make a very small difference. IMHO, this is not the right time to invest in an AthlonXP or Pentium4 setup. The A64 is in a totally different league. It doesn't even need the slightest bit of overclocking for its performance level to be realized. Futhermore, as an overclocking processor, the A64 DTR is outstanding. My 1700+ DLT3C required 1.95v on watercooling to accomplish 2.5GHz. This processor effortlessly makes it that far on aircooling with only 1.75v.

Some notes:
The process to get this far was not as smooth as one would have wished. For some reason, the processor defaults at 1.1v, but still at 2 GHz. I spent close to two hours trying to find out why it couldn't boot, swapping HD's, etc, only to find that it was running at such a low voltage. Bumped it up to 1.5v, and it was all good.

As I began to overclock, I noticed that Sandra and just about everything non-3D ran perfectly. However, running 3DMark or a game would cause the system to hang in 10 secs even with a modest overclock. I tried running at 250x4, and the same happened. What was causing this was that Clockgen was automatically raising the AGP bus as I changed the HT bus. Frankly, I was impatient and reckless, so I failed to notice this for a while, which also wasted about an hour. As soon as I took note of this, though, overclocking was as smooth as butter.

Is the A64 worth it at this time? Pass your own judgement. :D
 
First off Nice post Guatam Glad to see your results.And I might add your selection of hardware was very wise. The giga K8N series is a great choice for this platform, and the raid card perfect for the OC. As for your OC 2500 on air is great. Lots of the regular A64 3200+ top out between 2300-2400 on air so this is nice to see.

Nice Work :thup:
 
Hey, d]g[ts, thanks a ton for your help, bud. Couldn't have made it this far without ya. :clap:
 
"The only two PCI devices that I am running are a Promise FastTrack S150 TX2Plus Serial ATA controller, and a Soundblaster Audigy2. In order to avoid hard drive corruption that results from high PCI buses, a controller card supporting 66MHz PCI buses is required. This card fits the bill perfectly, and allows very aggressive overclocking without issue. The Soundblaster also can handle PCI buses in excess of 50MHz."

could you give me a link to that controller card? i am going to building a A64 soon and i think i might be getting alot of the same components as you, thanks
 
<Note to self>

....Buy Mobile FX (Socket 939):drool:
___________________________________

And to think that 754 is gonna be a "Low-end" platform:eek:

I cant wait till 939..... 939 with pwnz teh Int3L j00z:cool:

I cant even begin to imagine how well those suckers will OC!
 
Congrats and nice review.

It is nice to have people like you to voice different opinions and have the idea and courage to try out new hardwares and combinations, ..., especially they may not be the main consensus at a given time, ....

Experimenting new ideas and things are very important and they add to our knowledges, ....

Best.


BTW, what are the cache and memory bandwidth numbers (e.g. from Sandra), ....
 
devilscow22 said:
could you give me a link to that controller card? i am going to building a A64 soon and i think i might be getting alot of the same components as you, thanks

http://www.newegg.com/app/ViewProductDesc.asp?description=16-102-022&depa=1

It's got two serial ATA channels and one parallel ATA channel, and supports striping and mirroring. A very nice price for what you get.

Sentential said:
<Note to self>

....Buy Mobile FX (Socket 939):drool:
___________________________________

And to think that 754 is gonna be a "Low-end" platform:eek:

I cant wait till 939..... 939 with pwnz teh Int3L j00z:cool:

I cant even begin to imagine how well those suckers will OC!

IMHO, there won't be a considerable difference in performance between 754 and 939. If anything, I'd venture that 939 would actually turn out a little worse because of the lack of cache(except in the FX's of course). 940 is a loose indication of this. The memory bandwidth doesn't have much of an effect in general everyday performance.

hitechjb1 said:
Congrats and nice review.

It is nice to have people like you to voice different opinions and have the idea and courage to try out new hardwares and combinations, ..., especially they may not be the main consensus at a given time, ....

Experimenting new ideas and things are very important and they add to our knowledges, ....

Best.


BTW, what are the cache and memory bandwidth numbers (e.g. from Sandra), ....
Thank you. Feels great to hear that coming from you. :)

The memory bandwidth results in Sandra aren't too great. I'll put up a screenshot in a few minutes, but at those speeds, it got 2960 MB/sec. Rather disappointing considering the 500MHz effective data rate of the HT bus, and the 420MHz rate of the memory bus. Just 88% efficiency. I'll do some more tweaking and see how it goes. But as I said before, even though the synthetic numbers may not look too nice, its real-world performance is excellent. The extremely low latency due to the on-die memory controller offsets the low bandwidth very well.
 
Great review - now I know what to play with next ;) Odd that the memory scores are so low though. I was getting 6300+ with my FX51 on the Asus. Well this is not dual channel though correct?
 
I think we should not on one hand based solely on benchmark numbers, and on the other hand focus the discussion on subjective terms like smooth feeling, ....

The tradeoff and equivalence of cache latency, cache size and memory bandwidth have been studied to great extent, .... Their average and statistical relationship and impact on performance have been welll analyzed, ....

- Cache latency is the time or number of cycles to wait for getting the first data from the cache after a read command is issued.

- Cache size, the size of cache in bytes (for data/instruction), relates the probability (hit ratio) that data can be found in a cache. The larger the cache, the higher the chance to find the data in the cache. If data is not in L2 (cache miss), then data has to be looked for in the next stage cache or main memory (aka L3). Memory can be treated as the next level cache (L3) for L2.

- Memory latency is the time or number of cycles (many more cycles compared to L2 latency) to wait for getting first data from the main memory after a read command is issued.

- Memory bandwidth is the number of bits per sec (usually in MB/s) transfer to/from the memory after a transfer has started (i.e. after the latency).

Studies have shown that twice the size of L2 would translate into about 5% of average system performance, over a wide range of prorgrams (some benefited more and some less). This conurs with what we have been talking about the XP w/ 256KB L2 and the 512KB L2 Barton.

939 has twice the max memory bandwidth of a 754 due to the 128-bit memory bus in 939. This is in additon to the dual channel memory controller in 939. For applications with well structured, large data such as large matrices in scientific computations, video decoding, image processing, ..., these applications would be benefited directly from the 128-bit memory bus of 939 (vs the 64-bit of 754), ....


Studies have also shown that there is equivalence between cache size and cache latency, i.e. using larger memory/cache to tradeoff memory/cache with larger latency (not relevant here).
 
Last edited:
ok one last question about those controlers, i have one 80 gig HD and one 40 gig HD, i do not have then in a raid aray is there a cheaper card that would prevent the data curruption but support my two HHDs.

sorry my knowledge of harddrives is pretty much limited to knowing they store data, thanks

if i got 256x2 of muskins 222 special would i be able to run it on the K8N?
 
Great review - now I know what to play with next Odd that the memory scores are so low though. I was getting 6300+ with my FX51 on the Asus. Well this is not dual channel though correct?
Correct, the lack of a dual channel memory controller holds the S754-based A64's in memory bandwidth. However, my scores are looking low even for single channel.

I think we should not on one hand based solely on benchmark numbers, and on the other hand focus the discussion on subjective terms like smooth feeling, ....
Yep, although both deserve consideration.

939 has at least twice the memory bandwidth of a 754 due to the 128-bit memory bus in 939. This is in additon to the dual channel memory controller in 939. For applications with well structured, large data such as large matrices in scientific computations, video decoding, ..., these applications would be benefited directly from the 128-bit memory bus of 939 (vs the 64-bit of 754), ....
I concur. If you require a system to perform video editing or very math intensive tasks, the 939 would probably be a better bet for you. However, for most tasks, the 754 fits the bill just fine.

ok one last question about those controlers, i have one 80 gig HD and one 40 gig HD, i do not have then in a raid aray is there a cheaper card that would prevent the data curruption but support my two HHDs.

The Ultra100 TX2 (Or 133 if you have ATA133 drives) is cheaper, but hardly more economical. I'd say the extra $30 for the sake of upgradeability is well worth it, as a RAID 0 array helps a lot in making things feel slick, and cuts loading times in half.

Mushkin 222 should work great with the K8N. You'd have to run it asyncronously, but the tight timings are what the memory controller seems to love.
 
Incredible work Guatam!

This is really promising, and really gives us something to look forward to, as I'm sure most of us have Athlon 64 plans for the future (however far off it may be for some of us), of some sort or another ;) ;)

Which fan are you using with the 948-U, out of curiosity? I'm also wondering about the overvolting options that Gigabyte board has at stock, and whether you're overvolting the chipset, or running it totally stock. 250 X 10 is an incredible overclock, even moreso with the amount of Vcore you're using, and the with the fact that you're using air.

As for S939 Sentential, I highly doubt the pricing will ever be low enough to justify it's use in a proper rig; as illustrated above a S754 system thrashes pretty hard, and looks to be a lot of fun to play with, given the apparant abundance of headroom over stock with the right hardware.

Posted by Guatam

At stock speeds, it outdid my T-Bred at 2.6GHz by close to 1000 points.

I have a similar 9800 Pro / 2600 MHz Thoroughbred setup running right now, awaiting another Mobile Barton to try out; your 3DMark score is clearing my machine fully overclocked by a signifigant amount as well!
 
This is really promising, and really gives us something to look forward to, as I'm sure most of us have Athlon 64 plans for the future (however far off it may be for some of us), of some sort or another

Thanks, dude. Your cooling setup should complement an A64 wonderfully. ;)

Which fan are you using with the 948-U, out of curiosity? I'm also wondering about the overvolting options that Gigabyte board has at stock, and whether you're overvolting the chipset, or running it totally stock. 250 X 10 is an incredible overclock, even moreso with the amount of Vcore you're using, and the with the fact that you're using air.
I'm using a 10 yr+ Innovative fan. Couldn't find any specs of it online. It seems to put out somewhere between 30-40 CFM, definitely nothing spectacular. The mobo claims I'm at 37 load, but I think it's just underreporting ridiculously. I'll be getting a Torin centrifugal blower on Tuesday, which is comparable to a Tornado. It should help some. I'm not even sure if I'm maxed out as it is.

The Gigabyte manual claims that up to 1.85v is selectable, however, in reality it only offers 1.7v, and overvolts by about .05v. I didn't overvolt the HT voltage at all for this. I've heard that it isn't too beneficial, anyways. I did lower the HT from 3x to 2.5x. The stock HT speed is 3x200=600MHz, and I have it at 2.5x250 for 625MHz. From what I've seen, it can't go much further, but I'll try raising the voltage and see if it can get further. If you keep the HT at around 600MHz, there usually isn't any need to increase voltage.
 
Do you think you've pushed it to the end of its ability yet? I was hoping for a little more considering this is a new stepping. I was getting 2450 out of my old FX51 which was one of the early production runs last year. It hit 2900 with a vapochill and 2600 on ice water. I hope this new stepping can muster more than 50MHz higher.
 
No, not sure at all. 250x10 was just a nice round number. :D

I haven't even come close to pushing it to its limits yet. Bear in mind that I'm using a Gateway-pull fan from the early 90's on it, nothing more. :D

Although those FX51 results are extremely high. 2.9GHz is most certainly not normal. 2.8 on modded Prometeias seem more the norm, and 2.6-2.7 for Vapo PE's, from what I've seen. Sounds like you had a gem for its time, but things only seem to be getting better.
 
Very nice results Gautum, I'm glad to see you are happy with your move. :)

I can only hope that I will be as satisfied when it comes time for me to go to 939.
 
Gautam said:
The memory bus is running asyncronously to the HTT bus, at only 210MHz.

Why is this? Async memory is never a good thing if you can help it, right?

I've noticed your bus speed is 250mhz. Shouldn't it be 800mhz? Edit: or is it 600mhz with an nforce3-150?

With an out of spec fsb, are you having problems with onboard stuff? sound, lan, etc..

Sheesh, all this new athlon64 stuff is confusing. I feel like I'm having to learn computers all over again.
 
One more thing, does an expensive HSF seem to give the chip a boost in overclocking?

While it sounds like a dumb question, these chips run so cool that even the retail HSF is able to cool an overclocked a64 to well under 50C.
 
sappo said:
One more thing, does an expensive HSF seem to give the chip a boost in overclocking?

While it sounds like a dumb question, these chips run so cool that even the retail HSF is able to cool an overclocked a64 to well under 50C.
I'd be really careful with a retail hsf. I used one briefly with my FX51 and it was hitting 70C+ very quickly. In fact I'd put it up there with the prescott I had when it comes to using a retail hsf. It was more like 55C under water cooling.
 
sappo said:


Why is this? Async memory is never a good thing if you can help it, right?

I've noticed your bus speed is 250mhz. Shouldn't it be 800mhz? Edit: or is it 600mhz with an nforce3-150?

With an out of spec fsb, are you having problems with onboard stuff? sound, lan, etc..

Sheesh, all this new athlon64 stuff is confusing. I feel like I'm having to learn computers all over again.

Originally posted by Gautum
Unfortunately, I'm still using my old PC3200 memory, which tops out at about 228MHz.

He hasn't gotten new RAM.

His effective HT bus speed is multiplied by 3 by default to make 600, and he changed the setting to 2.5 and increased the bus to 250.

He's using a PCI soundcard that he said tolerates the higher PCI speed fine and I don't think he's using any other onboard stuff - that was the point of using the PCI cards he chose - it was a workaround for having no PCI locks. Maybe there are some onboard things hes using, but I didn't think there were.
 
Back