ATI Radeon HD 4000 Series discussion

[Vapor]

Opened.

Feel free to discuss when/where appropriate. No need to keep it super busy if there's nothing to say, it's a sticky and not going anywhere

When 4870 previews begin to emerge, we'll have a new thread for that. Old thread ditched since it's just too premature and the thread had basically 0 useful posts (including the first post).

[n91htmare]

holy batman temps. Just got mine today and I'm idling at 70!!!!

I think this card needs an aftermarket cooler BIG TIME. Other than that... for $150, I think It's an excellent performer.

[MAS]

holy batman temps. Just got mine today and I'm idling at 70!!!!

if it works without troubles who cares of temperature?

[Clint]

if it works without troubles who cares of temperature?

Please keep those stupid non-contributive comments to your self.

High temps will hamper your OC capability and shorten life length.

Seek, read and learn.

[darkskypoet]

It idles hot because the fan is turned right down and stays there, and power play seems to leave it run a bit high.. (500mhz). So as soon as we can turn the fan up, or power play updates from next catalyst turn freq down; the temp'll be a bit high. However, I haven't run in to any stability problems at all yet with mine.

Also, some have mentioned that pulling HS/Fan and changing HS paste to something better, drops the core temp. So you could always try that if you want.

Edit: My HIS HD3870 acted in a similar fashion; very low fan speed and higher temperature yet super quiet. My Sapphire 4850 is really, really quiet as well.

[mr2sw21]

so there is going to be 4 version HD4850 510mb, HD4850 1GB, HD4870 512MB, and HD4870 1GB. I am interested to the HD4870 1GB any one know when will that come out?

[mr2sw21]

by the way is any different in 512mb and 1gb in benching? For say any one try the HD3850 1gb is any different in benching than the HD3850 512mb?

[Stuen4y]

if it works without troubles who cares of temperature?

The other components in the case. The hot air doesn't go out of the case, so we CARE if the card is hot or not .

by the way is any different in 512mb and 1gb in benching? For say any one try the HD3850 1gb is any different in benching than the HD3850 512mb?

For me 1gb video memory is far much if you are not running on 24"+ display with resolution 1920x1200+. The price will be higher too and the 512mb model is more attractive(for me).

[darkskypoet]

The other components in the case. The hot air doesn't go out of the case, so we CARE if the card is hot or not .

True. But then again, most cards that exhaust into the case share this issue... As in all single slot cards? Also, the temp of the core doesn't quite equate to how much its heating the case. If a side issue is inefficient heat sink transfer material, then the core itself is reading a higher temp because of lack of proper heat transfer to heat sink, and then into case.

Example, if the core is reading say 90 degrees, and a portion of that is due to inefficient heat transfer... and a second card has a core temp of around 55 degrees... Its not a simple comparison as to which is heating the other components more. If the core on the second card is larger, and has very efficient transfer material; it could be dumping more heat into the case then the first card.

Long story short; the core temp doesn't indicate how much heat is being dumped into the case. That being said... The card idles @ too high a frequency IMO.. and as such, that could be fixed, and that would lower heat generation, which would lessen heat being dumped into case.

[MAS]

Please keep those stupid non-contributive comments to your self.

High temps will hamper your OC capability

are you going to OC using stock cooler?

and shorten life length.

the card will live 10 years instead of 15. cool

[Chickenfeed]

Example, if the core is reading say 90 degrees, and a portion of that is due to inefficient heat transfer... and a second card has a core temp of around 55 degrees... Its not a simple comparison as to which is heating the other components more. If the core on the second card is larger, and has very efficient transfer material; it could be dumping more heat into the case then the first card.

Long story short; the core temp doesn't indicate how much heat is being dumped into the case. That being said... The card idles @ too high a frequency IMO.. and as such, that could be fixed, and that would lower heat generation, which would lessen heat being dumped into case.

For sure. Its more down to the TDP of the gpu/pcb than anything. I'm still for dual slot coolers though as they do make a marked difference in both noise and case ambient temp.

Aren't the 4xxx supposed to have a 2D clock as well for idling like the GT200s?


As far as component life time due to overclocking, when done properly the hardware will be WAY obsolete before it ever fails on you so thats a moot point. If the cards don't exceed their rated temperature threshold they'll last a good long time and considering most auto fan profiles and down clocking mechanisms prevent this, it shouldn't ever be an issue.

[YukonTrooper]

Aren't the 4xxx supposed to have a 2D clock as well for idling like the GT200s?

Have to wait until PowerPlay is 4 series compliant. At least that's what I think I read in one of these threads.

[SNiiPE_DoGG]

Have to wait until PowerPlay is 4 series compliant. At least that's what I think I read in one of these threads.

well considering that both my 2900 xt and my 3870 lower clocks when not in 3d app I'm gonna go out on a limb and say ya, its gonna have a 2d clock :P

[Stuen4y]

Aren't the 4xxx supposed to have a 2D clock as well for idling like the GT200s?

To be honest I don't care. We just need new version of AtiTool and everyone can adjust the fan and the clocks on their taste.

[Kai Robinson]

Can someone help clear this up? I'm having a tough time trying to figure out how cards communicate over crossfire...and P45 issues - i was under the impression that the bridges that the cards connect to each other with, share data between them, but all the fetching of the textures and the like, are done over the PCIe bus, correct?

Now, the other thing is with P45. A single PCIe 1.0 16x slot provides 4GB/sec of bandwidth, and a single PCIe 2.0 16x slot provides 8GB/sec of bandwidth....P45's having PCIe 2.0 8x/8x setup for crossfire is the same as having a pair of PCIe 1.0 16x slots, as each PCIe 2.0 8x provides 4GB/sec of bandwidth, correct?

So why is it then, that in the Tweaktown review that was done that pitted the P45 vs the X48 have CF on the P45 tank so badly when it got to Crysis, even at low resolutions? The mild 10% performance hit i can fathom, and as well in the 2560xstupid resolutions they were using, but from 35fps at 1280x1024 to 21fps? I'm sorry but i dont buy that. A single HD4850 hands my 8800GT its own arse at 'Very High', and my 8800GT gets 29.03fps at 1440x900... so here i am wondering, is that review flawed, or have i dropped into a coma and all of this is some elaborate way of my mind telling me that nothing is real and i need to wake up....?

If its the former, can someone on the forum please do an independant CF test of P45 vs X38 (DDR2 to DDR2, not DDR2 to DDR3) and test Crysis? And if its the latter, can someone please ring Gene Hunt?

(cue David Bowies "Life on Mars")

[Chickenfeed]

A single PCIe 1.0 16x slot provides 4GB/sec of bandwidth, and a single PCIe 2.0 16x slot provides 8GB/sec of bandwidth....P45's having PCIe 2.0 8x/8x setup for crossfire is the same as having a pair of PCIe 1.0 16x slots, as each PCIe 2.0 8x provides 4GB/sec of bandwidth, correct?

That is correct, along with increased power delivery at the slot with the v2.0 spec.

As for the abnormal P45 performance I don't know what to make of that. I hope its merely a driver problem as I doubt that these cards would be over saturating the bus ( haven't done the math ) but it might be possible. If thats the case this sucks as you'd think the 4850s would be a very popular entry level crossfire option on a P45 board. I'd wait for some more numbers like you said before we panic though. I'm sure someone on the boards here has 2 4850s on a P45 already.

[SimBy]

Can someone help clear this up? I'm having a tough time trying to figure out how cards communicate over crossfire...and P45 issues - i was under the impression that the bridges that the cards connect to each other with, share data between them, but all the fetching of the textures and the like, are done over the PCIe bus, correct?

Now, the other thing is with P45. A single PCIe 1.0 16x slot provides 4GB/sec of bandwidth, and a single PCIe 2.0 16x slot provides 8GB/sec of bandwidth....P45's having PCIe 2.0 8x/8x setup for crossfire is the same as having a pair of PCIe 1.0 16x slots, as each PCIe 2.0 8x provides 4GB/sec of bandwidth, correct?

So why is it then, that in the Tweaktown review that was done that pitted the P45 vs the X48 have CF on the P45 tank so badly when it got to Crysis, even at low resolutions? The mild 10% performance hit i can fathom, and as well in the 2560xstupid resolutions they were using, but from 35fps at 1280x1024 to 21fps? I'm sorry but i dont buy that. A single HD4850 hands my 8800GT its own arse at 'Very High', and my 8800GT gets 29.03fps at 1440x900... so here i am wondering, is that review flawed, or have i dropped into a coma and all of this is some elaborate way of my mind telling me that nothing is real and i need to wake up....?

If its the former, can someone on the forum please do an independant CF test of P45 vs X38 (DDR2 to DDR2, not DDR2 to DDR3) and test Crysis? And if its the former, can someone please ring Gene Hunt?

(cue David Bowies "Life on Mars")

It could be that the Tweaktown review is simply loads of crap? Maybe some driver problems, who knows.

[bill_d]

i think P45 & [x38, x48] are all pcie 2.0, so P45 is at PCIe 2.0 8x/8x and x38,x48 is at PCIe 2.0 16x/16x
don't know where pcie 1.1 comes in to it

[Kai Robinson]

Bill - it was supposed to illustrate that PCIe 2.0 boards are a relatively new occurrence, and beforehand, when dealing with boards that have two PCIe 1.0 16x slots, Crossfire is fine. Now we have PCIe 2.0 8x (which is the same bandwidth), that suddenly, is no longer enough, and we're saturating the bus (supposedly).

I'm hoping its driver issues, at any rate.

[Leeghoofd]

I too want a 2nd opinion on the Tweaktown review... a lot of people will ditch the P45 easy setup and clocking for the more expensive X38/48 variant mobo's if this test is completely accurate... it's possible a simple bios update will cure this, also don't get why they didn't test a second P45 of another brand to be sure that their findings are spot on....

I find it pretty logic that there has to be a difference but this is a bit too much for all the Crossfire fans out there...

[adamsleath]

yeah, well, i was also thinking about the whole: is 8x8xpci-e2.0 enuf for 4850cf question...vs pci-e2.0 16x16x...but i didnt quite get to posting the question. some more reviews comparing 8x8x to 16x16x pci-e 2.0 would be handy to compare, confirm, disconfirm...
but if 8x8x is supposedly = to 16x16x pci-e1.0 i wouldnt have thought your "bus" would be clogged. some proof will pop up sometime this millennium

..you know with all those free samples i get i'll just pop off and do a test, and rewrite some drivers while im waiting

[Ral3l3it]

I just picked up 2 4850's for some crossfire for $300 usd out the door!!!
This should run all my games at 1920x1200 4xaa no issues.

Where did you get them for that cheep?

[alexio]

The bandwidth of 8x PCIe 2.0 is plenty for these cards unless Crossfire is now done very differently. The problem could be latencties, which I doubt make such an impact, or more likely some kind of bug.

[Kai Robinson]

I have a lot of hope for the P45 as i really do need the ICH10R, which neither the X38 or the X48 boards have

That and the Asus 16-phase PWM = winola and cake.

EDIT: http://www.tomshardware.com/reviews/...-0,1915-3.html

Take a look at that - you'll notice that 8x PCIe 2.0 is more than enough for even the 9800, the only app to really use the extra bandwidth being MS Flight Sim....but then, that game wouldn't be happy unless you threw BlueGene/L at it.

[adamsleath]

Once larger textures need to be accessed, as is the case in Crysis http://en.wikipedia.org/wiki/Crysis or Microsoft’s Flight Simulator http://en.wikipedia.org/wiki/Flight_simulator X, interface bandwidth becomes a crucial element. Any link width below x16 will noticeably limit these games’ playability. []

The answer, thus, has to be "yes": you want maximum bandwidth, and PCI Express 2.0, for all sorts of sophisticated 3D applications

although 8x8x surely enuf for older games.
er um P45 a single card mobo or low end CF then...?