ATI Radeon HD 4000 Series discussion

[STaRGaZeR]

One thing is clear in all those reviews: AA scaling is so much better in the 4850 compared to RV670. The perfomance hit is on par with NV cards, which is nice.

[Hornet331]

I don't think the PCGH article setup is bad, I mean it clearly shows that at times even the GTX280 somehow loses to the 4850, such as my Oblivion example.

The PCPER article should've clearly stated the 8800GT and 9800GTX were OC'd versions to eliminate confusion, but I guess they did a rush job to put the article up. So either way you look at it though, the 4850 hangs and even beats OC'd versions of the GTX.

nah they wernt oced, at least to there test system specs:

Testsystem und Konfiguration
CPU: Intel Core 2 Duo E8500 @ 3.600 MHz (400x9)
Board: Asus P5N-D (Nforce-750i-SLI-Chipsatz)
RAM: 4x 1.024 MiByte DDR2-800 (5-5-5-15)
OS: Windows Vista 64 Bit samt SP1
Driver:
• Forceware 177.34 (HQ)
• Catalyst 8.5 respektive 8.6 bei HD 4850 (AI def.)

VGA:
• Geforce GTX 280, 1.024 MiB GDDR3, 602/1.296/1.107 MHz
• Geforce 9800 GX2, 2x 512 MiB GDDR3, 600/1.512/1.000 MHz
• Geforce 8800 Ultra, 768 MiB GDDR3, 612/1.512/1.080 MHz
• Geforce 9800 GTX, 512 MiB GDDR3, 675/1.674/1.100 MHz

• HD 4850, 512 MiB GDDR3, 625/993 MHz (inklusive Crossfire)
• HD 3870 X2, 2x 512 MiB GDDR3, 825/901 MHz
• HD 3870, 512 MiB GDDR4, 776/1.125 MHz

but they used win vista for all tests

http://www.pcgameshardware.de/aid,64...=648091&page=1

[zerazax]

nah they wernt oced, at least to there test system specs:



but they used win vista for all tests

http://www.pcgameshardware.de/aid,64...=648091&page=1

I said the PCPER article, where they used the BFG 8800GT OCX and BFG 9800GTX and didnt state those two cards were both OC'd.

The PCGH article didn't use OC'd cards. But i'm still confused as to how the 4850 straight up beat the GTX280 while using the Qarls Texture Pack at 1920 x 1200. I thought cards with < 700MB choke with that pack, but the 4850 is eating a 1 GB card at 8xAA no less.

[cadaveca]

Perhaps, but you have no way of knowing if ATI has found away around that problem. This info certainly adds weight to the rumors of the two RV770s of R700 communicating to each other through memory though.

You are completely right...however I have been watching GDDR5 since quimonda first released specs(in such a way that I know that up pic there is from a quimonda document)...I fully understand it's possible(I was saying yes months ago, I had a big part in the shared framebuffer rumour, when ATI guys were saying no), but I question only whether the pcb design will affect overall cost as much as it should...given the intended prices for such cards. I see the same cost as making 512-bit membus pcb...which gives me a $575 retail price for 4870x2's...I am hoping for less cost..$499...

Again, I'm the ATI fanboi, but i like to keep it real. If they have pulled it off as I hinted like months ago, they've got nVidia under thier boot. If they don't have shared framebuffer, i'll still buy if i get more than 512mb of framebuffer for each gpu.

[ghost101]

lulz, why does it sound like fanboy talk... review is not like it was expected, so ignore it?

maybe you would consider that they use other bench settings (custom time demo etc.) befor saying they are rubish....

The PCGH review is the only inconsistent review out there. Yes, they could be the only review with the correct figures, but im gonna take the most likely possibility, there was a problem in their test setup. When you get the 4850 barely outperforming the hd 3870 and even falling behin in the higher res cod4 benchmark, it should tell you something. Am i a fanboy? I dont really care but I think the 9800gtx+ will do better than the hd4850 as the reviews show us, at normal resolutions of 1280*1024 and 1680*1050, the 9800gtx at the moment is pretty much neck and neck.

Yes there custom time demo might be different, but why should the relative positions be so significantly different? It would only affect absolute figures. Look at the pcgh graphs, the 4850 is absolutely awful in some of the games.

The PCGH article didn't use OC'd cards. But i'm still confused as to how the 4850 straight up beat the GTX280 while using the Qarls Texture Pack at 1920 x 1200. I thought cards with < 700MB choke with that pack, but the 4850 is eating a 1 GB card at 8xAA no less.

The HD3870x2 also does abnormally well as well. Maybe somethign to do with ATI drivers being better optimised in that particular scenario.

[Xello]

which gives me a $575 retail price for 4870x2's...I am hoping for less cost..$499...

Didn't all speculation so far point to a $599 launch for the dual gpu card? It's not out of the realm of possibility, it would still be cheaper than the 280, after all.

Is August still the best guess for a launch on it, anyone?

[cadaveca]

I don't consider the competition. I have rigs from both makers, but really could care less about what nvidia offers in this respect. Two different companies with different goals leads to very different products, and different apps will be better on either system.

but playing the ATI fanboi, I won't talk nV stuff often.


August seems fair...if 4870 is really dealyed until july though...4807x2 may take longer as board partners are responsible for pcb design, no? they've not had the gpu's for that long yet...

[Unbornchild]

No, $499,-
For instance:

When it will become available the 4870 X2 will hit the market for $499.

http://www.tgdaily.com/content/view/37093/135/

[YukonTrooper]

Wow, guys! These cards are really something! I love the peak power consumption. No need to upgrade my PSU!

[Alex-Ro]

well,when there will be any 4870 for sale ?

[cadaveca]

No, $499,-
For instance:
http://www.tgdaily.com/content/view/37093/135/

says the article from APRIL with 480 SP's in the title.

Radeon 4800 is pretty much set for a May 2008 launch

2 months ago, minus a week? they don't even have the launch date correct..by a MONTH!!!

New for the 4800 series is AMD’s decision to split the clock of the GPU and shaders, following a move that Nvidia made with the GeForce 8800: Back then, the shaders were clocked at 1.35 GHz, while the rest of the chip ticked at 575 MHz.

[hurleybird]

I see the same cost as making 512-bit membus pcb...which gives me a $575 retail price for 4870x2's...I am hoping for less cost..$499...

Well, GDDR5 can use different trace lengths to memory chips, which I'm sure you know. If the shared frame buffer rumor is true I'm guessing that the trace lengths between a memory chip and the two RV770s will need to be equal, but even then board design should probably be less complex and have less layers if the memory placement is good.

[ORBR]

STOP THE PRESS

TAKE A LOOK AT THIS



http://www.tweaktown.com/articles/14...nce/index.html




so a lot of review are pure sh_t -.-

there's a bug here, I suppose

[hurleybird]

Well, that certainly helps explain things!

[Unbornchild]

says the article from APRIL with 480 SP's in the title.

So what; this very thread started 28 March 2008, in it you'll find several times 499 speculated, this in response to Xello.

[ghost101]

STOP THE PRESS

TAKE A LOOK AT THIS

http://www.tweaktown.com/articles/14...nce/index.html

So would a single hd4850 be bottlenecked on a p35 board?

[cadaveca]

Well, GDDR5 can use different trace lengths to memory chips, which I'm sure you know. If the shared frame buffer rumor is true I'm guessing that the trace lengths between a memory chip and the two RV770s will need to be equal, but even then board design should probably be less complex and have less layers if the memory placement is good.

Yep, you are think as I am...trace length from one gpu to mem to next gpu might have to stay the same, or one gpu will always get "priority". I do forsee a way around this, but it's not good for cooling, imho, so who knows. It's a techinical challenge that should be shouted from the mountaintops if they pull it off...as they should have patents for such that would prevent nVidia from ever pulling off the same thing...making ATI king of dual gpus.


But since R300 ATI gpu's have been capable of multi-rendering(renderbeast), so they have far more experience in this field, with products in the marketplace far longer than NV has...so I will put nothing past them.



But let me say this much...R600 was reviewed as having UVD...even some reviews posted results of UVD(which were completely faked)...but no card actually was capable. Today this review is still posted in this fashion...which makes me wonder about AMD's rumour control...everything out in the public domain right now is just rumour, IMHO, and ones I helped start aren't gonna get much attention from me.

So what; this very thread started 28 March 2008, in it you'll find several times 499 speculated, this in response to Xello.

No, I was just poniting out that thier info need to be taken with a grain of salt, as does all other info.


I mean really, perfect example is GTX260/280 release prices...they were supposed to be what? And are actually how much more? It's just speculation in that article, as where the specs they listed, the prices, and everything else. That article is far too old to have any real info other than stuff purposely leaked to find holes in NDAs.

They even hype split clock domains...

"Our sources"....in other words, the info was not from ATI/AMD.

[ORBR]

Well, that certainly helps explain things!

[Jowy Atreides]

http://images.tweaktown.com/imagebank/4854845_g_12.gif

http://images.tweaktown.com/imagebank/4854845_g_09.gif

http://images.tweaktown.com/imagebank/4854845_g_08.gif

Forbidden

[Unbornchild]

I mean really, perfect example is GTX260/280 release prices...they were supposed to be what? And are actually how much more? It's just speculation in that article, as where the specs they listed, the prices, and everything else. That article is far too old to have any real info other than stuff purposely leaked to find holes in NDAs.

They even hype split clock domains...

"Our sources"....in other words, the info was not from ATI/AMD.

Yes you're right. I agree.

[gojirasan]

An intensive game like Crysis sees the cards get it at all resolutions, and quite significantly at that. The X48 is really able to stand out when compared to the P45.

WTF?! Why word it like that? How about something like: "The crossfire mode on the P45 seems to have some kind of problem". Oh wait. Is it because it would have negative implications for AMD and that might not be great for his shares? If he wanted to hide a crossfire problem with the P35/P45 chipset he should have just left that part out.

[m0da]

WTF?! Why word it like that? How about something like: "The crossfire mode on the P45 seems to have some kind of problem". Oh wait. Is it because it would have negative implications for AMD and that might not be great for his shares? If he wanted to hide a crossfire problem with the P35/P45 chipset he should have just left that part out.

I'm not sure if you're joking.
The whole article was trying to prove that 4850CF benefits from 2 PCI-E x16 lanes (X48 board), because when tested with 2 PCI-E x8 lanes (P45 board), the performance was noticably lower.

[Chickenfeed]

Hmm P45 is PCIe v2.0 2x 8 which is equivalent in bandwidth to 2x 16 PCIe v1.1. Its interesting if these cards actually require the additional bandwidth of full PCIe v2.0 16x. If this is truly the case and not merely a driver deficiency, I'm glad I didn't wait for a P45 and stuck with x48. 4870/X2s in crossfire will really hate P45 if this is the case.

@ghost

A single 4850 will stuff run full speed in 1 slot so it won't be bottlenecked.

[hurleybird]

@cadaveca: I was just in the shower when I had a "Eureka!" moment.

Wait for it....

Wait for it....


Who said that both rv770s need to write to the framebuffer?

[initialised]

I'm not sure if you're joking.
The whole article was trying to prove that 4850CF benefits from 2 PCI-E x16 lanes (X48 board), because when tested with 2 PCI-E x8 lanes (P45 board), the performance was noticeably lower.

This has been the case since early P35 boards ASUS released the Blitz with a crosslinx chip which made about a 13% difference and then x38 came out and made it redundant. Sadly the P45 chipset has kept the 16x + 4x or worse moved to an 8x + 8x layout.