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

[Shadowmage]

One of the GPUs can be use as a PPU

That's what I thought. So how can the 3rd graphics card improve the 3dmark06 score?

[CraptacularOne]

That's what I thought. So how can the 3rd graphics card improve the 3dmark06 score?

Because it's not exclusively used as a PPU.

[adamsleath]

rd790 mobo's in september ;
i cant wait for rumours to be put to rest one way or the other.

[Zytek_Fan]

Because it's not exclusively used as a PPU.

Yep. I'm not going to agree that they used 3 GPUs, because the Phenom rig photographed at GC 07 had 2.

Anyway, ATi developed a way to where the workload is spread across the 3 GPUs, each card generates different parts of the images, then the data is transferred through the PCIe slot or CrossFire connector (can't remember which, but it's one of those) to the master card (the top card, not the master card with the stupid dongle) where the images are put together and sent to your monitor.

[Donnie27]

That's what I thought. So how can the 3rd graphics card improve the 3dmark06 score?

Because it can increase the performance of the already fast Crossfire with Dual GPU. Folks are also using these to do CPU work.

http://www.bit-tech.net/news/2007/07...ed_ppu_cards/1

[Baron]

One of the GPUs can be use as a PPU

Same as with the Nvidia Tri SLI if they ever get round to doing it.

*see the pic in my sig*

[Zytek_Fan]

Same as with the Nvidia Tri SLI if they ever get round to doing it.

*see the pic in my sig*

WTF does it actually work? I know 8 series can do physics, but does it really work?

[Baron]

WTF does it actually work? I know 8 series can do physics, but does it really work?

Sadly not.

[Donnie27]

Yep. I'm not going to agree that they used 3 GPUs, because the Phenom rig photographed at GC 07 had 2.

Anyway, ATi developed a way to where the workload is spread across the 3 GPUs, each card generates different parts of the images, then the data is transferred through the PCIe slot or CrossFire connector (can't remember which, but it's one of those) to the master card where the images are put together and sent to your monitor.

But the other rig did have 3. I didn't see this one and I'm not sure what the setup was. Yet, that what the problem is here to start with. In 11 days we'll know. Some will say AMD was smart no matter what happens.
If it's fast, we see;
See, AMD is smart to have held this back, it rocks.
If it's slow we'll see;
AMD is smart to have hid this performance from the market.

What K10 does though, it will be great that the hiding and crap will be over at last. Maybe folks will be nicer?

[CraptacularOne]

Sadly not.

It works but they haven't developed a game yet that does physics through the use of shaders.

[Donnie27]

Same as with the Nvidia Tri SLI if they ever get round to doing it.

*see the pic in my sig*

Holy Crap that's sweet!

[LordEC911]

It works but they haven't developed a game yet that does physics through the use of shaders.

Plus no drivers other then the very, very early betas.

[Zytek_Fan]

Hey! Who's thread crapping by rating this thread bad? We've actually got a serious, good discussion about Phenom going on

[LordEC911]

Hey! Who's thread crapping by rating this thread bad? We've actually got a serious, good discussion about Phenom going on

All the flaming/trolling didn't help, at least the mods got that straightened out.

[oohms]

Wow its looking better by the day

I just hope they don't have too many yield problems with that huuge die and they can get it to scale past 3 ghz.

[dinos22]

Assuming this is true:
http://www.tweaktown.com/news/8067/w...emo/index.html
6400+ with 2x2900XT on 790G gave 12000 3DMark06.
Only the CPU change won't give 18K, but can the third card, and OCing the 3 cards give 12K even?

it's true as i was stationed just on the next bench table doing our stuff

http://www.tweaktown.com/news/8068/w...ghz/index.html

that's kayl with the chicks
i did the benching of course hhahahahah

[erwinz]

it's true as i was stationed just on the next bench table doing our stuff

http://www.tweaktown.com/news/8068/w...ghz/index.html

that's kayl with the chicks
i did the benching of course hhahahahah

nice dinos22 hehehe why are not with the girls..

[cky2k6]

heh, i just noticed inquirer disabled the e-mail page. i'm sure someone will share his email addy so that we can spam it to death if this turns out to be bull

[dinos22]

nice dinos22 hehehe why are not with the girls..

hahah i was too busy behind them checking our their arses j/k

kayl's a showman i'm an ocer.....

[Zytek_Fan]

heh, i just noticed inquirer disabled the e-mail page. i'm sure someone will share his email addy so that we can spam it to death if this turns out to be bull

Theo didn't want hate mail

[hurleybird]

That's what I thought. So how can the 3rd graphics card improve the 3dmark06 score?

I have a better question. How the hell is 'Trifire' even supposed to work with cat 7.7? It seems odd that such an old driver would support the new Crossfire, no? Also, correct me if I'm wrong, but Trifire has only been showcased on 3GHz systems, right? IIRC this system was a 2.5GHz system that got OCed to 3GHz.

[IntelUser]

Time for analyzation:

From HKEPC:

http://www.hkepc.com/bbs/news.php?ti...me=0&endtime=0

-Dual Crossfire gives 1.8x speed of single 2900XT
-Tri-Fire gives 2.6x speed of single 2900XT

1.8x/2.6x=69.2%

69.2% of 30k=20769

A more reasonable score.

BTW, one of AMD's presentation shows IPC improvements of 15% over K8, so I doubt it was done with dual x-fire.

http://www.ptcinnovationforum.ru/documents/AMD.pdf

[IntelUser]

English version: http://www.sunmicrosystems.se/virtua...f_Nordlund.pdf

[Aussie FX]

Could it crack 30,000 with an instuction set like say, SSE5?

[savantu]

savantu,

What is AMD's L2 latency on K10

Don't know , I expect it to be 12-14 cycles.The L3 is slow because you need arbitration and that eats plenty of cycles for 4 cores.
The solution is to make it large ( slow L3s are typically 16-64MB ) but the die is already huge.

You can imagine the L3 trashing when all 4 cores chew along different datasets...