ATI HD4800 Review Thread

[AliG]

AliG - 160GB/sec? Sorry but you've been on the funny sauce again. It's a 48 lane PCIe 2.0 Bridge chip from either PLX or IDT (more than likely PLX). Thats 16 lanes for each GPU, and 16 lanes for the bus connection, which means at most, 32GB/sec bandwidth between GPU's.

that's what the ATI representative said, don't shoot the messenger. I'll dig up the link. Only thing I don't get is why does ati need a fancy bridge chip when the 9800gx2 clearly did fine without one, all it does is add power consumption.

And you said I'm on the funny sauce again, when was the first time?

[halo112358]

hmm maybe needs further investigation.

Did you oc your card already?

Also can you see any physical damage?

No physical damage, nothing different than when I plugged the card in in the first place. Of course I OC'd it, duh. No voltage bumps, nothing out of the ordinary except routine catalyst overdrive changes.

[AriciU]

http://www.hardforum.com/showthread.php?t=1319753

2x4870 real life comparison against GTX 280 from an owner. Nicely written IMO.

[Calmatory]

that's what the ATI representative said, don't shoot the messenger. I'll dig up the link. Only thing I don't get is why does ati need a fancy bridge chip when the 9800gx2 clearly did fine without one, all it does is add power consumption.

And you said I'm on the funny sauce again, when was the first time?

X2 cards aren't regular corssfire, GX2 aren't crossfire, crossfire works in a different way SLI does. Does this help out?

[AriciU]

No physical damage, nothing different than when I plugged the card in in the first place. Of course I OC'd it, duh. No voltage bumps, nothing out of the ordinary except routine catalyst overdrive changes.

By OC you mean thru BIOS update with modded frequencies or thru ati catalyst?

[MomijiTMO]

What I can't believe is the power consumption of the 4870, especially in CF. It's just nuts.

[AliG]

X2 cards aren't regular corssfire, GX2 aren't crossfire, crossfire works in a different way SLI does. Does this help out?

I'm serious, I believe the only reason why they have the bridge chip is to increase the bandwidth between the two gpus for a faster interconnection as they've claimed no micro-stuttering with both the 3870x2 and 4870x2. Of course later on they admitted the micro-stuttering on the 3870x2 was horrible because the bridge chip didn't provide enough bandwidth (8 GB/s) and the extra latency made it worse than regular xfire.

[Lightman]

http://www.hardforum.com/showthread.php?t=1319753

2x4870 real life comparison against GTX 280 from an owner. Nicely written IMO.

That was interesting info for me:

The GTX 200 series power saving features do NOT work on an Intel platform. That’s right. I thought I was the odd one out when I saw countless other websites posting about the down-clocks and down-voltages, whereby the GPU settles at 300Mhz Memory to 100Mhz etc when no 3D app is running. What some of them failed to state was that they only achieved this on an Nvidia platform; while ALL the reviews out there failed to warn consumers that this does NOT work on an Intel platform at all. The writeups out there further confuse you by stating the downclocks are achieved in drivers, while only the full hybrid power features, i.e., shutting down the GPU and offloads 2D to the onboard IGP require an Nvidia platform. They just don’t know what they are talking about.

This was confirmed after a 30 minute phone conversation with an XFX engineer, the manufacturer of my GTX 280.

What this translates to is that all the idle power consumption graphs you saw on reviews over the GTX 200 series do NOT apply to an Intel platform. Again, out of the 15 odd reviews out there, not one, I repeat, NOT A SINGLE ONE, bothered to state this clearly. Whats the user footprint in the real world with an Nvidia platform? 0.1% of the PC population?

The GPU temp does drop back to the 50s while idling in Windows, but the clocks do not scale back, which leads me to believe there is no real power savings.

Anyone to confirm this from personal experience?

[SKYMTL]

That was interesting info for me:



Anyone to confirm this from personal experience?

Well, I know my GTX 280 downclocks fine on an X38 and P35 chipset. The power consumption figures I have seen follow the trend of it working when at idle.

[Flinch]

If you mean clocking down when not in 3D,that works just fine for me on an Intel chipset.

[m0da]

Edit:
There's been more than 1 response saying the downclock works. I stand corrected and it must have been due solely to my copy of the XFX card. The engineer I spoke to may not be in a very experienced position with the GTX 280 to identify whether its the card or the series at fault.

Guess it works.
That guy sounds like an AMD/ATI infomercial though! It almost sounds too good to be true, but I'm rooting for the underdog here... hope it is true

[Hornet331]

Guess it works.
That guy sounds like an AMD/ATI infomercial though! It almost sounds too good to be true, but I'm rooting for the underdog here... hope it is true

what an opportunity... he can discredit intel and nv in one blow, and forgot to mention that atis own powersaving is not working right now.

[SKYMTL]

what an opportunity... he can discredit intel and nv in one blow, and forgot to mention that atis own powersaving is not working right now.

Actually, it is in some cases and will soon be working with ALL cards. Only some of the first batch of cards had it disabled which should be fixed in a future driver release. Other cards had it working (most notably the MSI cards) without a problem.

[GoldiClocks]

http://www.hardforum.com/showthread.php?t=1319753

2x4870 real life comparison against GTX 280 from an owner. Nicely written IMO.

"Sorry for the wall of text, I hope you found the above a little more useful than a bunch of graphs from some teenage “me-too” review sites."

What bad have hardware sites ever done to this guy??

[Tonucci]

I stopped reading when he was attacking reviewers credibility and then proceeded to use his "trusty right foot" for temperature testing.

That, along with many other flaws/contradictions.

[STaRGaZeR]

I stopped reading when he was attacking reviewers credibility and then proceeded to use his "trusty right foot" for temperature testing.

That, along with many other flaws/contradictions.

What flaws/contradictions did you observe?

[Tonucci]

What flaws/contradictions did you observe?

Sorry, Im not going through that again. I would have to re-read, taking notes

[disruptfam]

would 850w psu be enough for 4870 1b cf?

[Truckchase!]

would 850w psu be enough for 4870 1b cf?

Name brand, yes..... assuming the load of the rest of your system is "average".

[Chickenfeed]

Well his personal review confirms for me that CF 4870s best the GTX 280 in AoC and that was my only remaining concern. Good to know as AoC is one of the more demanding games out currently and if the 4870s don't choke with high levels of AA in it, thats awesome.

[gojirasan]

Of course later on they admitted the micro-stuttering on the 3870x2 was horrible because the bridge chip didn't provide enough bandwidth (8 GB/s) and the extra latency made it worse than regular xfire.

Who is "they"? AMD? Did they really admit that? Since when do these large corporations ever admit to a flaw in their product? I had heard that both Nvidia and AMD are doing very well at the moment by not admitting to the existence of microstuttering or even the lack of multi-GPU driver support.

[zerazax]

They did indeed admit that they should have used the PCIE 2.0 bridge chip and not the PCIE 1.1 one, but they didn't say. But they did indeed admit using the wrong chip.

[Sr7]

They did indeed admit that they should have used the PCIE 2.0 bridge chip and not the PCIE 1.1 one, but they didn't say. But they did indeed admit using the wrong chip.

Uh, you'll see that has absolutely nothing to do with microstuttering though.

[zerazax]

Note they never said anything about microstuttering, they just admitted they made a mistake with the type of bridge chip used. As to for what reasons, I don't know. I'm just re-affirming the fact they did indeed state it.

[flopper]

Note they never said anything about microstuttering, they just admitted they made a mistake with the type of bridge chip used. As to for what reasons, I don't know. I'm just re-affirming the fact they did indeed state it.

More like with the performance output.

and the cards can always be faster and better designed