Sign in with
Sign up | Sign in
Your question

AMD Phenom 9600 benchmark - Sandra XII

Last response: in CPUs
Share
November 10, 2007 12:20:24 AM

thanks for the find. I'll run the benchmark on Q6600 @ 2.3Ghz to compare the score.
Related resources
November 10, 2007 12:37:53 AM

X-Files said:
It beats my system (Q6600 overclocked to 3.2 with 8800 GTS/640M at stock speed, which scored 11247):
http://www.expreview.com/news/hard/2007-11-06/119433358...


Compare your CPU score to their CPU score. They used 8800GTX for graphic part.

GP9600 lost to my Q6600 running at 2.3Ghz.

Correction: GP9600 scored about 400 points lower than my Q6600 at stock in 3DMark 06. I never benched with Q6600 @ 2.3Ghz.
November 10, 2007 1:38:19 AM

yomamafor1 said:
Compare your CPU score to their CPU score. They used 8800GTX for graphic part.

GP9600 lost to my Q6600 running at 2.3Ghz.

Correction: GP9600 scored about 400 points lower than my Q6600 at stock in 3DMark 06. I never benched with Q6600 @ 2.3Ghz.


You're right, I just re-run 3DMARK 06 again on my system and here's the result:
http://service.futuremark.com/orb/resultanalyzer.jsp?pr...

But don't forget - my CPU is oc to 3.2Ghz, their CPU is oc to only 3Ghz. I have 200Mhz advantage.
November 10, 2007 2:58:47 AM

you can also downclock it to 3.0Ghz, and see if it matches up.

So far GP9600 @ 2.3Ghz lost to my Q6600 @ 2.3Ghz in Cinebench, Sis Sandra, and other benchmarks.
November 10, 2007 5:35:07 AM

X-Files said:
Hello eveyone,

Here's another unofficial AMD Phenom 9600 benchmark:
http://my.ocworkbench.com/bbs/showthread.php?p=423034#p...

Don't hold your breath too much... ;) 


Here are the benchmarks OCW took.



My Q6600 @ 2.3Ghz scored 253133iit/s for Int x8, and 13777fit/s for Float x4. The Int test can be discarded since Phenom was running with Int x4 only.

Then onto Dhrystone/ Whetstone:


Again, 2.3Ghz Q6600 scored 42347 MIPS in Dhrystone (Int test), and 29590 MFLOPS in Whetstone (FP test). Since AMD has been dominating in the FP arena for a while, but what really surprised me is how low the FP score is. I was expecting somwhere along the 1.5x of Q6600 clock for clock.
November 10, 2007 6:05:11 AM

Did anyone else notice the CPU-Z screen shot. It shows that phenom with SSE 4A instruction set. I thought that AMD was by passing that instruction in favor of SSE 5. Can anyone clarify?
November 10, 2007 6:09:37 AM

ira176 said:
Did anyone else notice the CPU-Z screen shot. It shows that phenom with SSE 4A instruction set. I thought that AMD was by passing that instruction in favor of SSE 5. Can anyone clarify?


AMD only supports a subset of SSE4 instruction set, hence the SSE4"A". And you're correct, they're bypassing the majority of SSE4 instruction set for SSE5.
November 10, 2007 4:07:05 PM

yomamafor1 said:
you can also downclock it to 3.0Ghz, and see if it matches up.

So far GP9600 @ 2.3Ghz lost to my Q6600 @ 2.3Ghz in Cinebench, Sis Sandra, and other benchmarks.


I just downclocked my Q6600 to 3Ghz (333Mhz x 9) and run 3DMark06 again:
http://service.futuremark.com/orb/resultanalyzer.jsp?pr...

As you can see, their Phenom X4 (oc to 3Ghz) scored 4319 Marks compared to mine which scored 4769 Marks.

In regard to Sisoftsandra XII, I downclocked my Q6600 to 2.3Ghz and the result is in line with yours:
Multi-Media Float x4 iSSE2 = 137375 fit/s and Whetstone iSSE3 = 29500 MFLOPS.


I really start to wonder if it could even beat the Q6600 clock for clock...let alone the QXnnnn. Let's hope the final product will do better.
November 10, 2007 5:09:37 PM

why did they blur out the name of the proc in the sandra tests.
the x4 proc was at 2.3 it says. not 3.0
November 10, 2007 6:23:23 PM

Quote:
Id like to see benches on something other than a Nforce560. Maybe on AMD's own chipset to make it as fair as the C2Q with a Intel chipset.


You got it.
http://www.expreview.com/news/hard/2007-11-08/119451488...
Quote:

mobo%uFF1A790FX (AMD reference Board)
graphic%uFF1AGeForce8800GTX
graphic driver %uFF1A169.02
soundcard driver%uFF1ARealtek 1.80




Q6600@ 2.3Ghz
November 10, 2007 7:43:49 PM

yomamafor1 said:
You got it.
http://www.expreview.com/news/hard/2007-11-08/119451488...
Quote:

mobo%uFF1A790FX (AMD reference Board)
graphic%uFF1AGeForce8800GTX
graphic driver %uFF1A169.02
soundcard driver%uFF1ARealtek 1.80


http://www.expreview.com/img/news/071108/PhenomX4_2.jpg

Q6600@ 2.3Ghz
http://img.photobucket.com/albums/v30/johnli0615/Cinebench2300.jpg

I just wonder if there going to start benching the X4 in dual channel anytime soon as even the FX62 has a higher single core performance. The 5000+ brisbane is neck and neck with the X4 in single core.

November 10, 2007 8:22:17 PM

How close are the engineering samples to the released CPU's? While I don't expect AMD to equal or beat Intel until they go 45nm, I do expect them to have the real world value and mainstream segments champions.

Will those of us with 65 watt Athlon X2's see enough of a performance increase to justify upgrading, or will Phenom be for new value builds and upgraders leaving socket 939 behind? The quad and triple cores might be worth it if enough games and productivy support more than 2 cores.
November 10, 2007 9:17:04 PM

elbert said:
Why is there 2 QX6800 in the second benchmark and is the 9600 beating the crap out the green under clocked version? The 9600 looks to be a float monster at only 2.3GHz. I think this version has a little more truth.
http://my.ocworkbench.com/bbs/showthread.php?%20threadid=68717


That's because two of them are only dual cores...

The QX6700 and QX6800, and in this case, Q6600 @ 2.3Ghz, are all above GP9600.
November 10, 2007 9:59:53 PM

yomamafor1 said:
That's because two of them are only dual cores...

The QX6700 and QX6800, and in this case, Q6600 @ 2.3Ghz, are all above GP9600.

No clearly the second test has 2 QX6800 one in green and the other in purple. The only dual core in the second test is the orange X2 6400+. Q6600 is no where on either test.
November 10, 2007 10:11:32 PM

oh... I tested the Q6600 @ 2.3Ghz myself.
Quote:




Again, 2.3Ghz Q6600 scored 42347 MIPS in Dhrystone (Int test), and 29590 MFLOPS in Whetstone (FP test). Since AMD has been dominating in the FP arena for a while, but what really surprised me is how low the FP score is. I was expecting somwhere along the 1.5x of Q6600 clock for clock.
November 10, 2007 10:33:07 PM


Again, 2.3Ghz Q6600 scored 42347 MIPS in Dhrystone (Int test), and 29590 MFLOPS in Whetstone (FP test). Since AMD has been dominating in the FP arena for a while, but what really surprised me is how low the FP score is. I was expecting somwhere along the 1.5x of Q6600 clock for clock.



But the QX6800 doesn't even get 40000 so something's amiss.
November 10, 2007 10:40:00 PM

BaronMatrix said:

Again, 2.3Ghz Q6600 scored 42347 MIPS in Dhrystone (Int test), and 29590 MFLOPS in Whetstone (FP test). Since AMD has been dominating in the FP arena for a while, but what really surprised me is how low the FP score is. I was expecting somwhere along the 1.5x of Q6600 clock for clock.



But the QX6800 doesn't even get 40000 so something's amiss.


Please try again! :hello: 
November 10, 2007 10:57:58 PM

BaronMatrix said:

Again, 2.3Ghz Q6600 scored 42347 MIPS in Dhrystone (Int test), and 29590 MFLOPS in Whetstone (FP test). Since AMD has been dominating in the FP arena for a while, but what really surprised me is how low the FP score is. I was expecting somwhere along the 1.5x of Q6600 clock for clock.



But the QX6800 doesn't even get 40000 so something's amiss.


You have to look at the horizontal axes..
November 10, 2007 11:21:56 PM

elbert said:
No clearly the second test has 2 QX6800 one in green and the other in purple. The only dual core in the second test is the orange X2 6400+. Q6600 is no where on either test.



That's why we tried to overclock or underclock our Q6600 to 3Ghz/2.3Ghz and run the benchmark so we can compair to their result.

I just finished benchmarking my Q6600 (underclocked to 2.3Ghz, 8800 GTS as video card) in 3DMark06, here's the result:
http://service.futuremark.com/orb/resultanalyzer.jsp?pr...

As you can see, it cannot even beat my sys. this test, here's their result with 8800 GTX as video card:
http://www.expreview.com/news/hard/2007-11-08/119451488...

Look at the CPU score on both results.

According to them, their Phmenom X4 GP-9600 used in this 3DMark06 test is a retail version...


November 10, 2007 11:30:17 PM

X-Files said:
You're right, I just re-run 3DMARK 06 again on my system and here's the result:
http://service.futuremark.com/orb/resultanalyzer.jsp?pr...

But don't forget - my CPU is oc to 3.2Ghz, their CPU is oc to only 3Ghz. I have 200Mhz advantage.


Honestly, why in hell would you downclock your E6600 to 3.0GHZ in the first place? They would have made it 3.2 if the cpu could have handle it... no? About the E6600, almost all of them do 3.2 without even sweating it in most case. I guess it pretty fair as it is.

I just hope AMD can prove that their process of "building while improving the formula" will prove it's worth just about now. Otherwise I can't see why Phenom perform so poorly frequency-wise.
November 10, 2007 11:55:52 PM

X-Files said:
You have to look at the horizontal axes..



Can I say duuhhh? I'd better not as that will empower people. I guess I've never seen a graph like that. Usually the x-axis is block size or something. Maybe they should just use bars.
November 11, 2007 1:32:51 AM

BaronMatrix said:
Can I say duuhhh? I'd better not as that will empower people. I guess I've never seen a graph like that. Usually the x-axis is block size or something. Maybe they should just use bars.


Or maybe you should take a look at the x-axis and y-axis representations before you make outrageous statement? :pt1cable:  :pt1cable: 
November 11, 2007 10:41:22 AM

yomamafor1 said:
oh... I tested the Q6600 @ 2.3Ghz myself.
Quote:

http://my.ocworkbench.com/bbs/attachment.php?attachmentid=648&stc=1&d=1194573367


Again, 2.3Ghz Q6600 scored 42347 MIPS in Dhrystone (Int test), and 29590 MFLOPS in Whetstone (FP test). Since AMD has been dominating in the FP arena for a while, but what really surprised me is how low the FP score is. I was expecting somwhere along the 1.5x of Q6600 clock for clock.

I see but the green QX6800 looks like its getting beat. Is the QX6800 green underclocked to 2.2GHz? I see that your Q6600 beats the score but this test could be with much less ram. IE the systems could only have had say 1GB were you may have 2GB's or RAM. Your system then would not make a good compare.
November 11, 2007 11:17:20 AM

The 'green' QX6800 isn't even on the charts, I don't see a green dot, why are you reading so much into it?

Cmon Elbert, use some logic - do you really think a QX6800, even underclocked to 2.2GHz, would score well below the X2 6400+ when at 3GHz it DOUBLES the X2 6400+ score?

I'm sorry, the 9600 is hardly a 'floating point monster' if it's getting matched by an underclocked Q6600, and focusing on the 'green QX6800' (whatever that represents, it's probably a bug and should be showing purple instead, notice there is no purple label box?) is not going to change that. ;) 
November 11, 2007 11:23:10 AM

epsilon84 said:
The 'green' QX6800 isn't even on the charts, I don't see a green dot, why are you reading so much into it?

Cmon Elbert, use some logic - do you really think a QX6800, even underclocked to 2.2GHz, would score well below the X2 6400+ when at 3GHz it DOUBLES the X2 6400+ score?

I'm sorry, the 9600 is hardly a 'floating point monster', and focusing on the 'green QX6800' (whatever that represents, it's probably a bug and should be purple instead) is not going to change that. ;) 

Green and purple are the same thing - QX6800 from the database. Both the horizontal and vertical values must be identical. Thus, one dot would be superimposed on the other. You happen to see purple on top.
November 11, 2007 11:24:19 AM

WR said:
Green and purple are the same thing - QX6800 from the database. Both the horizontal and vertical values must be identical. Thus, one dot would be superimposed on the other. You happen to see purple on top.


That certainly makes sense...
November 11, 2007 1:07:47 PM

elbert said:
I see but the green QX6800 looks like its getting beat. Is the QX6800 green underclocked to 2.2GHz? I see that your Q6600 beats the score but this test could be with much less ram. IE the systems could only have had say 1GB were you may have 2GB's or RAM. Your system then would not make a good compare.


Their system uses 2GB of RAMS too:
http://www.expreview.com/news/hard/2007-11-08/119451488...

You can see it in the CPU-Z picture.
November 11, 2007 1:42:50 PM

epsilon84 said:
The 'green' QX6800 isn't even on the charts, I don't see a green dot, why are you reading so much into it?

Cmon Elbert, use some logic - do you really think a QX6800, even underclocked to 2.2GHz, would score well below the X2 6400+ when at 3GHz it DOUBLES the X2 6400+ score?

I'm sorry, the 9600 is hardly a 'floating point monster' if it's getting matched by an underclocked Q6600, and focusing on the 'green QX6800' (whatever that represents, it's probably a bug and should be showing purple instead, notice there is no purple label box?) is not going to change that. ;) 

I am reading nothing into it as you can see I have a question mark at the end of my last post to yomamafor1.
Quote:
I see but the green QX6800 looks like its getting beat. Is the QX6800 green underclocked to 2.2GHz? I see that your Q6600 beats the score but this test could be with much less ram. IE the systems could only have had say 1GB were you may have 2GB's or RAM. Your system then would not make a good compare.

Using some logic it could be under the red tag and be ahead of the X2. Using some logic one may question why they are use the same CPU on a test and why was the CPU's changed from the first test.

If you note the test the 2.3GHz float for the 9600 in the first test is quite high on the chart. 115225 only trailing the QX6700 by about 40787 at 156012. That again is a 2.3GHz v/s 2.67GHz. I hardly see a comparable Q6600 on the tests and all the tests done here are unusable as no way they have the same settings and hardware. All the benchmarks I see are 3dmarks of Q6700. On the second test the FPU(MFLOPS) for the 9600 are 29630 and the QX6700 isn't posted but looks to be about in the 33~34000 range. Again a 2.3GHz v/s a 2.67GHz so yes its a 'floating point monster' and most likely equals kentsfield clock for clock in these tests.
November 11, 2007 1:49:14 PM

X-Files said:
Their system uses 2GB of RAMS too:
http://www.expreview.com/news/hard/2007-11-08/119451488...

You can see it in the CPU-Z picture.


I'm sorry but I see 1GB in the last picture.
http://my.ocworkbench.com/bbs/showthread.php?%20threadid=68717
We were talking about the ocworkbench tests as you can see below.
Quote:
yomamafor1 wrote :

oh... I tested the Q6600 @ 2.3Ghz myself.
Quote :


http://my.ocworkbench.com/bbs/atta [...] 1194573367


Again, 2.3Ghz Q6600 scored 42347 MIPS in Dhrystone (Int test), and 29590 MFLOPS in Whetstone (FP test). Since AMD has been dominating in the FP arena for a while, but what really surprised me is how low the FP score is. I was expecting somwhere along the 1.5x of Q6600 clock for clock.

I see but the green QX6800 looks like its getting beat. Is the QX6800 green underclocked to 2.2GHz? I see that your Q6600 beats the score but this test could be with much less ram. IE the systems could only have had say 1GB were you may have 2GB's or RAM. Your system then would not make a good compare.
a c 100 à CPUs
November 11, 2007 2:23:46 PM

Quote:
Id like to see benches on something other than a Nforce560. Maybe on AMD's own chipset to make it as fair as the C2Q with a Intel chipset.


It's not going to make a big difference, especially in CPU-limited tasks. The memory controller sits in the CPU, so all the motherboard does is supply power and route the RAM data lines to the CPU. It's the same as with the K8s- the motherboard makes very little difference in performance in CPU-intensive tasks. Intel motherboard have the memory controller in the northbridge on the motherboard, so your choice of motherboard can affect performance.

The only thing that the motherboard and chipset on an AMD K8/10h platform really affects is I/O to the peripherals. You would see poor GPU, disk, USB, and Ethernet performance is the southbridge on the motherboard was terrible. The poor GPU bandwidth would cause a bottleneck in GPU performance, and thus make game benchmarks perform worse WRT other boards at higher resolutions.
November 11, 2007 3:01:27 PM

All I know AMD lost in a very little gap. Unlike to the P4 and Athlon race which makes Pentium 4 => Pentium Poor 'performance'
leaving it to dust.
3Ghz Pentium 4 = 2.0Ghz Athlon (what a shame!)

Now!
3Ghz C2D is greater than 3Ghz Phenom (overclocked) in just very few gap, not a shame at all. Few more improvement like increasing its L2 cache which C2D has 4x compared to AMD surely I agree its a fair specs and no more excuse.

FYI: Phenom on the benchmark is running on a chipset that does not support HT 3.0 so some descripancies may arise and the true speed may not be unleashed, unless it will be tested under AMD790FX chipset which fully support all features of the Phenom. (SANDRA XII TEST)
@ The expreview - they should test it also to Q6600 not the latest C2D because overclocked version is not consistent to stocked version (they are not equal). Clock per clock at stocked speed, no bias at all.
November 11, 2007 3:19:46 PM

elbert said:
I'm sorry but I see 1GB in the last picture.
http://my.ocworkbench.com/bbs/showthread.php?%20threadid=68717
We were talking about the ocworkbench tests as you can see below.
Quote:
yomamafor1 wrote :

oh... I tested the Q6600 @ 2.3Ghz myself.
Quote :


http://my.ocworkbench.com/bbs/atta [...] 1194573367


Again, 2.3Ghz Q6600 scored 42347 MIPS in Dhrystone (Int test), and 29590 MFLOPS in Whetstone (FP test). Since AMD has been dominating in the FP arena for a while, but what really surprised me is how low the FP score is. I was expecting somwhere along the 1.5x of Q6600 clock for clock.

I see but the green QX6800 looks like its getting beat. Is the QX6800 green underclocked to 2.2GHz? I see that your Q6600 beats the score but this test could be with much less ram. IE the systems could only have had say 1GB were you may have 2GB's or RAM. Your system then would not make a good compare.



You're right then, it seems they used only 1GB for that benching. However, I just finished running Sandra XII again with only 1GB of RAM and Q6600 underclocked to 2.3Ghz and the result is still in line with those from yomamafor1 which is around 42231 MIPS in Dhrystone and 29390 MFLOPS in Whetstones. In fact, the amount of RAM seems to have litle or no effect at all on the result in Sandra XII (look for the result with 4GB of RAM in my previous post).

Beside, you could also download Sandra XII and try it by yourself to confirm the result, it's free. Here's the link:
http://www.softpedia.com/get/System/System-Info/SiSoftw...

November 11, 2007 3:45:49 PM

yomamafor1 said:
Quick question:

How much RAM are you using? And what was your Sis Sandra scores for Multimedia as well as Arithmetic?


I used 4GB and 1GB of RAM in Sandra XII test:

With 4GB (dual chan.) of RAM, Q6600 = 2.3Ghz: Whetstone iSSE3 = 29500 MFLOPS.

With 1GB (single chan.) of RAM, Q6600 = 2.3Ghz: Whetstone iSSE3 = 29390 MFLOPS.
November 11, 2007 4:05:28 PM

elbert said:
I see but the green QX6800 looks like its getting beat. Is the QX6800 green underclocked to 2.2GHz? I see that your Q6600 beats the score but this test could be with much less ram. IE the systems could only have had say 1GB were you may have 2GB's or RAM. Your system then would not make a good compare.


We're only trying to speculate its performance, and I can only do so by using unofficially leaked benchmarks, as well as my machine. I tried my best to eliminate discrepancies, but I'm sure some minor ones still do exist. For example, I'm using 64bit Vista while the leaked authors used XP. I corrected the discrepancies by using 32bit benchmarks, but I don't know what kind of impact would it have on the scores if the program is still ran in 64bit OS.

Here are the scores with 4Gb RAM:
Processor Arithmetic:
-Dhrystone ALU: 42347 MIPS
-Whetstone iSSE3: 29590 MFLOPS

Processor Multi-Media:
-Int x8: 253133 iit/s (void)
-Float x4: 137771 fit/s

Here are the scores with 2Gb RAM:
Processor Arithmetic:
-Dhrystone ALU: 41611 MIPS
-Whetstone iSSE3: 29628 MFLOPS

Processor Multi-Media:
-Int x8: 253373 iit/s (void)
-Float x4: 137503 fit/s

So as you can see, the 2Gb results are pretty much inline with the 4Gb results. Sis Sandra said the error margin are within 5%~10%. Therefore these results are still valid.

Now let's compare with Phenom's results:
Processor Arithmetic:
-Dhrystone ALU: 35235 MIPS
-Whetstone iSSE3: 29630 MFLOPS

Processor Multi-Media:
-Int x4: 87564 iit/s (void)
-Float x4: 115225 fit/s
November 11, 2007 10:03:13 PM

yomamafor1 said:
We're only trying to speculate its performance, and I can only do so by using unofficially leaked benchmarks, as well as my machine. I tried my best to eliminate discrepancies, but I'm sure some minor ones still do exist. For example, I'm using 64bit Vista while the leaked authors used XP. I corrected the discrepancies by using 32bit benchmarks, but I don't know what kind of impact would it have on the scores if the program is still ran in 64bit OS.

Here are the scores with 4Gb RAM:
Processor Arithmetic:
-Dhrystone ALU: 42347 MIPS
-Whetstone iSSE3: 29590 MFLOPS

Processor Multi-Media:
-Int x8: 253133 iit/s (void)
-Float x4: 137771 fit/s

Here are the scores with 2Gb RAM:
Processor Arithmetic:
-Dhrystone ALU: 41611 MIPS
-Whetstone iSSE3: 29628 MFLOPS

Processor Multi-Media:
-Int x8: 253373 iit/s (void)
-Float x4: 137503 fit/s

So as you can see, the 2Gb results are pretty much inline with the 4Gb results. Sis Sandra said the error margin are within 5%~10%. Therefore these results are still valid.

Now let's compare with Phenom's results:
Processor Arithmetic:
-Dhrystone ALU: 35235 MIPS
-Whetstone iSSE3: 29630 MFLOPS

Processor Multi-Media:
-Int x4: 87564 iit/s (void)
-Float x4: 115225 fit/s

That is about as in depth as anyone can ask. You have done everything possible to validate your benchmark. It looks to me given the margin of error that Intel has a noticeable advantage in Dhrystone ALU. AMD's biggest problem seems to be the massive difference in the Int x4. I don't think AMD has a shot at this test give barcy also got beat on Sandra.
November 11, 2007 10:23:09 PM

X-Files said:
You're right then, it seems they used only 1GB for that benching. However, I just finished running Sandra XII again with only 1GB of RAM and Q6600 underclocked to 2.3Ghz and the result is still in line with those from yomamafor1 which is around 42231 MIPS in Dhrystone and 29390 MFLOPS in Whetstones. In fact, the amount of RAM seems to have litle or no effect at all on the result in Sandra XII (look for the result with 4GB of RAM in my previous post).

Beside, you could also download Sandra XII and try it by yourself to confirm the result, it's free. Here's the link:
http://www.softpedia.com/get/System/System-Info/SiSoftw...

The 29390 MFlops in Whetstones may be low enough for Phenom win but given its single channel that would be a stretch. yomamafor1's tests seem to have them at even match for the 9600.

I have used sandra a bit in my lectures and its a pretty good program for burn in tests. I neither have access to a quad here or at the school. Administration only likes the cheap dells. Guess I just have to remain envious until I can afford 1. I'm wanting one of the 2.5GHz Penryn and with any luck I should have the money to buy 1 around the time their set to launch.
November 11, 2007 10:38:24 PM

If it's so close to "launch" why aren't there real benchmarks?
November 11, 2007 11:31:16 PM

TechnologyCoordinator said:
If it's so close to "launch" why aren't there real benchmarks?


Well lets see...could it be that AMD's claim of K10's 40% advantage over Core2 was a lie and now they are waiting till the last minute. Why do that? Well I remember when barcelona was released they had a couple of top executives quit, now with phenom release I expect a couple of more rats to bail.
November 11, 2007 11:43:28 PM

qurious69ss said:
Well lets see...could it be that AMD's claim of K10's 40% advantage over Core2 was a lie and now they are waiting till the last minute. Why do that? Well I remember when barcelona was released they had a couple of top executives quit, now with phenom release I expect a couple of more rats to bail.

That's not nice...AMD is supposedly trying their best here, albeit using underhand methods such as lying. I'm sure if the situation was flip flopped, Intel would do the same. They are trying to compete against a much larger company, so give them some slack here...
a c 100 à CPUs
November 12, 2007 2:22:54 AM

pogsnet said:

Now!
3Ghz C2D is greater than 3Ghz Phenom (overclocked) in just very few gap, not a shame at all. Few more improvement like increasing its L2 cache which C2D has 4x compared to AMD surely I agree its a fair specs and no more excuse.


First, AMD's cache size-performance ratio is different than Intel's because the memory accessing and cache setup is done much differently. Intel has a single 128-bit memory controller in the northbridge that runs at the system bus speed (200/266/333 MHz) and communicates to the CPU over a quad-pumped, 64-bit FSB. The CPU has a two-level cache arrangement with a shared, inclusive L2 cache and discrete L1 data and instruction caches. AMD's CPUs have two independent 64-bit memory controllers on die running at 200 MHz under core clock speed, a three-level cache structure with a shared victim L3 cache and discrete inclusive L2 caches and then L1 D and I caches for each core.

Secondly, it would be very hard for AMD to increase cache size at the moment. The quad-core Phenom/Opteron is already a 285 mm^2 chip, which is quite large. Adding more cache would just make it bigger, more expensive, and harder to get good yields from. AMD pretty well has to wait until they get to the 45 nm process node to increase the cache. Roadmaps have 12 MB L3 on 45 nm quad-core Opterons.

Quote:
FYI: Phenom on the benchmark is running on a chipset that does not support HT 3.0 so some descripancies may arise and the true speed may not be unleashed, unless it will be tested under AMD790FX chipset which fully support all features of the Phenom. (SANDRA XII TEST)


HyperTransport doesn't do much for a single-socket chip like the Phenom. It serves only to communicate with the southbridge, which doesn't even saturate the existing 2.0 GT/sec (1000 MHz) HT links. Now if you're talking about the Opteron 23xx/83xx which use NUMA over HT to do remote memory access, then it is a big deal. About the only thing the Phenoms gain from the new 790-chipset AM2+ boards would be support for split IMC/CPU power planes and clock speeds, which lets the memory controller run faster than it would on a single-plane AM2 board. This can increase performance.

Quote:
@ The expreview - they should test it also to Q6600 not the latest C2D because overclocked version is not consistent to stocked version (they are not equal). Clock per clock at stocked speed, no bias at all.


The IPC of a chip should not change with the clock speed unless you are running into some serious bus/memory bottleneck issues. That does not seem to be the case with the bandwidth offered by DDR2-800+ being more than enough to feed CPUs out there today. Plus, you generally overclock the RAM when you overclock the CPU by increasing system bus speed unless you have a CPU with an upwardly-adjustable multiplier. So overclocking the CPU means that the overall performance goes up with the CPU's clock speed.

I understand what you were getting at. Benchmarking a stock CPU versus an overclocked one gives a relative performance advantage to the overclocked CPU. Benching two overclocked CPUs against each other isn't indicative of what's actually being sold. But the benchmarks still have some merit as a lot of people around here do overclock. They don't care about what a 2.40 GHz Phenom X4 will do against a 2.40 GHz Q6600 because they'll run the chips at higher (and relatively different) clock speeds. If the 2.40 Phenom beats the 2.40 Q6600 by a small margin but only gets to a a top 3.00 GHz OC while the Q6600 gets to 3.40 GHz, then the Q6600 is faster. It all depends on what you intend to do with your CPUs, which is why benchmarks are run at both stock and overclocked speeds.
November 12, 2007 2:38:52 AM

Evilonigiri said:
That's not nice...AMD is supposedly trying their best here, albeit using underhand methods such as lying. I'm sure if the situation was flip flopped, Intel would do the same. They are trying to compete against a much larger company, so give them some slack here...


No slack. They are trying to seperate people from their hard earned money by promising product perfromance they dont deliver on. Intel did do the same thing with crapburst. It cost them almost 30% of the total market and rightly so.

Honesty goes much further than BS. Ultimately, either AMD or Intle can spread all the BS they want. There are too many independent tech sites that will expose the truth, and as the past 18 months have proven, the vast majority of people/busineses will gravitate towards that which they want, high performance, value, or some balance of the 2. And no amount of BS will mask poor performance or value, not anymore. It can only bring shame and bad PR to a company.
!