ATI Radeon HD 4000 Series discussion

[Eastcoasthandle]

I'm assuming you have or will update your own results?

[Spectrobozo]

so, 32 or 40 tmus?

[ryboto]

so, 32 or 40 tmus?

They're saying 40.

[Xoulz]

Yeah preliminary crossfire results of the 4850 are showing that some games scale at 90% even when CF is working. And supposedly the RV770 architecture has many improvements in how the chip communication is designed. Looked almost like a dual core or quad core CPU in layout...

It seems that perhaps AMD's influence on chip design is paying off. I'm curious to see the actual architecture the new R7xx..!

Anand Shimpi has hinted that ATI's core is quite impressive/interesting, even though his hands are tied by the NDA, it seemed as if he really wants to discuss what ATI/AMD have done.

So depending on how far ATI/AMD is thinking ahead about scalability, I really do see the 4870 is a premier Card. At least for the majority of gamers that I know of. Most if not all gamers are using 21" & 24" monitors, resolutions beyond 1920 x 1200 really are meaningless to most. So I think ATI is going in the right direction.

A 1gig 4870 for $350 sounds about right for 80% of the people I know!

[cadaveca]

so, 32 or 40 tmus?

10 effective pipelines, 4 per pipe.

atis card also only has 160cores, 800 is only marketing.

it's actually almost more than that, thanks to added functionality of the "slave" shaders. There's absolutely no direct comparison now. We need a standardized gpgpu app to test cards to determine which is faster, really..even games aren't accurate testing as scheduling of pixels in flight can change from driver to driver, completely skewing any results...think like G80...with the "missing MUL"....it's only missing in certain situations(most situations, but that's besides the point).

[AliG]

hmm, I guess that we'll never truly see the full potential of any r600 derivitive (including the r600 itself) for a long time, but once that happens, NVIDIA's going to need a life time supply of butt-plugs because they're going to be raped

[Helmore]

10 effective pipelines, 4 per pipe.



it's actually almost more than that, thanks to added functionality of the "slave" shaders. There's absolutely no direct comparison now. We need a standardized gpgpu app to test cards to determine which is faster, really..even games aren't accurate testing as scheduling of pixels in flight can change from driver to driver, completely skewing any results...think like G80...with the "missing MUL"....it's only missing in certain situations(most situations, but that's besides the point).

Yup 10 quads of TUs (40 TUs), these are not the full TMUs that the R(V)6xx series had though.
To Hornet331, there are only 10 cores actually, just like on the GT200 which also only has 10 cores. Compared to the 8 cores on G80 and 4 on the RV670/R600.
About that MUL, the MUL on G80 was basically nowhere to be found, there are some guys that tried to find it (on beyond3d for example) but without any succes. This is different on the GT200, there the SPs can do this 'dual issuing' (calculating 1 MADD + 1 MUL (3 FLOPs per SP in other words)). This 3 FLOPs per SP per clock is what gives the GTX280 it's 933 GigaFLOPs, it would have 622 GigaFLOPs if it could not do this. The 8800GTX was advertised as being able to do 518 GigaFLOPs, but that is with being able to do this MUL and in reality it could do 346 GigaFLOPs. But the GTX2x0 is not always utilizing this extra MUL, I'll quote TechReport on this one:

Trouble is, that additional MUL wasn't always accessible on the G80, leading some folks to muse about the mysterious case of the missing MUL. Nvidia won't quite admit that dual-issue on the G80 was broken, but it says scheduling on the GT200 has been massaged so that it "can now perform near full-speed dual-issue" of a MAD+MUL pair. Tamasi claims the performance impact of dual-issue is measurable, with 3DMark Vantage's Perlin noise test gaining 16% and the GPU cloth test gaining about 7% when dual-issue is active. That's a long way from 33%, but it's better than nothing, I suppose.

http://techreport.com/articles.x/14934/5
Oh and it should give the GT200 +50% extra computing power, they are a little off with their +33%. Those Vantage tests are pretty much completely shader dependent BTW.

[Helmore]

hmm, I guess that we'll never truly see the full potential of any r600 derivitive (including the r600 itself) for a long time, but once that happens, NVIDIA's going to need a life time supply of butt-plugs because they're going to be raped

I think you are a exaggerating a bit, but ATI/AMD sure is in a lot better shape than it was about 1 year ago.

[cadaveca]

Anyone tried these yet? I would like to know what kind of performance gains is will give, if any....

I'm seeing 5-8% in vantage, most notibly the flag test and Jane Nash. '06 performance is basically the same.


And yeah, I'm aware of "the MUL".

[Helmore]

I'm seeing 5-8% in vantage, most notibly the flag test and Jane Nash. '06 performance is basically the same.


And yeah, I'm aware of "the MUL".

The NVIDIA way of wishing you luck goes something like this:

"May the MUL be with you, Luke".

[ChaosMinionX]

I hope R700 is a beast....

[cadaveca]

LEt's put it this way....3870 Quadfire gets about E62xx in Vantage @ 877/954(CCC max), Phys-X and 4ghz QX9650.

http://service.futuremark.com/compare?3dmv=179377

4870x2 is supposed to get E7xxx?

[Helmore]

It seems that perhaps AMD's influence on chip design is paying off. I'm curious to see the actual architecture the new R7xx..!

Anand Shimpi has hinted that ATI's core is quite impressive/interesting, even though his hands are tied by the NDA, it seemed as if he really wants to discuss what ATI/AMD have done.

Well that probably because RV770 is a bigger change in microachitecture from RV670 than GT200 was from G92. Quite a lot has changed actually, no ringbus but more of a hub, TMUs are different, Shader units have also gotten some refinement, RBEs are larger and there is extra functionality build in to make sure that the R700 can be made. That list one is quite a departure from the R680, here they are claiming (dutch source, you might have to grab yourself a translator) that the RV770 chip has it's own PCIe bridge build in and therefore won't be needing a PLX chip, or get a smaller (proprietary) one. The R700 board is smaller because of this, quite a surprise actually as R700 will probably draw more power than R680 and it will still be a shorter board.
There probably are a couple more things to mention but these are probably the most significant changes.

[AnXioZ]

RV770 Debut - ATi Radeon HD4850 @ VR-Zone.

[Helmore]

That could explain the early arrival of that hot-fix, there barely are any results in that article .
BTW AnXioZ, your avatar is disgusting .

[AAbenson]

to the guys who already got one 4850 - is the mount mechanism(mount holes) the same as on the 3870?...
...so that coolers which can be used with the 3xxx series can be used also here with the 4xxx series

[AnXioZ]

The article is yet to be posted on their website that's why
BTW Thanx

[gojirasan]

4870 is out on the 25th of June while the 4870x2 is unknown but is looking like July/August

Actually there are no reports of a July release for the R700 that I am aware of. For now, we are talking August 15th as a best case scenario. It could easily be delayed further. If there are any unforeseen problems it could be October before we see it.

Let's not forget how ambitious the R700 is if it's true that it seeks to connect the two GPUs in some way that is not Crossfire. Sort of like what the CPU folks have been doing for a while now, but for GPUs. If AMD can make the two GPUs look like one functional unit to the OS and drivers it will be a major milestone for PC graphics cards. A huge step forward. And it will vindicate AMD's new design philosophy for GPUs. The 6 month refresh could be a 4870x4, and then a summer refresh could be a 4870x8, and then a 4870x16 etc. Although, as some have mentioned wrt a GTX295 GX2, cooling could be an issue. Remember that GPUs don't have the scaling problem that CPUs have. Graphics processing is an "embarrassingly parallel" problem domain. SLI and CF (usually AFR) is basically a crude 3DFX era hack. That's why it doesn't scale much past 2. If doubling the amount of stream processors is so effective for this generation there is no reason why doubling the number of GPUs couldn't do the same. At least in theory.

And to those of you who are reminding us of how a 4870x2 should utterly clobber a GTX280 in every game, remember that that is only true if AMD succeeds with this idea of eliminating "crossfire" from their dual GPU cards. Otherwise it would only be true for those games that fully support crossfire. Currently there are quite a few that do not. And for older games that isn't likely to change no matter how popular the R700 becomes. And that is not even considering the microstuttering or lag issues that continue to plague SLI/CF. Until someone is willing to break NDA and actually tell us, it will remain a mystery.

[STaRGaZeR]

RV770 Debut - ATi Radeon HD4850 @ VR-Zone.

With the ATi HD4850 displaying so many quirks in our tests, many of you must be expecting the product to be slammed down the depths of Hell and banished to the Zone's bin. Many of you may too wonder, why VR-Zone's testing seem to be choppy seas to the Little Red Raft when a great other review sites have their HD4850s successfully complete the plethora of tests.

The reason is simple: we test the HD4850 in a variety of situations that you, as an hardware enthusiast, is likely to put the hardware through. Best part of the story? ATi will continue to support the HD4850 with new drivers along the way to iron out bugs and stabilize the product.

Haha what is this noob saying.

[G.Foyle]

to the guys who already got one 4850 - is the mount mechanism(mount holes) the same as on the 3870?...
...so that coolers which can be used with the 3xxx series can be used also here with the 4xxx series

Yes. The four main holes are exactly the same.

[CBONE]

Haha what is this noob saying.

Oh, SNAP!!!! You mean they will support the card they have just released with new drivers in the future? Wait! These "drivers" will "iron out" bugs(what are those? I don't get the jargon.) and stab-o-rize the product? AMD is too good to us!

This whole supporting your product with beneficial "drivers" could flip the tech industry on its ear! Bravo, Vr-zone, for getting the scoop on this whole product support business! Bravo!

[gojirasan]

Whats the best brand that has a better customer service, warranty and rma of the ati cards
TIA

I am also curious about this. The last time I was looking at ATI cards (an X850 maybe?) Sapphire was regarded as an excellent manufacturer IIRC, with lots of fancy cooling solutions etc. Now I am told not to buy anything from them, ever. So who is the best then? Is there an Evga of the ATI world? It looks like Asus and MSI make cards. My immediate inclination would be to buy my 4870x2 from Asus or MSI, since they are familiar to me as motherboard manufacturers.

[Dark-Energy]

Asus is pretty good. HIS has good coolers.

[AliG]

Speaking of HIS, I wonder if they'll reveal the Ice Q 4 cooler, the 3 was good but I heard it wasn't enough for the 850 core 3870

[perkam]

We no longer need this thread now that the new reviews are out. Plus the 4850 Preview thread needs to be renamed to the Official 4850 Review Thread.

Perkam