AMD's 3GHz K10 to break 30,000 3DMark06 (Inq)

[SEA]

Well maybe that is the truth behind all of this. They were actually doing some testing with 3 Graphic cards. So maybe Theo did not notice it.

Maybe. But original article states this explicitelly:

The particular processor was none other than a single socket Barcelona or Agena FX, call it what you will. The reference motherboard containing RD790 chipset packed two HD 2900XT cards, and the memory installed was Corsair’s Dominator PC2-9136C5D, or the same ones we have been using ever since they came out. There was a Raptor hard drive, and that was about that. OCZ will like the fact that PP&C Quaddie CrossFire PSU was installed in the system.

[onewingedangel]

I find this extremely unlikely, but would like to be proven wrong

PCI-E 2.0 would add nothing - r600 is only pci-e x16 so no additional bandwidth is made avaliable to the cards.

It's possible (though doubtful) that x2900 crossfire has been platform limited all along as you either have to go for pcie x8/x8 on i975 or resort to the rd600's hit and miss overclocking in order to pair it with the fastest processors. Of course someone with a x38 board could test this possibility by comparing dual x16 to dual x8.

[Stapler]

Isn't stock frequency for PCIe 2.0 175mhz, so it would infact increase 3dmark score, but I don't think it'll do 30k on air.

[[XC] flat-four]

I'm sure someone, somewhere is reading this thread and laughing because they have a Phenom in their possession and know what the results are either way.

[Shintai]

How much would another graphics card add to the score?

Roughly it would add enough that K10 would be around C2Q speeds +/- 10-20%. Ofcourse we need tripple CF numbers to know precisely.

But its all guesswork ofcourse.

[SEA]

In addition: I think that the difference 23k and 30k in 06marks is for sum of processor, video and onboard memory, and PCI-E 2.0 of course.
I mean 23k for 2.5GHz CPU, default GPU/VidMem, default DDR2
30k for all OC-ed, including DDR2 @ 1066 MHz.

Khm. Don't forget single die 4 cores, granting better scalability for 4 cores.

[cadaveca]

I will only make one comment on this.


Intel crossfire = 8x8 PCI-E 1.0 + DDR2/3 MemoryControl on Northbridge.

RD790 Crossfire = 16x16 PCI-E 2.0 + DDR2 IMC.


Now, most comments about HD2900XT performance don't have the cards. I do.

Nobody has posted RD580(16x16 SKT 939/AM2) 2900XT Crossfire numbers.


BUT....

@ 3ghz K8 SKT939, RD580, DDR1 @ 250mhz 3-3-3-8, X1900Crossfire =10.5k '06.

http://service.futuremark.com/compare?3dm06=376335


@ 3ghz Core2, SKT 775, D975X, DDR2 @ 833mhz 4-4-4-8, X1900Crossfire =10.2k '06.


http://service.futuremark.com/compare?3dm06=639467


Now, although these numbers are NOT R600, they illustrate how limits imposed by PCI-E bandwidth @ 8x8 afect Intel's 3dmark scores.


Like it or lump it, clock for clock AMD really isn't that far behind Intel in '06. The only thing currently holding AMD back in benching is lack of high clockspeeds on K8. Add in IPC improvements on the new AMD core, higher PCI-e Speed, and greater overall bandwidth, even without the extra PCI-E speed...things don't seem too "fake" to me....they seem right where I expected, actually.

[Lane-k]

You are not right, The 2900Xt have been designed at start for PCI ex 2.0, (but no mobo was available).. ask overlclockers, the card begin just fly with 150mhz on PCI express. More you increase the PCI ex frequencies, more the performance are.. (and this without the bandwith gain of a 16x 16x platform)

Second thing, i don't know if we can trust 30K with this K10 but the first result with 23.7K look for me totally possible.. Now, the only thing we don't know, is how the K10 scale with overclocking, A64 was scale with low increase of frequencies.. from 2.8 -> 3ghz -> 3.2 etc.. totally invert of old Intel and CD2, who scale from 2.66-2.93 -> 4ghz -> 5ghz... More the architecture of K10 with a 4 in one QUad ( vs the 2x dual core side by side of the QX intel) scale perhaps better of the increase of frequencies.. don't forget the Intel was the first Quad architecture and we have in real no idea of the real efficacity of the architecture, like we know only this one, and we have no other point of comparaison... Look like 4x CPU (or 2 x dual core) on professional platftorm, they have a better render of efficacity in multithreaded appliccation compared to the Quad Intel... and better performance too

Anyway i don't say it's true, i just say not impossible at all.

[Nedjo]

Looks very, VERY fake.

Theo doesn't got any reason to gamble with his reputation and to make up some fake figures!

[madcho]

How much would another graphics card add to the score?

nothing, 3 graphics card are physic crossfire, 4 graphics card are normal crossfire. don't thing they used more than 2.

[SEA]

Theo doesn't got any reason to gamble with his reputation and to make up some fake figures!

Right, particularly just 10 days before K10 launch.

[Metroid]

It is difficult to know how RD790 crossfire, bios and most of the settings were configured as how it works, Shamino and Kinc got 27000 with a pair of ASUS Radeon HD 2900 XT overclocked to 1175/950MHz and with a Kentsfield at 5.1 GHz.

Total Points: 27039

SM 2.0 Score 11273 Marks
SM 3.0 Score 12088 Marks
CPU Score 7619 Marks

If Phenom is far superior than Kentsfied it is clearly possible. CPU score was 7619 marks, maybe a fast IMC and some advantages combined would result in this outcome. It is hard to believe and I doubt that a Phenon clocked at 3.0 GHz is faster than a 5.1 GHz kentsfield.

[cadaveca]

It is hard to believe and I doubt that a Phenon clocked at 3.0 GHz is faster than a 5.1 GHz kentsfield.

Maybe the fault in your thinking is that the increase in score is due to cpu alone...and I'll tell ya something...it's not.

[kemo]

I will only make one comment on this.


Intel crossfire = 8x8 PCI-E 1.0 + DDR2/3 MemoryControl on Northbridge.

RD790 Crossfire = 16x16 PCI-E 2.0 + DDR2 IMC.


Now, most comments about HD2900XT performance don't have the cards. I do.

Nobody has posted RD580(16x16 SKT 939/AM2) 2900XT Crossfire numbers.


BUT....

@ 3ghz K8 SKT939, RD580, DDR1 @ 250mhz 3-3-3-8, X1900Crossfire =10.5k '06.

http://service.futuremark.com/compare?3dm06=376335


@ 3ghz Core2, SKT 775, D975X, DDR2 @ 833mhz 4-4-4-8, X1900Crossfire =10.2k '06.


http://service.futuremark.com/compare?3dm06=639467


Now, although these numbers are NOT R600, they illustrate how limits imposed by PCI-E bandwidth @ 8x8 afect Intel's 3dmark scores.


Like it or lump it, clock for clock AMD really isn't that far behind Intel in '06. The only thing currently holding AMD back in benching is lack of high clockspeeds on K8. Add in IPC improvements on the new AMD core, higher PCI-e Speed, and greater overall bandwidth, even without the extra PCI-E speed...things don't seem too "fake" to me....they seem right where I expected, actually.

System 1:
Radeon X1900 XT CrossFire
1GB (512MBx2) DDR2 800 4-4-4-4-12
2.66GHz Core 2 Duo
Intel D975XBX "BadAxe"
Maxtor 6L300SO HD
Antec 550w PSU

System 2:
Windows XP SP2
ATI Catalyst v6.6 Beta System 2:
Radeon X1900 XT CrossFire
1GB (512MBx2) DDR 400 2-2-2-5
A64 FX-60 @ 2.8GHz
Asus A8R32-MVP Deluxe
Maxtor 6L300SO HD
Antec 550w PSU

Windows XP SP2
ATI Catalyst v6.6 Beta



i thought the Asus A8R32-MVP Deluxe use dual 16x slot ,,, i think you forgot the card clocks on the C2D run

[informal]

I emailed Theo about 3-Fire question.I will post his response as soon as i get it .

PS I believe Theo is telling the truth(ie. K10 scored 30K+ with 2 CF cards)

[onewingedangel]

Now, although these numbers are NOT R600, they illustrate how limits imposed by PCI-E bandwidth @ 8x8 afect Intel's 3dmark scores.

anyone with a x38 board to test?

[biohead]

You are not right, The 2900Xt have been designed at start for PCI ex 2.0, (but no mobo was available).. ask overlclockers, the card begin just fly with 150mhz on PCI express. More you increase the PCI ex frequencies, more the performance are.. (and this without the bandwith gain of a 16x 16x platform)

that is correct when dealing with extreme frequencies, not with normal air clocks (~850/1000)

[Metroid]

I emailed Theo about 3-Fire question.I will post his response as soon as i get it .

PS I believe Theo is telling the truth(ie. K10 scored 30K+ with 2 CF cards)

I believe that too. So are 2 believers along.

[kemo]

I believe that too. So are 2 believers along.

i want to believe ,,,,,,,,, but still i cant

[Vapor]

What does C2Q at 3GHz and 2900XTs (OC'd similarly) get at 640x480, 800x600, and 1024x768?

At least then we'll get some baselines (and yeah...might as well throw in default 1280x1024 in for kicks, though I SERIOUSLY doubt 30k was achieved at 1280x1024).

[Baron]

My one and only comment.

Its the Inq. Ill believe it when i see it from a more reputable source.

End of.

[kemo]

What does C2Q at 3GHz and 2900XTs (OC'd similarly) get at 640x480, 800x600, and 1024x768?

At least then we'll get some baselines (and yeah...might as well throw in default 1280x1024 in for kicks, though I SERIOUSLY doubt 30k was achieved at 1280x1024).

now i believe 640x480

[Yoxxy]

I hope it is not 3 graphics cards, because then benchmarking has become even more expensive...

[Metroid]

now i believe 640x480

lol. Well when people make benchmarks without even state its resolution, meaning as default at 1280 x 1024, but who really knows

[Gautam]

that is correct when dealing with extreme frequencies, not with normal air clocks (~850/1000)

Quite the opposite from what I've found. I can crank my PCI-E speeds up quite easily below 900MHz core. Around 1000MHz things get significantly harder. The gains are always pretty decent, though they sometimes get blown out of proportion.

This is all CrossFire mind you, I haven't done much single card testing TBH.