Radeon HD 4890 3DMark Performance Revealed

[ShogoXT]

I hate that my work blocks pictures of yalls results. Looks to be a nice little improvement, but ATI needs to get on the Physx deal.

Not really, its only serving to make Nvidia look better in this synthetic benchmark. Is that worth paying Nvidia money for?

EDIT: Sorry if that sounded fanboish, just annoyed how that works in Vantage. It screws the score....

 

[TheMailMan78]

Not really, its only serving to make Nvidia look better in this synthetic benchmark. Is that worth paying Nvidia money for?

Well this may be news but some games use Physx. To bad Vantage doesnt use Havok instead.

 

[farlex85]

The results are comparable IMO. Many here believe that Vista is just as good as XP. Furthermore, the 3D06 is CPU bound. The results with a quad core should have been higher then a dual core using XP.

It's been shown time and time again the in 3d06 xp yeilds better scores than in vista. Note that the i7920 did score 1500 points more in cpu score at 1ghz lower clock than the e8400. Sounds about right. This is a preliminary score w/o proper drivers and completely at stock (who here benches their stuff at stock?). It's also a 4890, not a 5870. Once it is released it will be a bit faster than the 4870, but not by a landslide. These seem about right considering.

 

[EastCoasthandle]

You are on XP, results are not comparable (XP scores higher).

WarEagleAU: EastCoasthandle got 16155 3DMarks SM2:6907 SM3:8525 CPU:3364 on XP

The results are comparable IMO. Many here believe that Vista is just as good as XP. Furthermore, the 3D06 is CPU bound. Having a quad core can increase the overall results of the benchmark when compared to a dual core. Therefore, IMO, the 920 quad core should have been higher then a overclocked dual core using XP. The results of the 920 clearly show a higher benchmark result the E8400 at 3.60GHz.

It's been shown time and time again the in 3d06 xp yeilds better scores. Note that the i7920 did score 1000 points more in cpu score at 1ghz lower clock. Sounds about right. This is a preliminary score w/o proper drivers and completely at stock (who here benches their stuff at stock?). It's a 4890, not a 5870. Once it is released it will be a bit faster than the 4870, but not by a landslide. These seem about right considering.

I would rather wait for other benchmark results. Do to the higher CPU scores there doesn't appear to be any correlation with overall results. If you look at both SM2.0 and SM3.0 they are much lower and I've seen higher results then that when using a Q2C.

 

[erocker]

All I meant by my card, was that I was comparing a heavily OC'd GTX 260, to effectively a heavily OC'd 4870, so thought they would be closer matched than these results.

That's just speculation. It's not a heavily OC'd 4870 it's a stock 4890. We have no idea what kind of overclocks these cards get yet. I love my GTX 260 too, but since I love new cards, it's getting a playmate!

 

[alexp999]

That's just speculation. It's not a heavily OC'd 4870 it's a stock 4890. We have no idea what kind of overclocks these cards get yet. I love my GTX 260 too, but since I love new cards, it's getting a playmate!

Even if I hadnt made that comment, you must agree that these clocks seem unexplainably low. Iam basing my comments of what I know now, which afaik, is that the RV770 brings nothing more than higher clocks.

Take EastCoastHandle's results for example. His CPU score is lower, yet the SM scores are higher with the 4870 clocked at the same speed.

 

[farlex85]

I would rather wait for other benchmark results. Do to the higher CPU scores there doesn't appear to be any correlation with overall results. If you look at both SM2.0 and SM3.0 they are much lower and I've seen higher results then that when using a Q2C.

Even if I hadnt made that comment, you must agree that these clocks seem unexplainably low. Iam basing my comments of what I know now, which afaik, is that the RV770 brings nothing more than higher clocks.

Take EastCoastHandle's results for example. His CPU score is lower, yet the SM scores are higher with the 4870 clocked at the same speed.

Drivers, this is an unreleased card

 

[cdawall]

i could see it running with a GTX260 considering with both cards oc'd my G92 8800GTS keeps up with a 4850 oc'd and the 4870 runs with a 9800GTX

 

[alexp999]

Drivers

But if it is based off an RV770 then there should be no excuse on the drivers front.

 

[farlex85]

But if it is based off an RV770 then there should be no excuse on the drivers front.

I was under the impression it was an RV790. Even if it's similar, it's going to be a tad different. For all we know, the only difference may be on the software level. Either way, it's a preliminary benchmark of an unreleased card (thus not to be expected as full results).

 

[johnnyfiive]

That seems a little low. My GTX 260 beats that.

It should, physx.

 

[muon]

That seems a little low. My GTX 260 beats that.

Because your cpu (physx) score is through the roof.

-----------------------------------------------------------------------------------------------

For comparison:

http://service.futuremark.com/compare?3dmv=895112

HD4870 using cat 9.2, at 850mhz and q9300 @ 3.3ghz. I get a gpu score of 10,034 and overall score of P10613

The HD4890 is an overclocked hd4870, which is basically what the name of the card suggested.

 

[btarunr]

But if it is based off an RV770 then there should be no excuse on the drivers front.

It's not based off an RV770.

 

[muon]

It's not based off an RV770.

What makes you so sure? I think it is if these results are true.

 

[btarunr]

What makes you so sure? I think it is if these results are true.

Results aside, we were talking about drivers. It's not that existing drivers (the ones used in this test) seamlessly support RV790 just because it has a lot in common with the RV770.

 

[alexp999]

It's not based off an RV770.

You're article said it was!

Because your cpu (physx) score is through the roof.

-----------------------------------------------------------------------------------------------

For comparison:

http://service.futuremark.com/compare?3dmv=895112

HD4870 using cat 9.2, at 850mhz and q9300 @ 3.3ghz. I get a gpu score of 10,034 and overall score of P10613

The HD4890 is an overclocked hd4870, which is basically what the name of the card suggested.

I was only referring to the GPU score as I know physx affects the CPu score.

 

[muon]

I was only referring to the GPU score as I know physx affects the CPu score.

Fair enough. What do you get with the 260 gtx? Is it overclocked?

 

[alexp999]

Fair enough. What do you get with the 260 gtx? Is it overclocked?

I get about 10,800 GPU score and yes it is overclocked. But this card seems way too far behind. I know the 4870 cant quite match a GTX 260, but I only have a 192SP version, and I would have thought the RV790 would be capable of more than that.

Now I'm not throughing doubt on this, I'm just speaking my mind, when I say this.
How do we know this is even genuine? You could quite easily edit the fields in the screenies. Only reason I say this, is cus they have blanked out the BIOS version. WHY???

 

[niko084]

http://pchardwareblips.dailyradar.com/story/rv790_to_become_radeon_hd_4890/

Looks like the news is, it is indeed a different core.

 

[btarunr]

You're article said it was!

"It is effectively only an overclocked RV770" true, since its specs are identical to the point where clock speeds get in. But it wasn't said they had the same IDs, or the same ASICs.

 

[alexp999]

"It is effectively only an overclocked RV770" true, since its specs are identical to the point where clock speeds get in. But it wasn't said they had the same IDs, or the same ASICs.

If it has the same fab process, same number of stream processors, TMU's and ROP's, same memory interface. That in my book means based off. I didn't say it was an RV770, but the architecture must be so close to being the same (if not the same), then I cant see a reason for the drivers not to be fully optimized.

 

[farlex85]

I get about 10,800 GPU score and yes it is overclocked. But this card seems way too far behind. I know the 4870 cant quite match a GTX 260, but I only have a 192SP version, and I would have thought the RV790 would be capable of more than that.

Now I'm not throughing doubt on this, I'm just speaking my mind, when I say this.
How do we know this is even genuine? You could quite easily edit the fields in the screenies. Only reason I say this, is cus they have blanked out the BIOS version. WHY???

It could be fake. Given the circumstances it seems far more likely that it's just not running on all software cylinders yet. Either way, it's the first numbers, never trust them fully.

If it has the same fab process, same number of stream processors, TMU's and ROP's, same memory interface. That in my book means based off. I didn't say it was an RV770, but the architecture must be so close to being the same (if not the same), then I cant see a reason for the drivers not to be fully optimized.

Again the main difference b/t the cards may lie at the software level. Perhaps you are expecting too much.

 

[niko084]

I get about 10,800 GPU score and yes it is overclocked. But this card seems way too far behind. I know the 4870 cant quite match a GTX 260, but I only have a 192SP version, and I would have thought the RV790 would be capable of more than that.

Now I'm not throughing doubt on this, I'm just speaking my mind, when I say this.
How do we know this is even genuine? You could quite easily edit the fields in the screenies. Only reason I say this, is cus they have blanked out the BIOS version. WHY???

We don't realistically.

But also the 4870 can come pretty close...
http://service.futuremark.com/resultComparison.action?compareResultId=522177&compareResultType=19

We have to take into account that this card is at "Stock Clocks" if it's even real, and is on pre-release drivers.

Do NOT take any of the numbers posted up as accurate, expect anywhere up to 20% or so off, if you look at a lot of cards numbers that seemingly get released before the release, they range from 5-20% off, in either direction.

 

[EastCoasthandle]

It could be fake. Given the circumstances it seems far more likely that it's just not running on all software cylinders yet. Either way, it's the first numbers, never trust them fully.



Again the main difference b/t the cards may lie at the software level. This is certainly the most likely cause for low numbers.

If anything was tweaked I am willing to believe it was to optimize the shaders making them more efficient. Other then that we really have to wait and see once the card is released in order to know if anything was changed or not. That would IMO determine if drivers for the 4870 wouldn't be as effective for the 4890, for example.


Furthermore, I honestly don't believe those results are correct. Here is a i7 920 and 4870 (750/900) stock using Vista:
3D Marks 16057
SM2.0 = 6013
SM3.0 = 7481
CPU = 5050
CPU clocked at 2.673GHz
source

Yet they are claiming 16096 using 850/975 also using a i7 920. Come on, something is obviously not adding up with that result.
3D Mark 16096
sm 2.0 = 6155
sm 3.0 = 7521
cpu = 4836

 

[btarunr]

If it has the same fab process, same number of stream processors, TMU's and ROP's, same memory interface. That in my book means based off.

Your book excludes the ASIC design itself as a factor.