Wednesday, March 18th 2009

Radeon HD 4890 3DMark Performance Revealed

Taiwan-based English tech website OC Heaven has disclosed some performance figures of the upcoming ATI Radeon HD 4890 1 GB graphics accelerator. The tests run are two of the most popular synthetic benchmarks: 3DMark Vantage and 3DMark06. Also disclosed, rather verified, are the card's clock speeds as read by ATI Catalyst Control Center and GPU-Z. The test bed, from what the 3DMark06 window shows, consists of an Intel Core i7 920 CPU running at 2.66 GHz, coupled with 3 GB of system memory. In 3DMark06, the HD 4890 accelerator secured a score of 16,096 points, with SM 2.0 score of 6155, HDR/SM 3.0 score of 7521, and CPU score of 4836. In 3DMark Vantage, it secured a score of P10996. Catalyst Control Center reveals the card's memory bandwidth to be 124.8 GB/s, up from the 111 GB/s on its predecessor, the HD 4870. The early driver in use makes provides "RV790" as the device string to GPU-Z.

Update (03/19):Ukrainian website Overclockers.com.ua has come up with a more comprehensive 3DMark shootout between cards in this segment. Radeon HD 4890 and HD 4890 CrossFireX are part of the comparison. The testers used an Intel Q6600 CPU running on an X48 motherboard with 4 GB of memory. The article can be read (Google-translated to English) here.

Sources: OC Heaven, Overclockers.com.ua
Add your own comment

103 Comments on Radeon HD 4890 3DMark Performance Revealed

#2
Black Hades
WarEagleAU said:
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.
ATi will get on the Physx deal when DirectX11 is out ;) , and then where do you think Physx will stand?

I'm pretty neutral to the whole AMD/Nvidia thing but I see this as reallity, Physx is going to be just as mainstream as Itanium was (and is) for Intel. No matter how much Nvidia tries to push the matter or promote it.
Posted on Reply
#4
btarunr
Editor & Senior Moderator
Here are some more benchmarks, and comparisons with cards in its range. Also included is CrossFireX scaling in 3DMark. (Google Translated from Russian to English): http://translate.google.ru/translate?prev=_t&hl=ru&ie=UTF-8&u=http%3A%2F%2Fwww.overclockers.com.ua%2Fnews%2Fhardware%2F2009-03-19%2F103501%2F&sl=ru&tl=en&history_state0=&swap=1

The good news is HD 4890 CrossFire should outperform GTX 295 whilst being the cheaper solution (eventually) a-la 2x GTX 260 vs. HD 4870 X2 in those days.

TheMailMan78 said:
I wonder if they are reliable.
I don't :)
Posted on Reply
#5
lemonadesoda
ROPs!? We need more TMUs!

Agreed. From analysis of 4830 performance, it is quite clear that the bottleneck is ROP and TMU not shader.

HOWEVER, scaling shaders is much much easier than redesigning the ROP/TMU stages. It's pimping shader statistics because it is the easiest and cheapest modification.

*****

I hope w1z will benchmark this against the RV770 with identical shader clocks. That way we can see if there really has been any shader optimisations, or if, in reality, the RV790 is *only* an overclocked RV770 (with marginal internal layout changes to avoid hotspots etc.)
Posted on Reply
#6
TheMailMan78
Big Member
btarunr said:
I don't :)
You don't what?
Posted on Reply
#7
btarunr
Editor & Senior Moderator
TheMailMan78 said:
You don't what?
..find them unreliable, though we never know. HD 4850 was listed till the last moment as having 480 SPs. But looking at these benches, I don't think anything more than 800 SPs are on that card, given that it's already clocked high.
Posted on Reply
#8
TheMailMan78
Big Member
btarunr said:
..find them unreliable, though we never know. HD 4850 was listed till the last moment as having 480 SPs. But looking at these benches, I don't think anything more than 800 SPs are on that card, given that it's already clocked high.
:laugh: Ok. Your word is gold then :toast:

My "sources" say that shader count will become more and more important in the coming months. Disclaimer: My source is an AMD shareholders letter. Which Iv used to wipe my ass with in the past.
Posted on Reply
#9
PCpraiser100
Maybe its a driver issue,

All ATI cards start slow thanks to drivers so give it a month or two and it'll be quite a show!:toast:
Posted on Reply
#10
EastCoasthandle
Ok, here is what I read so far...one reviewer said that he benchmarked the video card using Vantage at 1280x1024.
Based on a Q9450 @ 3.60Ghz (testbed appears to be similar between video cards):
4870............ (750/900): P7715
4870 ............ (850/975 @ 3.0Ghz unless that was a typographical error): P8696
HD4890......... (850/975): P9801
GTX280 (600/1300/1100): P9367
GTX285 (650/1475/1240): P10436


DailyTech believes the drivers are not optimized and they will benchmark the card this weekend with another set of drivers. Hopefully we will see some game benchmarks soon.
Posted on Reply
#11
farlex85
EastCoasthandle said:
Ok, here is what I read so far...one reviewer said that he benchmarked the video card using Vantage at 1280x1024:
Based on a Q9450 @ 3.60Ghz a
4870............ (750/900): P7715
HD4890......... (850/975): P9801
GTX280 (600/1300/1100): P9367
GTX285 (650/1475/1240): P10436


DailyTech believes the drivers are not optimized and they will benchmark the card this weekend with another set of drivers.
Where did the 9801 come from? And yes, drivers. When is this card to be released?
Posted on Reply
#12
EastCoasthandle
farlex85 said:
Where did the 9801 come from? And yes, drivers. When is this card to be released?
I've updated my other post to include one more result (4870 at 850/975). The 9801 is the GPU result only from vantage. You can click on the "4890"<--to read that post. Along with this post


4890 top, 4870 bottom...been seen in a few threads already around the net.
Posted on Reply
#13
btarunr
Editor & Senior Moderator
Different ASIC :)

An RV770 from our reviews:


The RV790:


Look at the arrangement of the components on the package, around the die.
Posted on Reply
#16
btarunr
Editor & Senior Moderator
I can confirm that the die-size of RV790 is greater than that of RV770. There is 'more' of something in this GPU.
Posted on Reply
#17
EastCoasthandle
Well that throws the "it's just a higher clocked 4870" out of the window.

Fuds dispelled so far:
It is not a 8+6 pin design specifically
Die size is not the same therefore not just a higher clocked 4870

done by someone else

more to come...
Posted on Reply
#18
btarunr
Editor & Senior Moderator
The PCB has provision for 8+6 pins, but that's very optional. Manufacturers may choose use a 8+6 pin design, if they feel it helps their factory-overclocked models. You will note that the reference PCB for GeForce GTX 285 has provision for 8+6 pins, though the product ended up with only 6+6.

The RV790 is a 'reworked' RV770, not an RV770 with higher clocks. So when one says 'it's just an overclocked HD 4870", it should be taken in context of the point that both GPUs have near-identical specs (so far), and not literally.
Posted on Reply
#19
EastCoasthandle
btarunr said:
The PCB has provision for 8+6 pins, but that's very optional. Manufacturers may choose use a 8+6 pin design, if they feel it helps their factory-overclocked models.

The RV790 is a 'reworked' RV770, not an a RV770 with higher clocks. So when one says 'it's just an overclocked HD 4870", it should be taken in context of the point that both GPUs have near-identical specs (so far), and not literally.
From the context that I've read posted on the issue it did read to me "just a 4870 clocked higher'. In reality (which others have pointed out) that this is a refresh of the RV770. No that does not mean it's a next gen GPU or a different current gen GPU. Just a refresh...which from my prospective is more then just an overclocked 4870. And cannot draw any comparisons of the use of such a term based on the information so far (unless there is actual information to suggest otherwise). However as with any new release like this I look forward to knowing more about what this RV790 is when it's publicly reviewed.
Posted on Reply
#20
TheMailMan78
Big Member
btarunr said:
I can confirm that the die-size of RV790 is greater than that of RV770. There is 'more' of something in this GPU.
Hmmmmmm This is becoming very interesting. Maybe it does in fact have more than 800 shaders.
Posted on Reply
#21
DeathTyrant
Looking pretty tasty. Come on then, let's get some game benchies please!
Posted on Reply
#23
EastCoasthandle
Salsoolo said:
any info on TDP?
I've only found a pic showing idle temps of 900/1000. Which IMO are decent when compared to the 4870 during its release.


Until the card is release we have to take all this pre-released information with a grain of salt.
Posted on Reply
#24
Tatty_One
Senior Moderator
Hmmmm i wonder if these will work in Hybrid with a 4870x2 in CrossfireX??
Posted on Reply
#25
Hayder_Master
ShadowFold said:
Shaders? They need more ROPs!
agree 100% , let they made it with 640SP but with more ROP'S and more texture units
Posted on Reply
Add your own comment