Investigations into Socket 939 Athlon 64 Overclocking
by Jarred Walton on October 3, 2005 4:35 PM EST- Posted in
- CPUs
RAM BIOS Settings
The final area that will need attention with overclocking is the RAM. The type of RAM used will dictate the approach to overclocking that RAM.
Even if you select the slowest possible timings, some DIMMs won't be able to run much above DDR400 speeds. Value RAM in particular is often what is left after all the high speed blanks are binned out, which is why it can be as cheap as half the cost of quality RAM. The above BIOS screens show our settings for 9x300 with the value and performance RAM - you can see that we had to drop to DDR266 vs. DDR333, since not even the OCZ ran optimally at DDR600. (At least, that's what we needed on this particular board with this particular set of DIMMs. 3-4-4-8 with 2T command rate worked, but was slower than DDR333 2.5-3-3-8 with 1T.) If you're going for top performance, value RAM is a poor choice. If, on the other hand, you want to save money, $85 RAM is a lot more attractive than $150+ RAM. The trick to using value RAM is that you basically have to keep it at or below DDR400 speeds, and this is accomplished by the use of memory ratios. There's a stigma against using higher memory ratios, because it "negates the performance gain." That can be true on some platforms, but it doesn't seem to affect Athlon 64 quite as much due to the integrated memory controller. Let's talk about why.
On the P4 chipsets, the NB talks to the CPU at one speed, and it can talk to RAM at a slower or faster speed. If it talks to RAM faster than it talks to the CPU (i.e. DDR2-533 or DDR2-667 with FSB800), there will be fewer "empty trains" going to the CPU. If it talks slower to the RAM than the FSB, however, there will be more empty slots on the FSB. When we look at Pentium overclocking in the future, we'll cover this more, but the basic idea is that you want the RAM to run at FSB speed or faster if at all possible. Dropping even to DDR333 (which is actually DDR320 for most Intel chipsets) can cause a 5 to 10% performance loss, and DDR266 would be even worse. So, why is Athlon 64 different?
An interesting corollary to the above discussion is in the Northbridge link of AMD vs. Intel. On AMD, the Northbridge is connected via the HyperTransport link. Running at 1000 MHz base (double-pumped makes it equivalent to 2000 MHz), the HT bus is 16-bits wide (2 bytes) with dedicated upstream and downstream links. All told, that gives:
The net result is that while a lower RAM clock speed can still impact performance on Athlon 64 systems, it will not do so as much as on a P4 style configuration. That's the theory, anyway, and we hope to support it with results from several planned overclocking articles.
To illustrate the above points, we've created a brief list of cycle rates and timings of RAM. We'll start with some common RAM speeds, but the actual speed of your RAM will depend on the CPU multiplier and CPU bus speed, among other things. With the clock speeds and timings, we can also calculate the latency and estimate the total memory latency. Remember that the RAM timings are relative to the base bus speed, not the doubled speed, so DDR400 has a 200MHz base clock speed. There is also latency involved internally in the CPU (typically at least a single RAM cycle each way is spent in the memory controller), as well as in sending the address request across the memory bus and receiving the data back across the bus to the CPU (we guessed at two RAM cycles each way). Command rate adds another cycle, and maybe we're wrong or missed some other potential delays.
Everything takes time, and the memory timings generally only reflect the delays caused internally by the RAM. Different clock speeds and timings should produce a different total latency, and we estimate the total memory subsystem latency below. We'll check the actual latency in our benchmarks later. In practice, going from DDR200 to DDR400 RAM will not cut the RAM latency in half, even at the same timings. For reference, our estimated latency is calculated as:
Even with all this information, we haven't even come close to touching on every BIOS option or memory timing that's available. That is, simply put, beyond the scope of this article. There are over a dozen memory timings listed in the DFI BIOS, and we left the majority of them set to Auto. The only items that we changed are CAS, tRCD, tRP, tRAS, and CMD. We may try to look at the other options in the future, but most of the extra settings are not available on the majority of the motherboards, and we want the information contained in this OC Guide to be applicable to more than just the "elite" motherboards.
The final area that will need attention with overclocking is the RAM. The type of RAM used will dictate the approach to overclocking that RAM.
Click to enlarge.
 |
 |
| Click to enlarge. | |
Even if you select the slowest possible timings, some DIMMs won't be able to run much above DDR400 speeds. Value RAM in particular is often what is left after all the high speed blanks are binned out, which is why it can be as cheap as half the cost of quality RAM. The above BIOS screens show our settings for 9x300 with the value and performance RAM - you can see that we had to drop to DDR266 vs. DDR333, since not even the OCZ ran optimally at DDR600. (At least, that's what we needed on this particular board with this particular set of DIMMs. 3-4-4-8 with 2T command rate worked, but was slower than DDR333 2.5-3-3-8 with 1T.) If you're going for top performance, value RAM is a poor choice. If, on the other hand, you want to save money, $85 RAM is a lot more attractive than $150+ RAM. The trick to using value RAM is that you basically have to keep it at or below DDR400 speeds, and this is accomplished by the use of memory ratios. There's a stigma against using higher memory ratios, because it "negates the performance gain." That can be true on some platforms, but it doesn't seem to affect Athlon 64 quite as much due to the integrated memory controller. Let's talk about why.
On the P4 chipsets, the NB talks to the CPU at one speed, and it can talk to RAM at a slower or faster speed. If it talks to RAM faster than it talks to the CPU (i.e. DDR2-533 or DDR2-667 with FSB800), there will be fewer "empty trains" going to the CPU. If it talks slower to the RAM than the FSB, however, there will be more empty slots on the FSB. When we look at Pentium overclocking in the future, we'll cover this more, but the basic idea is that you want the RAM to run at FSB speed or faster if at all possible. Dropping even to DDR333 (which is actually DDR320 for most Intel chipsets) can cause a 5 to 10% performance loss, and DDR266 would be even worse. So, why is Athlon 64 different?
An interesting corollary to the above discussion is in the Northbridge link of AMD vs. Intel. On AMD, the Northbridge is connected via the HyperTransport link. Running at 1000 MHz base (double-pumped makes it equivalent to 2000 MHz), the HT bus is 16-bits wide (2 bytes) with dedicated upstream and downstream links. All told, that gives:
2000 MHz * 2 bytes = 4000 MB/sThe total HT bandwidth is thus 8000 MB/s - 4000 MB/s in each direction. Here's the catch, though: this bandwidth is separate from the memory bandwidth. If you've ever tried overclocking the HT bus speed and found the results to be of little help, performance-wise, that should hopefully clear things up. There is rarely (if ever) 4 GB/s of data in either direction between the CPU and Northbridge. It's like putting an eight-lane highway through a rural farming community; since there's very little traffic to begin with, the extra lanes (bandwidth) won't help much. That's not to say this is a poor design decision - better to have too much available bandwidth than not enough! The P4 design doesn't suffer from a lack of bandwidth either; the problem is that the time in which it takes to get data from the RAM to the CPU is quite a bit longer - in other words, higher latencies.
The net result is that while a lower RAM clock speed can still impact performance on Athlon 64 systems, it will not do so as much as on a P4 style configuration. That's the theory, anyway, and we hope to support it with results from several planned overclocking articles.
To illustrate the above points, we've created a brief list of cycle rates and timings of RAM. We'll start with some common RAM speeds, but the actual speed of your RAM will depend on the CPU multiplier and CPU bus speed, among other things. With the clock speeds and timings, we can also calculate the latency and estimate the total memory latency. Remember that the RAM timings are relative to the base bus speed, not the doubled speed, so DDR400 has a 200MHz base clock speed. There is also latency involved internally in the CPU (typically at least a single RAM cycle each way is spent in the memory controller), as well as in sending the address request across the memory bus and receiving the data back across the bus to the CPU (we guessed at two RAM cycles each way). Command rate adds another cycle, and maybe we're wrong or missed some other potential delays.
Everything takes time, and the memory timings generally only reflect the delays caused internally by the RAM. Different clock speeds and timings should produce a different total latency, and we estimate the total memory subsystem latency below. We'll check the actual latency in our benchmarks later. In practice, going from DDR200 to DDR400 RAM will not cut the RAM latency in half, even at the same timings. For reference, our estimated latency is calculated as:
7 * (Cycle Latency) + CL + 0.1 * tRCD + 0.05 * tRPIt would be different for other platforms, and it's probably off by as much as 20%, we'd guess. The "7" represents the command rate, CPU to memory controller delays, and memory controller to RAM delays. We did use some tests to try to come up with a good estimate, but take the estimated latency in the following table with a serious dose of skepticism. For a P4 platform, the major change is that the inclusion of the NB will change the "7" factor to something much larger - perhaps 14 to 20. (We'll look at that in a future article.)
| RAM Speeds; Timings and Theoretical Latencies | |||||||||||
| RAM Rating | Base Speed | ns per cycle | CAS | tRCD | tRP | tRAS | CL ns | tRCD ns | tRP ns | tRAS ns | Estimated Latency |
| DDR266 | 133.33 | 7.5 | 2 | 2 | 2 | 5 | 15 | 15 | 15 | 37.5 | 69.75 |
| DDR266 | 133.33 | 7.5 | 2.5 | 3 | 3 | 7 | 18.75 | 22.5 | 22.5 | 52.5 | 74.63 |
| DDR266 | 133.33 | 7.5 | 3 | 4 | 4 | 8 | 22.5 | 30 | 30 | 60 | 79.5 |
| DDR333 | 166.67 | 6 | 2 | 2 | 2 | 5 | 12 | 12 | 12 | 30 | 55.8 |
| DDR333 | 166.67 | 6 | 2.5 | 3 | 3 | 7 | 15 | 18 | 18 | 42 | 59.7 |
| DDR333 | 166.67 | 6 | 3 | 4 | 4 | 8 | 18 | 24 | 24 | 48 | 63.6 |
| DDR400 | 200 | 5 | 2 | 2 | 2 | 5 | 10 | 10 | 10 | 25 | 46.5 |
| DDR400 | 200 | 5 | 2.5 | 3 | 3 | 7 | 12.5 | 15 | 15 | 35 | 49.75 |
| DDR400 | 200 | 5 | 3 | 4 | 4 | 8 | 15 | 20 | 20 | 40 | 53 |
| DDR450 | 225 | 4.44 | 2 | 2 | 2 | 5 | 8.89 | 8.89 | 8.89 | 22.22 | 41.33 |
| DDR450 | 225 | 4.44 | 2.5 | 3 | 3 | 7 | 11.11 | 13.33 | 13.33 | 31.11 | 44.22 |
| DDR450 | 225 | 4.44 | 3 | 4 | 4 | 8 | 13.33 | 17.78 | 17.78 | 35.56 | 47.11 |
| DDR500 | 250 | 4 | 2 | 2 | 2 | 5 | 8 | 8 | 8 | 20 | 37.2 |
| DDR500 | 250 | 4 | 2.5 | 3 | 3 | 7 | 10 | 12 | 12 | 28 | 39.8 |
| DDR500 | 250 | 4 | 3 | 4 | 4 | 8 | 12 | 16 | 16 | 32 | 42.4 |
Even with all this information, we haven't even come close to touching on every BIOS option or memory timing that's available. That is, simply put, beyond the scope of this article. There are over a dozen memory timings listed in the DFI BIOS, and we left the majority of them set to Auto. The only items that we changed are CAS, tRCD, tRP, tRAS, and CMD. We may try to look at the other options in the future, but most of the extra settings are not available on the majority of the motherboards, and we want the information contained in this OC Guide to be applicable to more than just the "elite" motherboards.
Facebook
Twitter
RSS
101 Comments
View All Comments
edlight - Tuesday, January 3, 2006 - link
I've found a way to overclock and retain the Power Now/Cool'n'Quiet.I let the motherboard do it's Cool'n'Quiet thing but I don't load the AMD driver. I run CrystalCPUID, which lets me set up the multiplier and voltage of each of the 3 cpu steps.
The voltage setting of the motherboard, for my Gigabyte, has to be on Auto for Crystal to be able to change it.
The highest Crystal can go with my 1.4v 3000+ Winnie is 1.45v.
This let me take it up to 3800+ -- a speed of 2.4. 240 x 10.
So it's running at 1.2 -- 240 x 5 -- most of the time. I set it at 1.2v there and froze it there and p95'd it overnight, as I did to the other 2 speeds.
For me this is a great compromise between running "cool 'n quiet" and high performance.
It's only a small percentage speed jump to 2.6, but requires alot of voltage and heat.
I can't say what the maximum voltage would be for a Venice. Crystal lets me choose higher voltages than 1.45, but it doesn't actually set them.
RaulAssis - Wednesday, December 21, 2005 - link
Some people reported that the Cool 'n' Quiet feature could work in a OC system. Maybe not all bioses support correct scaling of voltages when the system is OC and the Cool 'n' Quiet feature is turned on.JarredWalton - Wednesday, December 21, 2005 - link
With any moderate OC, CnQ is going to cause problems. It dynamically adjusts multipliers and voltages... something that will usually screw up an overclocked system. I would strongly discourage trying to use CnQ with an OC'ed setup. Some motherboard BIOSes actually disable CnQ automatically if you enable overclocking features.mrmoti - Wednesday, November 30, 2005 - link
If I understand correctly:Performance RAM running at DDR400 2-2-2-8 and Value RAM running at DDR400 2.5-3-3-8
At same OC on the processor, Performance RAM outperformed the Value RAM by 5% to 10%, being the price something between 80% to 100% more.
So, what's the impact of runnig faster memory at high lateny? Say DDR500 at 3-4-4-8
Because looking at the table of estimated latencies, (Performance) DDR400 2-2-2-5 has an estimated latency of 46.5, where (Value) DDR400 2.5-3-3-7 has an estimated latency of 49.75, an improve of 6.5% being in the range of 5% to 10% better.
By the same table, DDR500 3-4-4-8 has an estimated latency of 42.4, an improvement of 8.8% over the Performance DDR400 and 14.7% over the Value DDR400, based only in latencies.
Can anybody run a benchmark confirming/denying this?
Being the case that the price of DDR500 with those timings is in the middle between Performance and Value RAM
T Rush - Sunday, October 16, 2005 - link
One of the main focuses of this article seems to be value -vs- performance RAMs when over clocking, but you chose to run the performance RAM at settings where is doesn't perform, shame on you Jarred Walton, very disappointedIf you look at the settings you used to test the two RAMs at...
http://images.anandtech.com/reviews/cpu/amd/athlon...">http://images.anandtech.com/reviews/cpu...niceover...
...you see that the MAX speed you where able to run the OCZ Rev2 at was not in it's "performance envelope", as the OCZ Rev2 is one of the worst performers in this speed range http://www.anandtech.com/memory/showdoc.aspx?i=256...">http://www.anandtech.com/memory/showdoc.aspx?i=256... <OCZ Rev2 at 266MHz, and all the other "performance RAMs" beat it
But if you look at how the OCX Rev2 does work at much higher speeds, where it does perform...
http://www.anandtech.com/memory/showdoc.aspx?i=256...">http://www.anandtech.com/memory/showdoc.aspx?i=256...
...you find that it is performing much differently than what you tested at, and would have shown a much larger performance lead over the value RAM
How did you get this on to Anandtech? How could you show such a bad comparison of value -vs- performance RAM on a site which has always shown so much information about how these RAMs perform?
I not only blame you, but also the editors for not catching how badly you have managed to make performance RAM look. It is clear you were trying to prove that cheap RAM can falsely perform as well as high-end performance RAMs. If you truly wanted to show what performance RAM can do when over clocking you either needed to run the OCZ Rev2 at much faster speeds, or use a different performance RAM that works well at the sub 270MHz speeds you tested at.
JarredWalton - Sunday, October 16, 2005 - link
This is one set of RAM run through extensive tests on one platform. I've seen the same RAM run faster in some other systems, but not a whole lot. Just because some DIMMs reach DDR636 doesn't mean that all of them do. I could run this RAM at 3-4-4-8-2T timings at DDR600, but it actually ran worse than 2.5-3-3-8-1T with the lower memory ratio.What is clear is that I wasn't trying to "prove" anything. I was running some comparison tests with a system using two different types of RAM, and I'm sorry that you don't like the results. What I did prove was that someone one a budget could build a very fast system. An FX or San Diego core with higher quality RAM and a better motherboard would be better overall, but price/performance it would get stomped by this <$1000 setup.
T Rush - Monday, October 17, 2005 - link
I don't find the OCZ Rev2 to be a good example of the high performance RAM everyone thinks it is, as it doesn’t perform well at the speeds you (and most everyone else) use…not compared to other good over clocking RAMsGranted some of the other performance RAMs do cost much more than the value RAMs, and even more than the OCZ Rev2, but they would have shown a greater performance difference than the value RAM which in your tests was not able to run any faster than its stock rated speeds or timings
Your testing shows that running RAM at faster speeds adds very little performance over stock speed value RAM, and that is because the timings/speed relationships of that peculiar performance RAM at those peculiar higher speeds were not good.
As I said before, all the other performance RAMs beat it, and perform much better at speeds under 270MHz than the OCZ Rev2 does
Using the right RAM at the right speeds to run the best timings is the true art to over clocking, as RAM timings and speeds can allow the AMD64 to perform at much higher levels when over clocked
I do not disagree with your results, as that is how those RAMs perform:
Value RAM only being able to run its rated speeds and timings, not being able to over clock at all.
OCZ Rev2 running at higher speeds but with such bad timings that it shows very little performance gains unless you are able to run it at CAS 2.5 in the 300MHz range
I am not a fan of the OCZ Rev2 because of this, but I am a fan of performance RAM over value RAM, even on a budget system.
If you read clue22’s reply “so basically what the everybody is saying about the value RAM vs. low latency more expensive RAM is that for the athlon 64 it is basically a waste of money (i.e. you only get about 5% performance gain), but usually spend 100% or more money to get the "better" RAM.”
…and cyptonomicon’s “and its nice to see those ram comparisons. good to see those results on the latest a64 platform and confirm once again that the ram makes only a few percentage points difference”
…next intellon’s “I understand how/why the memory quality is not too imoprtant (5-9% increase for 100 bucks = not worthy)”
Clearly by using the OCZ Rev2 you did not show what spending a little bit more for better performing RAM can do. You have shown that running RAM at speeds with timings where it doesn’t perform well is a waste of money, but this does not answer any questions about value -vs- performance RAM
JarredWalton - Monday, October 17, 2005 - link
I've got X2 benchmarks with four different types of RAM in the works:OCZ VX
OCZ Plat Rev2 (TCCD)
Mushkin Value
PDP 2x1GB 2-3-2-5-1T
Other than the fact that 2GB of RAM helps out certain tasks (BF2 load times!), the total performance difference with those configurations is still not huge. With a 3.5V RAM voltage, the VX would do better, but even then the difference isn't above 10%.
T Rush - Thursday, October 20, 2005 - link
for a budget system I would say the socket 754 is better...as the motherboards and CPUs are cheaper...and you can get ClawHammer CPUs with the larger performance 1MB L2 cachethe only thing you miss out on with the 754 is the dual channel memory mode(which only adds very little performance anyway)...but by over clocking the core:memory speed you can easily match the performance gained by the greater bandwidth of the dual channel mode (this could be why the socket 939 doesn't show large gains from overclocking with the memory 'in-sync'..as it can't use all the bandwidth the faster memory gives)
with a mid-range system you could pick a 10X multi 3200+ Venice, or even a 11X multi 3500+ Venice(either of those would have a much better CPU multi for overclocking than the 3000+'s 9X multi) which would allow you to keep the HTT/HTL speed at a more reasonable level(270MHz and 245MHz to reach 2.7GHz CPU speed...where a 9X multi CPU would need a 300MHz HTT speed to run the CPU that fast)
...but the 3200+ and 3500+ are costing $190 to $250...so for not much more you could have a performance San Diego core on the 939 platform (3700+ 11X multi SD is only $267 now)
...so for $80 to $40 more I would go for the larger San Diego core...I would also spend the ~$20 more for CAS 2 RAM (over $90 Value RAM)...thats like just $100...for a computer with much better parts...and say you use this system for 18 months, that works out to less than $6 a month for a using higher quality parts
Deathcharge - Saturday, October 15, 2005 - link
Jarred this was a great article and did come at a great time as i am in the market for buying a bang for the buck system. One thing you didnt mention (although i saw that in the CPU-z screen shots) is the CPU steppinghttp://www.amdcompare.com/us-en/desktop/default.as...">http://www.amdcompare.com/us-en/desktop/default.as...
the 3200+ venice core comes in 3 different stepping and i belive the one you used in your article is the E3 stepping which is being replaced with the E6 stepping. Any info on how well the new stepping OC? initial reports from around the net indicate that it doesnt OC very well for some reason would love to read your comments on this.
Do you know if it is possible to OC to 2500 or 2600O with stock HSF as i would really like to save the money spent on the TT-90 and get a 7800GT (as opposed to x800xl). one final thing would OCZ value VX require active cooling?
thanks and keep up the good work, really enjoyed reading it and would look forward to future articles