Sign in with
Sign up | Sign in
Your question

AMD Phenom vs INTEL C2D and C2Q - 3D rendering and gaming benchmarks - Page 2

Last response: in CPUs
Share
a b à CPUs
January 27, 2008 12:21:35 PM

Would that be down to the individual transistor design or more a question of the process used in manufacture?

I ask because IBM have gotten SOI above 4Ghz ... and more.

Why can't AMD then?

And thanks for the interesting information ... a refreshing change.
January 27, 2008 1:43:58 PM

More Pheom results here. According to Xbitlabs, Q6600 is 7% faster than Phenom 9900 2.6GHz while using considerably less power.

If you believe the stories of Intel's latest price cuts going around, the Q6600 will be $224, the Q6700 $266, the Q9300 for $266 and the Q9450 for $316.
January 27, 2008 2:03:46 PM

it could have a performance hit, but there's already this phenom-only feature that no one benchmarked yet, can't recall the name of the feature but makes each core use a separate dimm for its own, you would need 4.
Related resources
January 27, 2008 2:04:24 PM

EVIL PUMPER!!!

turpit said:
Meh, we all know Im just an evil paid Intel fanboy pumper thats just here to insitgate trouble and dilute facts ;) 

a b à CPUs
January 27, 2008 2:06:09 PM

I'd call you a paid pumper but at 24 your still a kid just out of college.

January 27, 2008 2:10:03 PM

Reynod said:
I'd call you a paid pumper but at 24 your still a kid just out of college.


If you're referring to me, I am currently pursuing my masters degree, so I guess I am still in college.
a b à CPUs
January 27, 2008 2:13:06 PM

Good on you

:) 
January 27, 2008 3:25:26 PM

Mandrake_ said:
More Pheom results here. According to Xbitlabs, Q6600 is 7% faster than Phenom 9900 2.6GHz while using considerably less power.

If you believe the stories of Intel's latest price cuts going around, the Q6600 will be $224, the Q6700 $266, the Q9300 for $266 and the Q9450 for $316.


Since the author mentions the TLB but specifically omits what version of bios he was using... we will have to assume he was using the latest bios that was available at that time. Benchmarking with an older bios on a brand new motherboard would be contrary to current industry practices. Besides... it appears he uses DDR2-1066 memory. On that motherboard there were difficulties with older bios and that memory speed. SO... If he WAS using the latest bios... and he completely fails to mention it... then we have to assume that the benchmarks in this article were created WITH the TLB patch in place.

If we go on that premise... then we can't really use any of the benchmarks. (And I would suggest throwing out this set of suspect benchmarks.)

However, if you insist on using it, the we will have to add an average score to all benchmarks of about 6%-10% to account for the TLB patch. When we do that... the Phenom 9700 is closer to the Q6600 and the 9900 is closer to the Q9700.

======================================================
ADDED CONFIRMATION THAT YOUR LINK HAS THE BIOS PATCH IN PLACE:

After I wrote the information above I did even more research to verify my conclusion.

We can look at the Cinebench rendering in an article where the TLB BIOS patch was not available: (This author later did an article with the patch available and turned on so we can compare and further verify this theory.) http://techreport.com/articles.x/13633/12

That benchmarks shows that the numbers in your link are either WAY off... or they have the Bios patch in place.

If we take your link's numbers and add about 8% you get similar results to the link I provided. This further verifies my premise that the 9700 is more comparable to the Q6600 and that the 9900 is more akin to the Q6700.

==================

Thanks for making me do the research! Between you and several others that make me dig this stuff up... you are verifying (at least to me) that the "No Phenom can match the available Intel quads" myth is false. (So I can laugh at the people that keep posting commentsl like that as if it was some kind of "mantra". Just because people keep repeating it doesn't make it true.)
January 27, 2008 4:33:50 PM

Reynod said:
Would that be down to the individual transistor design or more a question of the process used in manufacture?

I ask because IBM have gotten SOI above 4Ghz ... and more.

Why can't AMD then?

And thanks for the interesting information ... a refreshing change.


I'll answer the question with an example -- intel's 90nm. You've got the NetBurst stuff and the Mobile stuff. And it's on the exact same process. One is leaky as hell but runs extremely fast, the other is lots more efficent but can't get anywhere close to the raw clockspeeds of the first. We all know which is actually the better part.

So, the data I show demonstrates "ultimate" transistor performance. Which is different than CPU performance. What you can grasp from the data is, for "identical" design, intels transistors will either -- have better thermals (due to lower current required) at the same clockspeed, or have a higher clockspeed ceiling. Thus they will win both the performance/watt and retain the ultimate performance crown. Assuming the same design. We all saw how a non-ideal design can modulate this.
January 27, 2008 10:30:13 PM

keithlm said:
If we take your link's numbers and add about 8% you get similar results to the link I provided. This further verifies my premise that the 9700 is more comparable to the Q6600 and that the 9900 is more akin to the Q6700.


Thats called sudo-science my little friend. And your conclusion is all wrong.

keithlm said:
We can look at the Cinebench rendering in an article where the TLB BIOS patch was not available: (This author later did an article with the patch available and turned on so we can compare and further verify this theory.) http://techreport.com/articles.x/13633/12


Sorry but that techreport page, although it did not include any 3D rendering, clearly shows the Q6600 handing the 2.6ghz phenom is ass. (although the phenom does win in the firefox test!). The author really should have included 3d rendering tests... since he was comparing it to his previous benchmark.

http://www.techreport.com/articles.x/13741/3


January 27, 2008 11:20:38 PM

keithlm said:
Since the author mentions the TLB but specifically omits what version of bios he was using... we will have to assume he was using the latest bios that was available at that time. Benchmarking with an older bios on a brand new motherboard would be contrary to current industry practices. Besides... it appears he uses DDR2-1066 memory. On that motherboard there were difficulties with older bios and that memory speed. SO... If he WAS using the latest bios... and he completely fails to mention it... then we have to assume that the benchmarks in this article were created WITH the TLB patch in place.

If we go on that premise... then we can't really use any of the benchmarks. (And I would suggest throwing out this set of suspect benchmarks.)

However, if you insist on using it, the we will have to add an average score to all benchmarks of about 6%-10% to account for the TLB patch. When we do that... the Phenom 9700 is closer to the Q6600 and the 9900 is closer to the Q9700.

======================================================
ADDED CONFIRMATION THAT YOUR LINK HAS THE BIOS PATCH IN PLACE:

After I wrote the information above I did even more research to verify my conclusion.

We can look at the Cinebench rendering in an article where the TLB BIOS patch was not available: (This author later did an article with the patch available and turned on so we can compare and further verify this theory.) http://techreport.com/articles.x/13633/12

That benchmarks shows that the numbers in your link are either WAY off... or they have the Bios patch in place.

If we take your link's numbers and add about 8% you get similar results to the link I provided. This further verifies my premise that the 9700 is more comparable to the Q6600 and that the 9900 is more akin to the Q6700.

==================

Thanks for making me do the research! Between you and several others that make me dig this stuff up... you are verifying (at least to me) that the "No Phenom can match the available Intel quads" myth is false. (So I can laugh at the people that keep posting commentsl like that as if it was some kind of "mantra". Just because people keep repeating it doesn't make it true.)


WoW!!! Lots of Wild Assumptions.

Actually, on the Date of that article the TLB patch was not generally available.
So, no you cant "add" 8% for the fun of things.
The Fact it was not mentioned would indicate it was not available.

However, the Site did complete testing w/ and w/o the patch after it became available.
http://xbitlabs.com/articles/cpu/display/amd-phenom9600...

The Phenom did not keep up.
January 27, 2008 11:40:38 PM

zenmaster said:
The Phenom did not keep up.


HA, thats a nice way of saying it.
January 28, 2008 1:20:36 AM

Actually, personal experience with the phenom myself, as you can see in my sig. Depending on the Benchie, TLB fix has a HUGE effect. I'm currently refusing to use the 1.2 Bios on my mobo because of the fact that it is enabled in that bios without a way to turn it off. Which is upsetting because I had a stable OC to 2.7Ghz on my proc. Rolled back to beta bios 1.13 and had to knock my OC down to 2.5 because of it. Though even with losing 200Mhz I'm still putting out better numbers than I had at 2.7 with the TLB fix.

A lot of the Bench marks that are in the article that the OP posted, would also be affected by an increase in NB/IMC, and Hypertransport speed.

TLB fix also effects memory latency in a very severe way. My memory latency with the TLB fix was 109.7ns, without it, my latency is 62.7ns according to Everest. I can pretty much go by bench results from everest, for almost the whole thing. It knocks down the memory bandwidth by a lot, which will have an effect on a lot of things.

For example the CPU PhotoWorxx benchmark on everest. With TLB Fix at 2.7Ghz I was getting around 22k, without the TLB fix at 2.5 I'm getting a little over 47k. I suspect that has to do with the memory latency almost being doubled with the TLB fix in place.

Once I get ahold of a decent BIOS, I'm going to play around more with things. And find out how much the NB/IMC, and HTT speeds effect things. But, they aren't doing as badly as they're made out to be. But, there are definitely some huge issues that I see that need to be fixed on them. Some kind of a hot spot seems apparent like someone else said. I believe the cores themselves actually never do get above 30c on my setup. But then when you look at the actual CPU temp in Everest it's showing 38c average under 100% load from Prime 95 small FFTs, never goes above 40c though on that.

January 28, 2008 1:25:48 AM

skittle said:
Thats called sudo-science my little friend. And your conclusion is all wrong.



Sorry but that techreport page, although it did not include any 3D rendering, clearly shows the Q6600 handing the 2.6ghz phenom is ass. (although the phenom does win in the firefox test!). The author really should have included 3d rendering tests... since he was comparing it to his previous benchmark.

http://www.techreport.com/articles.x/13741/3


Actually my conclusion is not "wrong" as you claim. You might deem it not the same as your conclusion... but you can't change my conclusion just by telling me it's wrong. Especially when the numbers I see contradict your opinion.

Personally I don't just look at a benchmarking article and blindly accept the results as a statistic. I have to look at each benchmark and decide for myself what benchmarks are more important. For example if looking at games... then multi-threaded games would carry much more weight than single threaded games. (And multi-threaded benchmarks of other types would be also be more important.)

In addition you seem confused about the techreport page. That page showed the Phenom 9900 ahead of the Q6600 for the benchmark in question. Oh... and it IS a 3d rendering benchmark.
January 28, 2008 2:03:19 AM

zenmaster said:
WoW!!! Lots of Wild Assumptions.

Actually, on the Date of that article the TLB patch was not generally available.
So, no you cant "add" 8% for the fun of things.
The Fact it was not mentioned would indicate it was not available.

However, the Site did complete testing w/ and w/o the patch after it became available.
http://xbitlabs.com/articles/cpu/display/amd-phenom9600...

The Phenom did not keep up.


No wild assumptions needed. You may consider my analysis to not carry a lot of value because you have personal standards of measurement... but my analysis is valid.

The patch WAS available... claiming it wasn't available won't change the fact. (And claiming it wasn't applied because it wasn't mentioned is also not a good assumption.) We would need to contact the author to find out exactly what bios was used.

It is good that the author went back and RE-DID the benchmark after a method of disabling the patch become available.

But then there are OTHER discrepancies completely separate from this issue: If you look at the benchmark values between the tests... some of the Intel benchmark numbers went up between the two benchmarks... while some of the Phenom values went down. Sure it was only 6 or 7 points up here... and/or 3 or 4 points down there... but it is shows an inconsistent sets of results.

For me this brings into question the validity of these reviews; especially when the results are vastly different than the results of similar reviews from other people and websites. (I'm sure based on your post you won't really care about this issue.)

Sure... on these FORUMS my comments will be ridiculed and met with derision... but that's okay... it doesn't change my mind or my opinion. It doesn't make me "right" or "wrong". (And it doesn't change the validity of my analysis... since analysis are very subjective anyway.)

I admit bias because of the following:

Intel made very bad decisions with the Netbust architecture. Then they tried to use larger and larger caches as a method to increase performance. This led to chips that couldn't multi-task very well. (And then throw in the hyperthreading fiasco. Sometimes it seemed to "work". More often it actually hurt performance. I ended up disabling it to get better overall performance.) Sadly I still have to use this sorry excuse for a machine. (Yes it is also a "Dell". My company wouldn't listen when I asked them to let me build a machine for the same price that would have MUCH better parts. It's a corporation thing.)

ANYWAY: Intel is now starting to use this same "trick" again with their current chips. Some people don't care... or don't remember. I personally am keeping an eye on this. I fully expect that this little trick will lead to the same end results. (Although probably not as bad... but that doesn't make it any better.) I believe this shortcut will end up coming back and biting them. Yet again.
January 28, 2008 2:25:45 AM

keithlm said:
Actually my conclusion is not "wrong" as you claim. You might deem it not the same as your conclusion... but you can't change my conclusion just by telling me it's wrong.

Yes I can.

keithlm said:
Personally I don't just look at a benchmarking article and blindly accept the results as a statistic.


Oh really? Then whats with this statement:
keithlm said:
...If we take your link's numbers and add about 8% you get similar results to the link I provided. This further verifies my premise...



keithlm said:
In addition you seem confused about the techreport page. That page showed the Phenom 9900 ahead of the Q6600 for the benchmark in question. Oh... and it IS a 3d rendering benchmark.


No, im not confused... I was talking about the other techreport page where they compared the phenom with the patch and without: http://www.techreport.com/articles.x/13741/3
January 28, 2008 2:26:04 AM

keithlm said:
ANYWAY: Intel is now starting to use this same "trick" again with their current chips. Some people don't care... or don't remember. I personally am keeping an eye on this. I fully expect that this little trick will lead to the same end results. (Although probably not as bad... but that doesn't make it any better.) I believe this shortcut will end up coming back and biting them. Yet again.



Benchmarks don't lie my friend. Decide what you want the machine for, pick the relevant benchies, and then make an informed decision.

If you want to use the machine for 24/7 memory bandwidth testing, go with AMD.

But if you're going to use the machine to run any of those pesky "applications", go with Intel. You can say that Intel's trick is cache, but I'll tell you what, if AMD could do it, they would. AMD can't even get their small amount of cache to work properly on their brand new K10 (TLB bug has to do with addressing of cache).

So who cares why Intel is better, but rather the fact that it is better. You can go for AMD's "true quad core" but you'll also be getting truly sub-par performance with no overclocking headroom.
January 28, 2008 3:13:45 AM

TechnologyCoordinator said:

So who cares why Intel is better, but rather the fact that it is better. You can go for AMD's "true quad core" but you'll also be getting truly sub-par performance with no overclocking headroom.


Ah but see that is the problem.

"Better" is not the same for everyone.

I don't blindly look at 15 benchmarks where 12 of them show one chip winning and 3 of them show the other chip winning and blindly say: "Oh.. the one chip is better. I'm done. There can be no question about it."

I have to know WHAT the benchmarks are actually testing.

For example: In several reviews I've seen the Phenom looking really good on real world application or synthetic benchmarks... and then suddenly start looking bad when a bunch of games are tested. I don't necessarily consider this situation a "loss". I will review the game benchmarks and if I see a lot of single threaded applications... I don't place a lot of weight on them; what matters is the NEWEST games that are multi-threaded. And guess what? The Phenom doesn't fall behind as much when only these newer games are considered. (Depending on the reviewer... it goes both ways.)

On the non-game benchmarks... I might throw out benchmarks that seem too high or too low as being too anomalous. Or I'll want more information as to why they are anomalous. If a vastly weird result can be explained by an architecture feature then I will reconsider that benchmark as valid. For example: We know why AMD does better on memory benchmarks. That is explainable and expected. If the AMD did 4x better on the YIZMO application... I'd want to know WHY and if I couldn't determine an answer I would have to throw out or at least greatly discount the results of that benchmark; at the very least consider the result suspect.

You see... contrary to what some people are saying: sometimes benchmarks DO lie.

You can't remove the human factor 100% from these benchmarks. Otherwise why would one reviewer have dramatically different results from another on the same benchmark when they have similar hardware?

BUT EVEN if I just take the hard numbers... on the benchmarks in question the 9900 is benchmarking much closer to the Q6700 more often than it comes close to the Q6600. So when I see someone say "The Q6600 is much better than even the best Phenom" I have to discount that opinion... it does not match the facts.

(ALSO: I generally try to look at more than one review/reviewer/website to make an opinion. To blindly accept one would be going down the rosy path.)

Oh and BTW: The AMD might be behind some of the Intel chips on some benchmarks... but it is not "sub-par" as you and many others erroneously claim. Actually I find it fascinating that so many people can be so obtuse. A true "tie" would be a 50%/50% situation. I consider this closer to a 66%/33% situation. But that is NOT "truly sub-par".

As for the overclocking headroom... that has little weight for me. I don't overclock. (Other than as an amusing thing to try out for a few hours.)
January 28, 2008 3:24:24 AM

keithlm said:
"Better" is not the same for everyone.


What?

Quote:
bet·ter1
-adjective, compar. of good with best as superl.
1. of superior quality or excellence: a better coat; a better speech; a better processor.
2. morally superior; more virtuous: They are no better than thieves.
3. of superior suitability, advisability, desirability, acceptableness, etc.; preferable: a better time for action.


Phenom < C2Q

End of story.
January 28, 2008 4:07:56 AM

What are all you guys arguing about?

Phenom 9900 looks to perform around a Q6600 and/or Q6700, modulo TLB.
Phenom 9900 is priced much closer to a Q6600 than a Q6700.

There are some applications which perform better clock-for-clock on a Phenom than a Kentsfield. There are many more which perform better on a Kentsfield. I would buy a Kentsfield (well, Yorkie when they come out) over a Phenom, but I'm not going to discount brand and application-specific loyalty.

The real issue is that it's AMDs high-end (on the desktop) which is competing against Intels mid-range/low end. *this* is why people have been disparaging Phenom.

We can even argue if this is even the right comparison to be making, since the server sales/performance are arguably more important to the company's health than the desktop performance.

And there, you can not dispute the advantage the IMC plays for AMD in keeping them in the game. Folks, you may not like it much, but I'm not sure AMD *cares* about desktop performance at the moment. I think that's short-sighted, but it's about the only thing they can do to stay afloat. They have to maximize their bandwidth/memory advantage as much as they can because it's clear that their architecture+process is somehow nerfed to the point where they have absolutely NO headroom to ramp up clockspeed. Whereas the success in overclocking the Core2 chips points to a large amount of headroom, if needed, to answer AMD. It's a good thing that, for a lot of the high-margin business, things get rate limited by memory throughput and not raw CPU horespower.

And that's also precisely why the TLB bug is so troubling -- it hits AMD where it hurts the most. Now that it's out there, they will lose any credibility they have left with the OEMs if they don't get the fix correct.
January 28, 2008 12:58:38 PM

ryman554 said:
And that's also precisely why the TLB bug is so troubling -- it hits AMD where it hurts the most. Now that it's out there, they will lose any credibility they have left with the OEMs if they don't get the fix correct.


Plus the fact that you can't get a K10 based Opteron right now!
January 28, 2008 1:01:43 PM

TechnologyCoordinator said:
If you're referring to me, I am currently pursuing my masters degree, so I guess I am still in college.


What are you studying? how old are you? and whats your favorite colour?


lol u dont have to answer that
January 28, 2008 3:32:00 PM

spoonboy said:
What are you studying? how old are you? and whats your favorite colour?


lol u dont have to answer that


MSIA (Masters of Science, Information Assurance)
24
Detroit Red Wings Red, and it's a color, not colour


:)  There you go Mr. Spoon!
January 28, 2008 3:47:11 PM

Reynod said:
Would that be down to the individual transistor design or more a question of the process used in manufacture?

I ask because IBM have gotten SOI above 4Ghz ... and more.

Why can't AMD then?

And thanks for the interesting information ... a refreshing change.


AMD certainly can clock Phenom above 4Ghz... but the question is, at what cost? IBM's Power chips are aiming at different market than AMD's. Similar to Itanium, which has their yield completely in the toilet, IBM can charge their Power chips an insane amount of markup, that made manufacturing them still economical.

On the other hand, AMD can't rely on the hope of producing a 4Ghz Phenom, at the cost of 300 wasted wafers. Sure they can produce a golden chip once in a while, but the majority of other chips still don't yield. What would happen if AMD does put out a 4Ghz Phenom product, but with the price tag of 10k?

Afterall, IBM's Power chips are architecturally different than AMD's K10. The routing of wires simply made AMD's 4Ghz Phenom attempt unfeasible.
January 28, 2008 3:51:38 PM

yomamafor1 said:
AMD certainly can clock Phenom above 4Ghz...


I highly doubt they could cherry pick more than a handful of those.



I said it before: Where the hell are the K10 dual cores!? It just doesn't make sense. They would yeild so much better than a monolithic quad, they would clock higher, and would score higher in single threaded games (most of today's games).

Where are the dualies!
January 28, 2008 4:25:44 PM

It's so sad that people cant do basic math. Phenom + mobo + 2G of ram = $300. Find me a similar intel quad combo for $300? Oh shucks that's right, you cant, because intel is oh so much better! In a nation swimming and drowning in debt, it is NO surprise that cost is so marginalized.
January 28, 2008 4:31:58 PM

shadowmaster625 said:
It's so sad that people cant do basic math. Phenom + mobo + 2G of ram = $300. Find me a similar intel quad combo for $300? Oh shucks that's right, you cant, because intel is oh so much better! In a nation swimming and drowning in debt, it is NO surprise that cost is so marginalized.


Be careful, you may get what you wish for.

If Intel sells a Q6600 for let's say, $150, do you know how that would affect AMD?!
January 28, 2008 4:50:29 PM

shadowmaster625 said:
It's so sad that people cant do basic math. Phenom + mobo + 2G of ram = $300. Find me a similar intel quad combo for $300? Oh shucks that's right, you cant, because intel is oh so much better! In a nation swimming and drowning in debt, it is NO surprise that cost is so marginalized.



Phenom 9500: $199.99
2GB of ram: ~$50
Motherboard: What $50 motherboard can you use? Sounds like you cant do the math.

In any case, a Q6600... for $55 more... outperforms that phenom (even at stock speeds) and clocks (much) higher. If im going out and spending $1000+ on a new system... I think I can afford that extra $55 to get a way better system.

Now if that phenom could run on my AM2 motherboard, and not take a huge performance hit... I might consider it. As of today, my board still does not support phenom... although the latest BIOS update did fix sempron temperature issues!
January 28, 2008 5:51:21 PM

TechnologyCoordinator said:
Be careful, you may get what you wish for.

If Intel sells a Q6600 for let's say, $150, do you know how that would affect AMD?!


Not sure, but I bet it won't only be Baron who's screaming "predatory pricing" :kaola: 
a b à CPUs
January 30, 2008 11:53:20 AM

mathos keep us posted on your efforts with the tlb fix turned off.

This sounds of real interest.

That's a big latency hit to chew up 200mhz of additional perfomance.

Take a look at Scott Wasson's site where he has done some interesting benchies on the latency issue.

I imagine things will improve with the B3 respin ... how much is the interesting point of conjecture.

If they also tighten the L3 latency and increase the IMC to the same level as the ES benchies released earlier I'd say the results will be welcome.

Obviously the headroom isn't likely to increase much as the process still can't meet Intel's ...

cheers
January 30, 2008 5:00:59 PM

yeah i reckon they'll have changed more with b3 than just the tlb error. Maybe the north bridge wont have a small stroke if you try and overclock.
January 30, 2008 5:38:05 PM

skittle said:
how come i havnt seen phenom benched on a am2 chipset? its always benched on the 790fx chipsets... kind of throws a wrench in the whole cheaper is better campaign


Perhaps they're attempting to mitigate the effect of older technology. They are attempting to benchmark the CPU and the "new" platform.

Once Phenom has its act together we'll probably see more reviews attemping to review the AM2 socket and whether or not it's comparable to the 790. The key word is "attempt". As Tom's already displayed, majority of manufacturers aren't supporting the first phenom chips on their AM2 boards.
January 30, 2008 5:41:33 PM

An after-thought...

Maybe the manufacturers decided not to support the Phenom as their price margins on these cheap mobo's are already low enough. Why not force them to upgrade to another $80 board?

This might be all instigated as a business decisision. Personally, I wouldn't want to throw away more money developing drivers and bios revisions on a board that might put 5USD in my pocket.

I have a feeling 45nm might let the K10 stretch its legs some.
      • 1
      • 2 / 2
!