AMD: R7xx Speculation

[XMAN26]

Heh, heh

You know I don't understand why the 3870X2 has such a bad rep. Looking over some recent numbers it consistently outpaces Nvidia's fastest single GPU cards. There are a few instances where multi-GPU rendering breaks down but for the most part it seems to do pretty well. Since Nvidia won't have a GX2 this time around it looks like AMD should be able to claim the performance crown relatively easy.

Really?

http://www.hardocp.com/article.html?art=MTUwNywzLCxoZW50aHVzaWFzdA==

They used the 8.5s in these tests.

 

[Pete]

So, it seems 4870 will be twice as fast as a 3870. That implies twice the texture units, no?

 

[AlphaWolf]

So, it seems 4870 will be twice as fast as a 3870. That implies twice the texture units, no?

Clock speed?

 

[Anarchist4000]

Clock speed?

>1.5GHz core clock?

Do we know that AFR is being used on the 4870x2 cards? Would be interesting to know if the x2 cards are using AFR or something else. hint hint

 

[trinibwoy]

Really?

http://www.hardocp.com/article.html?art=MTUwNywzLCxoZW50aHVzaWFzdA==

They used the 8.5s in these tests.

Yeah I guess it doesn't look too hot there. But check out computerbase.de's latest stuff that uses 8.5 as well. They test a much broader range of titles.

 

[Pete]

Yeah, I haven't read up on Vantage so I don't know how much AF limits performance. 4870'll have more bandwidth than 3870 either way, but a 1GHz clock is just a 30% bump in texturing performance assuming the same # of (equally-capable--i.e., no doubling of INT8 performance) units. I guess it would've been safer to assume Vantage is still harder on shader than texture units (and moreso as the res increases, from what I've gleaned from hanging around here), but it's still a little surprising to think* of Vantage performance scaling purely with shader speed (480/320 * 1.0/.775 = 1.95x), not hampered by texturing performance at all.

On the one hand, hoom's #s show a big hit with AF--albeit in 3DM06. OTOH, those CB.de AF-only #s Jawed linked don't show as big of a drop as I expected, given the speculation that it wasn't AA but AF that was hampering ATI's AA+AF #s. (I'm ignoring the "HQ" perf. drop, as disabling CatAI may introduce other variables.)

* By think I mean with my gut, not my brain.

 

[AlphaWolf]

>1.5GHz core clock?

It couldn't be some combination of clock, number of units and other improvements?

 

[Anarchist4000]

I'm sure it's some combination of improvements but just 16 beefed up texture units don't seem like they'd be enough.

 

[Wirmish]

Vantage Xtreme profile:
GeForce GTX 280 ······· X4100
GeForce GTX 260 ······· X3800
GeForce 9800GX2 ······· X3600
GeForce 8800 Ultra ····· X2400
Radeon HD 4870 XT ···· X2600
Radeon HD 3870X2 ····· X2500
Radeon HD 4850 Pro ··· X2000
Radeon HD 3870 ········ X1400

1. HD3870 = X1400, HD3870X2 = X2500 -> HD2870X2 / 1.78 = X1400
2. CJ -> 2 * X2600 (R700) = X5200+
3. X5200+ / 1.78 = X2900+ (HD4870)
4. X2900+ (HD4870) = ~2 * HD3870 !!

Wow! ... if it's true.

"If it seems too good to be true, it probably is." - a Reader's Digest sweepstakes "winner"

 

[ZerazaX]

While those 4870 and 4850 numbers from that link were stated fake by CJ, he did claim those GTX280/260 numbers and if the 4870X2 at 5200+ is true, those are some massive numbers

 

[XMAN26]

Yeah, I haven't read up on Vantage so I don't know how much AF limits performance. 4870'll have more bandwidth than 3870 either way, but a 1GHz clock is just a 30% bump in texturing performance assuming the same # of (equally-capable--i.e., no doubling of INT8 performance) units. I guess it would've been safer to assume Vantage is still harder on shader than texture units (and moreso as the res increases, from what I've gleaned from hanging around here), but it's still a little surprising to think* of Vantage performance scaling purely with shader speed (480/320 * 1.0/.775 = 1.95x), not hampered by texturing performance at all.

On the one hand, hoom's #s show a big hit with AF--albeit in 3DM06. OTOH, those CB.de AF-only #s Jawed linked don't show as big of a drop as I expected, given the speculation that it wasn't AA but AF that was hampering ATI's AA+AF #s. (I'm ignoring the "HQ" perf. drop, as disabling CatAI may introduce other variables.)

* By think I mean with my gut, not my brain.

http://www.pcgameshardware.de/aid,6...icle&image_id=826013&article_id=644504&page=1

http://www.pcgameshardware.de/aid,6...r_Catalyst_85_im_Hands-On-Test_Update/&page=1

Found this, OC'd Quad and an X2 and the 8.5s seem to hurt performance.

 

[Arty]

I'm getting a sneaky suspicion that CJ's alarm clock didnt go off.

Spoiler

Or he might have hit the snooze button!

 

[annihilator]

Vantage Xtreme profile:
GeForce GTX 280 ······· X4100
GeForce GTX 260 ······· X3800
GeForce 9800GX2 ······· X3600
GeForce 8800 Ultra ····· X2400
Radeon HD 4870 XT ···· X2600
Radeon HD 3870X2 ····· X2500
Radeon HD 4850 Pro ··· X2000
Radeon HD 3870 ········ X1400

1. HD3870 = X1400, HD3870X2 = X2500 -> HD2870X2 / 1.78 = X1400
2. CJ -> 2 * X2600 (R700) = X5200+
3. X5200+ / 1.78 = X2900+ (HD4870)
4. X2900+ (HD4870) = ~2 * HD3870 !!

Wow! ... if it's true.

I don't know why the idea of 4870 being two times the fast as 3870 is weird to so many people. If the rumors are correct, it's kind of obvious. Everything just seems doubled; which isn't the biggest of deals since NVidia seems to be doubling as well ((at humongous price points)

 

[Lukfi]

(from the GT200 thread)

Lukfi, there's another thread for that, but if they could get a dual-GPU R700 idling like a single GT280, I'd be impressed.

Very low consumption in idle is something I deem almost certain for RV770-based cards. I think idle power draw around 30 watts should be doable with R700.

 

[CJ]

I'm getting a sneaky suspicion that CJ's alarm clock didnt go off.

Spoiler

Or he might have hit the snooze button!

You were right.

Anyway, I got some updates.

The X4100 number for the GTX 280 was correct with the drivers at the time. Seems that NV was able to push it to X4800 with the latest drivers. They might add some more points when they enable PhysX on their GPU, but that kinda goes against Futuremark rules... since a PPU can assist for a higher score and not a GPU... this could add up to 1000 points in the CPU test... so we'll have to see what Futuremark decides about including the PhysX into the score....

But here's what you all wanted to hear.... According to the same source R700 scores about X5500> in Vantage.
Oh and my buddies at VR-Zone tested a HD4850 OC 700Mhz. P64xx score in Vantage.

 

[mozmo]

http://vr-zone.com/articles/Radeon_HD_4850_(RV770PRO)_Tested/5829.html

We have spotted a RV770PRO demo system over at Computex and we have checked out the settings of the beta Catalyst drivers for RV770 series. The card can be easily overclocked to 700MHz core and as a result, the score for 3DMark Vantage is P6466.

The vr-zone link CJ mentioned.

 

[igg]

http://vr-zone.com/articles/Radeon_HD_4850_(RV770PRO)_Tested/5829.html

The vr-zone link CJ mentioned.

Whats the PXXXX score of the X2 or GTX280?

 

[Pete]

HD3850 scores ~P4100, which is 12x10 1x1. It goes down to ~H2150, which is 16x10 2x8 (among other things).

A 3870 scores ~P4700 and ~X1500 here, same CPU (QX9770).

 

[Arun]

3850 scores ~P4100....

On a: Intel Core 2 Extreme QX9770 (3.2GHz)

 

[Lukfi]

Did anyone notice the core runs at 625 MHz? (Now where does that lead to?)