I looked at the FX-53 benchmarks on THG, and their benches vs. the FX-51 are right in line with what they should be... I don't see any descrepency as with AT's tests.
From this I conclude that AT must indeed have tested the FX-53 with the OCZ memory, but used the Mushkin memory with the FX-51. That difference must explain the higher than expected performance of the FX-53.
That also suggests that OCZ memory is pretty good stuff...
To #27: that's my point. For the FX-53 to perform more than 9.1% faster than the FX-51, *SOMETHING* must be different. FactMan (#25) suggests the FX-53 is indeed a new stepping with some improvements.
In my second note, I also observed that AT tested the FX-51 & FX-53 with different memory, though the article is none too clear on this point.
Finally, as I pointed out in post 23, ECC is NOT required for the Athlon64 FX. The FX clearly *supports* ECC, but it is not required. Registered memory IS required for both the Opteron and Athlon64 FX.
why cant the fx-53 perform more than 9.1% better? it could have some revised coding within it which allows for better allocation of data, and improved prefetch, etc.
According to AMD and factually speaking after running a slew of benchmarks, the FX-53 performs only 108% faster than the speed of the 64 3400 Athlon. This number was in overal~gaming prowess, period. I'm not too concerned with encoding, compiling, blah blah blah so I leave that to the folks who are. Man it's too hard to justify the extra cost of the FX-53 in relation to the 3400 with only 6-8% difference in speed, which at any rate is most likey un-noticeable anyhow. For me, and this is purely my humble opinion, the extra money price difference would better be suited a mess load of BAWLS.
WHY CAN'T THEY COLOR CODE THE CHARTS! I have to spend three times as long reading each chart to determine who placed where as I do on a color coded chart such as those posted on Toms site. It is really annoying. I color code my charts at work. Every college class teaches color codeing. My 6 year old daughter color codes her drawings! Come on!!
TO #18, Athlon64Boy: The Athlon 64 FX retains the Registered requirement of the Opteron, but they were able eliminate the ECC requirement.
After I posed my question, I did look around a bit for PC3200 Registered Non-ECC memory: it is something of a rare animal, but it does exist. (I think it's pretty A64FX specific). On Newegg.com, only CorsairXMS is offered in Reg/Non-ECC. Their website specifically touts this memory as being tested in AthlonFX motherboards:
I suppose AT may have run their tests with ECC disabled, which as far as I know would eliminate any extra performance penalty. It may also be that the A64FX memory controller runs ECC and Non-ECC with equal efficiency... but there's always been a performance penalty in the past with ECC, registered or not.
I've been hearing "Hold off til socket 939" for almost 6 months now. It appears Anandtech.com has become a PR mouthpiece for AMD, or at the very least, become irresponsible to the point of recommending hardware that no one has publicly tested, priced, or seen. The future is now and socket 940 exists now. Don't forget the definition of the term, "vaporware".
#17,Johnsonx, well noticed. All a bit strange given there is only one Skt 940 m'board. I dont know why Derek would get problems with the FX53 and not the FX51. After all they are both running at the same clock speed and hence RAM speed. That damn memory controller perhaps!
"It may seem like that's what we've said at the end of every single CPU or graphics review for the past few months, but the giant caution sign will soon be taken down. "
"But just wait if that's at all possible. The end is near, we promise."
Talk about foreshadows. I wonder how long the chip in question has been in their lab. Forget the NDAs, tell us the real story!
Hi #17. The Athlon FX is basically an Opteron rebranded as an Athlon FX. It is a 940 Pin Processor, just like the Opteron. When AMD releases their 939-pin Athlon, this will take non-ecc ram. This 939-Pined athlon 64 will be the head cheese. It will have 128-bit Dual Channel ram, and not be using the ECC ram. The Athlon 64 has single channel Ram. Hope that helped
Ok, one final question: Why are the FX processors being tested with ECC memory? Doesn't ECC further increase the performance penalty vs. regular DDR400 memory? Don't most people building FX boxes use Registered non-ECC RAM?
So that was more than one question... doesn't three questions about the same thing count as one?
Ok, the article isn't exactly clear about this, but it looks like the FX-51 and FX-53 were tested with different RAM. There are two types of registered RAM listed, and the article does say the FX-53 was tested with the OCZ 3200 2:3:3 RAM. It doesn't say so, but I guess that implies the FX-51 was tested with the Mushkin "High Performance" 2:3:2 RAM.
By the timings listed, the Mushkin RAM used with the FX-51 should have been faster than the OCZ RAM used with the FX-53, but we all know that memory often performs differently in reality than it does on paper.
So if it is true that the two CPU's were tested with different RAM, and that the OCZ RAM is faster in actual use than the Mushkin (nevermind the spec'd timings), then I guess that explains the benchmark discrepancies.
Some of these benchmark numbers are a bit suspect. Compared with the FX-51, the FX-53 has a 9.1% higher clock speed. Therefore, all else being equal, 9.1% is the most extra performance the FX-53 can have compared to the FX-51... yet:
In DIVX encoding, the FX-53 scores 52 to the FX-51's 46.3: a gain of 12.3%
In 3DStudio, the FX-53 takes 2.78 minutes to render while the FX-51 takes 3.15 minutes: a drop in rendering time of 11.7%
In Lightwave the rendering time drop is 10.2%
SysMark 2004 DataAnalysis shows a whopping 19.1% performance increase, more than double what should be theoretically possible! All the rest of the SysMark 2004 scores show gains of 10 to 11 percent, still more than should be possible.
I do realize there is a margin of error in benchmarking, but I just don't see these inconsistencies for any of the other processors. All the other gains are comfortably less than theoretical maximums. I don't normally go through AT reviews calculating the percentage difference for all the scores measured, but the Divx score just jumped out at me (that was before I noticed the SysMark2004 Data Analysis score).
What can be made of this? Perhaps AT could re-check the numbers for the FX-51 and FX-53 alone on the suspect scores? If the benchmarks are indeed correct, then this implies AMD has made a stepping change that dramatically improved the efficiency of something in the CPU. Something that has to do with manipulating large data sets...
Aquamark Warcraft3 Q3 Arena Jedi Academy (beat the FX53)
equalled it in:
Halo (also FX53)
Got beaten (by 0.25%)in:
Gun Metal Wofenstein UT2003
That's a majority. If you count +/- 3% as significant (performance or error statistic wise) then the 3400+ is equal in all tests. You'd be a fool to buy an FX51 (or even FX53) for double the money.
The 3400+ beat the FX51 in most games and equalled it in the bandwidth intensive 2D apps encoding and rendering etc.. it even beat the FX53 in Jedi knight. Most a64s get a 10% overclock. So you'd have FX53 performance or better (because of o'clocked RAM and mobo) for less than half price (even less if accounting for non reg. single module memory).
Unless, of course, you are an "enthusiast" (AMD defined)!
The FX53 might not o'clock so well under standard cooling because it is at the present K8 headroom limit and suffers from the dual bank mem. controller and reg. memory which are less conducive to o'clocking.
The scalability is DEFINATELY encouraging for AMD users. If the FX-55 was out and scaled porportionately as well, it wouldn't have been much of a contest.
According to the test setup, they're using "ATI Catalyst 4.10"...
That's 7 driver revisions in the future! There is no 4.10, the newest version is 4.3. Perhaps they mean 4.1, but nevertheless they're reporting an impossible driver revision.
Anand, I think it is time to put your programming knowledge to use and begin work on an open source benchmark. The goals could include:
1. Making the methodology of a benchmark open to the public, so that there is no question of integrity.
2. Having a frequently updated database of benchmark scores, sorted by application, so that someone could enter in the applications they use most and have a benchmark chart custom generated for their typical usage.
But you know what irritates me. Is when these sites always benchmark the processors, they rarely close with the fact that AMD is a 64-bit capable processor. AMD can whip up on an Intel, even in todays 32 bit software. They fail to mention the 64 bit future.
Someone would be dumb to go out and buy a Pentium 4. When the AThlon 64/FX line hasn't even shown its true potential .... yet.
all your base are belong to us! Kudo's to AMD. It's about time that AMD start whooping ass on the evil~Intel camp. And although prices are high the AMD chip is still cheaper. And who even cares about price at this point. I think the purpose is to see who is extracting the most performance from their silicon. I mean shaZb0t, gimme a break already.
Man, when the only difference between a 3.4 EE and a 3.2EE is 3 tenths of a frame a second, you know these architectures and instruction sets reached their limit.
Definitely just a bragging rights CPU. This is turning into a pissing contest.
Ok, The Athlon FX brand is going Socket 939... But any Socket 940 board will be able to hold an Opteron 1XX processor with no problem. And guess what? These are NOT going away anytime soon. I think you do a disservice to people who chose the benefits of the platform.
So, with a thousand dollar processor budget, what is the best chioce? Honestly, the best choice is to wait. It may seem like that's what we've said at the end of every single CPU or graphics review for the past few months, but the giant caution sign will soon be taken down
It was so annoying that I almost didn't pay attention to the results (I like the pretty pictures). You have AT LEAST one spelling error per page! If this were a newpaper, it would even get close to being published. Show us some AT pride in your articles and fix it! I know you can do better.
We’ve updated our terms. By continuing to use the site and/or by logging into your account, you agree to the Site’s updated Terms of Use and Privacy Policy.
30 Comments
Back to Article
johnsonx - Tuesday, March 23, 2004 - link
I looked at the FX-53 benchmarks on THG, and their benches vs. the FX-51 are right in line with what they should be... I don't see any descrepency as with AT's tests.From this I conclude that AT must indeed have tested the FX-53 with the OCZ memory, but used the Mushkin memory with the FX-51. That difference must explain the higher than expected performance of the FX-53.
That also suggests that OCZ memory is pretty good stuff...
nourdmrolNMT1 - Monday, March 22, 2004 - link
yea, i ment registered, my bad.MIKE
johnsonx - Saturday, March 20, 2004 - link
To #27: that's my point. For the FX-53 to perform more than 9.1% faster than the FX-51, *SOMETHING* must be different. FactMan (#25) suggests the FX-53 is indeed a new stepping with some improvements.In my second note, I also observed that AT tested the FX-51 & FX-53 with different memory, though the article is none too clear on this point.
Finally, as I pointed out in post 23, ECC is NOT required for the Athlon64 FX. The FX clearly *supports* ECC, but it is not required. Registered memory IS required for both the Opteron and Athlon64 FX.
nourdmrolNMT1 - Saturday, March 20, 2004 - link
why cant the fx-53 perform more than 9.1% better? it could have some revised coding within it which allows for better allocation of data, and improved prefetch, etc.it requires ecc. so you have to run with ecc
MIKE
truApostle - Friday, March 19, 2004 - link
According to AMD and factually speaking after running a slew of benchmarks, the FX-53 performs only 108% faster than the speed of the 64 3400 Athlon. This number was in overal~gaming prowess, period. I'm not too concerned with encoding, compiling, blah blah blah so I leave that to the folks who are. Man it's too hard to justify the extra cost of the FX-53 in relation to the 3400 with only 6-8% difference in speed, which at any rate is most likey un-noticeable anyhow. For me, and this is purely my humble opinion, the extra money price difference would better be suited a mess load of BAWLS.FactMan - Friday, March 19, 2004 - link
In Reply to 16There are two things that are likely to be the reason for the superlinear performence improvement.
1) The integrated memory controller runs at core speed, hence increased clockspeed makes the mem controller faster and reduces latency.
2) It's build upon the newer CG stepping. This stepping fixes and improves several things, mainly to the memory controller.
bldkc - Friday, March 19, 2004 - link
WHY CAN'T THEY COLOR CODE THE CHARTS! I have to spend three times as long reading each chart to determine who placed where as I do on a color coded chart such as those posted on Toms site. It is really annoying. I color code my charts at work. Every college class teaches color codeing. My 6 year old daughter color codes her drawings! Come on!!johnsonx - Friday, March 19, 2004 - link
TO #18, Athlon64Boy: The Athlon 64 FX retains the Registered requirement of the Opteron, but they were able eliminate the ECC requirement.After I posed my question, I did look around a bit for PC3200 Registered Non-ECC memory: it is something of a rare animal, but it does exist. (I think it's pretty A64FX specific). On Newegg.com, only CorsairXMS is offered in Reg/Non-ECC. Their website specifically touts this memory as being tested in AthlonFX motherboards:
http://www.corsairmicro.com/corsair/products/specs...
I suppose AT may have run their tests with ECC disabled, which as far as I know would eliminate any extra performance penalty. It may also be that the A64FX memory controller runs ECC and Non-ECC with equal efficiency... but there's always been a performance penalty in the past with ECC, registered or not.
yumarc - Friday, March 19, 2004 - link
I've been hearing "Hold off til socket 939" for almost 6 months now. It appears Anandtech.com has become a PR mouthpiece for AMD, or at the very least, become irresponsible to the point of recommending hardware that no one has publicly tested, priced, or seen. The future is now and socket 940 exists now. Don't forget the definition of the term, "vaporware".Xaazier - Friday, March 19, 2004 - link
when is socket 939 due?Pumpkinierre - Thursday, March 18, 2004 - link
#17,Johnsonx, well noticed. All a bit strange given there is only one Skt 940 m'board. I dont know why Derek would get problems with the FX53 and not the FX51. After all they are both running at the same clock speed and hence RAM speed. That damn memory controller perhaps!GoatHerderEd - Thursday, March 18, 2004 - link
"It may seem like that's what we've said at the end of every single CPU or graphics review for the past few months, but the giant caution sign will soon be taken down. ""But just wait if that's at all possible. The end is near, we promise."
Talk about foreshadows. I wonder how long the chip in question has been in their lab. Forget the NDAs, tell us the real story!
athlon64boy - Thursday, March 18, 2004 - link
Hi #17. The Athlon FX is basically an Opteron rebranded as an Athlon FX. It is a 940 Pin Processor, just like the Opteron. When AMD releases their 939-pin Athlon, this will take non-ecc ram. This 939-Pined athlon 64 will be the head cheese. It will have 128-bit Dual Channel ram, and not be using the ECC ram. The Athlon 64 has single channel Ram. Hope that helpedjohnsonx - Thursday, March 18, 2004 - link
Ok, one final question: Why are the FX processors being tested with ECC memory? Doesn't ECC further increase the performance penalty vs. regular DDR400 memory? Don't most people building FX boxes use Registered non-ECC RAM?So that was more than one question... doesn't three questions about the same thing count as one?
johnsonx - Thursday, March 18, 2004 - link
Ok, the article isn't exactly clear about this, but it looks like the FX-51 and FX-53 were tested with different RAM. There are two types of registered RAM listed, and the article does say the FX-53 was tested with the OCZ 3200 2:3:3 RAM. It doesn't say so, but I guess that implies the FX-51 was tested with the Mushkin "High Performance" 2:3:2 RAM.By the timings listed, the Mushkin RAM used with the FX-51 should have been faster than the OCZ RAM used with the FX-53, but we all know that memory often performs differently in reality than it does on paper.
So if it is true that the two CPU's were tested with different RAM, and that the OCZ RAM is faster in actual use than the Mushkin (nevermind the spec'd timings), then I guess that explains the benchmark discrepancies.
Can AT confirm this? Derek?
johnsonx - Thursday, March 18, 2004 - link
Some of these benchmark numbers are a bit suspect. Compared with the FX-51, the FX-53 has a 9.1% higher clock speed. Therefore, all else being equal, 9.1% is the most extra performance the FX-53 can have compared to the FX-51... yet:In DIVX encoding, the FX-53 scores 52 to the FX-51's 46.3: a gain of 12.3%
In 3DStudio, the FX-53 takes 2.78 minutes to render while the FX-51 takes 3.15 minutes: a drop in rendering time of 11.7%
In Lightwave the rendering time drop is 10.2%
SysMark 2004 DataAnalysis shows a whopping 19.1% performance increase, more than double what should be theoretically possible!
All the rest of the SysMark 2004 scores show gains of 10 to 11 percent, still more than should be possible.
I do realize there is a margin of error in benchmarking, but I just don't see these inconsistencies for any of the other processors. All the other gains are comfortably less than theoretical maximums. I don't normally go through AT reviews calculating the percentage difference for all the scores measured, but the Divx score just jumped out at me (that was before I noticed the SysMark2004 Data Analysis score).
What can be made of this? Perhaps AT could re-check the numbers for the FX-51 and FX-53 alone on the suspect scores? If the benchmarks are indeed correct, then this implies AMD has made a stepping change that dramatically improved the efficiency of something in the CPU. Something that has to do with manipulating large data sets...
Pumpkinierre - Thursday, March 18, 2004 - link
#13 yeah I read it right.the 3400+ beat the FX51 in:
Aquamark
Warcraft3
Q3 Arena
Jedi Academy (beat the FX53)
equalled it in:
Halo (also FX53)
Got beaten (by 0.25%)in:
Gun Metal
Wofenstein
UT2003
That's a majority. If you count +/- 3% as significant (performance or error statistic wise) then the 3400+ is equal in all tests. You'd be a fool to buy an FX51 (or even FX53) for double the money.
retrospooty - Thursday, March 18, 2004 - link
The 3400+ beat the FX51 in most games ?What review did you read ? ;)
Pumpkinierre - Thursday, March 18, 2004 - link
The 3400+ beat the FX51 in most games and equalled it in the bandwidth intensive 2D apps encoding and rendering etc.. it even beat the FX53 in Jedi knight. Most a64s get a 10% overclock. So you'd have FX53 performance or better (because of o'clocked RAM and mobo) for less than half price (even less if accounting for non reg. single module memory).Unless, of course, you are an "enthusiast" (AMD defined)!
The FX53 might not o'clock so well under standard cooling because it is at the present K8 headroom limit and suffers from the dual bank mem. controller and reg. memory which are less conducive to o'clocking.
Jeff7181 - Thursday, March 18, 2004 - link
The scalability is DEFINATELY encouraging for AMD users. If the FX-55 was out and scaled porportionately as well, it wouldn't have been much of a contest.Guspaz - Thursday, March 18, 2004 - link
According to the test setup, they're using "ATI Catalyst 4.10"...That's 7 driver revisions in the future! There is no 4.10, the newest version is 4.3. Perhaps they mean 4.1, but nevertheless they're reporting an impossible driver revision.
msva123 - Thursday, March 18, 2004 - link
Anand, I think it is time to put your programming knowledge to use and begin work on an open source benchmark. The goals could include:1. Making the methodology of a benchmark open to the public, so that there is no question of integrity.
2. Having a frequently updated database of benchmark scores, sorted by application, so that someone could enter in the applications they use most and have a benchmark chart custom generated for their typical usage.
athlon64boy - Thursday, March 18, 2004 - link
I love AMD, I am a big fan.But you know what irritates me. Is when these sites always benchmark the processors, they rarely close with the fact that AMD is a 64-bit capable processor. AMD can whip up on an Intel, even in todays 32 bit software. They fail to mention the 64 bit future.
Someone would be dumb to go out and buy a Pentium 4. When the AThlon 64/FX line hasn't even shown its true potential .... yet.
Cybercat - Thursday, March 18, 2004 - link
GO AMD! :p So I'm a little biased, every hardware company has their fanboys. ;)CrystalBay - Thursday, March 18, 2004 - link
Drool...:)...truApostle - Thursday, March 18, 2004 - link
all your base are belong to us! Kudo's to AMD. It's about time that AMD start whooping ass on the evil~Intel camp. And although prices are high the AMD chip is still cheaper. And who even cares about price at this point. I think the purpose is to see who is extracting the most performance from their silicon. I mean shaZb0t, gimme a break already.pwn4g3
Regs - Thursday, March 18, 2004 - link
Man, when the only difference between a 3.4 EE and a 3.2EE is 3 tenths of a frame a second, you know these architectures and instruction sets reached their limit.Definitely just a bragging rights CPU. This is turning into a pissing contest.
dweigert - Thursday, March 18, 2004 - link
Ok, The Athlon FX brand is going Socket 939... But any Socket 940 board will be able to hold an Opteron 1XX processor with no problem. And guess what? These are NOT going away anytime soon. I think you do a disservice to people who chose the benefits of the platform.mcveigh - Thursday, March 18, 2004 - link
So, with a thousand dollar processor budget, what is the best chioce? Honestly, the best choice is to wait. It may seem like that's what we've said at the end of every single CPU or graphics review for the past few months, but the giant caution sign will soon be taken downWHAT YOU TEASE!!!
whats around the corner?????????????????????
WooDaddy - Thursday, March 18, 2004 - link
DEREK, DEREK, DEREK!!!Speeling errurs are awl ovur the plase!
THIS IS PITIFUL!!! SPELL CHECKER!!!
It was so annoying that I almost didn't pay attention to the results (I like the pretty pictures). You have AT LEAST one spelling error per page! If this were a newpaper, it would even get close to being published. Show us some AT pride in your articles and fix it! I know you can do better.
Try it again....