ATI Radeon HD 4890: Playing To Win Or Played Again?

[maxcue]

Does anyone know for sure if adding an ATI card FASTER than my 4850 in Crossfire defaults to the slower card, like when you add faster RAM and it slows the faster RAM to the slower speed, or would the 4850 act as I hope, a performance boost for a faster card? I'm thinking just adding a cheap second 4850 is the best value but would love adding a 4870 1GB. I really appreciate your help and the best way to answer is at my email: maxcue@gmail.com but the correctness of the answer is more important. Thanks again! Dave S.

 

[2shea]

that 2 gig card will most likely perform the same as the 1 gig version, that memory isn't going to be accessed faster or used quicker.
The die build doesn't let the gpu use that much memory efficiently.

 

[Onus]

URL: http://media.bestofmicro.com/9/2/193430/original/image032.png
Category: none

Chris, I can no longer view your benchmark graphics (and many other graphics from THG) because the metadata field for the type of site has not been filled in. My company has recently blocked ALL sites with this metadata blank due to security concerns. If you can have this checked, I (and possibly others with similar issues) would appreciate it. Thanks.

 

[cangelini]

[citation][nom]gerrieM[/nom]Would just like to point out a small error. Farcry 2 was not developed by Crytek it's a ubisoft montreal game. On page 6 under the benchmark score you say "In sharp contrast to Crysis, Crytek’s Far Cry 2 is easily playable at the highest resolutions.". Monteal guys might get abit miffed[/citation]

Thanks for the catch Gerrie, fixed!

 

[cangelini]

[citation][nom]jtt283[/nom]URL: http://media.bestofmicro.com/9/2/1 [...] age032.png Category: none Chris, I can no longer view your benchmark graphics (and many other graphics from THG) because the metadata field for the type of site has not been filled in. My company has recently blocked ALL sites with this metadata blank due to security concerns. If you can have this checked, I (and possibly others with similar issues) would appreciate it. Thanks.[/citation]

Thanks for the head's-up jtt--I'll have this looked into!!

 

[Guest]

What about Performance per dollar?

 

[bounty]

I find the power consumption characteristics between review sites interesting. I know everyone has different setups etc, but I would think the cards power consumption would remain stable relative to each other.

It doesn't seem to happen that way though. So a 4890 should usually be a little lower than a 4870 at idle and 25-35 higher at peak etc. w/o overclocking. A 285 should be lower at idle and peak etc. The Xbit review makes the 4890 look like the power freaking master and the 285 a sloppy furnace. Interesting.

-Bounty

 

[crockdaddy]

I am sure most of you read two to three tech sites. This helps get a best spread for reviews. Typically for me I hit Toms, Anands, Xbit, and a few others from time to time.

 

[sandmanwn]

Speaking of Anand, boy did they fudge up a review over there. Some secret handshakes going on to get a 275 Nvidia part that won't hit the shelves any time soon and handicapping a brand new 4890 with 8.12 hotfix drivers from last year and 4 versions back.

The THG review was worlds better. Good one Chris.

 

[bustapr]

couldn't you guys just use another benchmark stting for the 4870 512mb for gta4? I mean the 4890 is a good card , but i believe that some people would also like to know of the way cheaper 4870 512mb and how it does in gta4. You cant just leave a blank benchmark on a card that, according to you guys, can play gta4 at resolutions such as 1680x1050. I would really have liked to see those benchmarks.

 

[swat565]

Any possibility of getting some benchmarks for 4890's in CF, or did they only send 1 card?

 

[cangelini]

Only sent one card, but I'll work on lining up more cards, getting my GTX 295 replaced (it died unceremoniously as I was doing SLI testing on the GTX 275), and putting up a follow-up piece.

 

[Ryun]

Nice review. After having read others, in particular the XbitLabs review: http://www.xbitlabs.com/articles/video/display/radeon-hd4890.html the power consumption at idle there was drastically lower. Other reviewers had power consumption similar to what you were getting, or a little bit less.

However, I'm wondering why the variance. The drivers are mostly the same everywhere else, and so are the platforms. That got me thinking that maybe the Core i7 platform is having issues with the drivers again (like the low performance you saw with the GTX 280 on a i7 platform). I doubt the processor is doing anything to influence the idle power draw, so maybe it's the X58 chipset?

At any rate, XbitLabs tested power draw on a platform that differed from their Core i7 testbed for performance testing. I know it's a lot to ask but is there any way you could do the power tests with the 4890 again on a different platform?

 

[rmicro1]

I thought I read somewhere that there were architual changes that would need to be taken advantage of with software updates. So basically games now do not use these advantages. I will try to find where I read this to confirm.

 

[cangelini]

Ryun,

I'll keep this in mind for any update with CrossFire/SLI and see if I can re-test on a different platform.

I was also quite surprised by the numbers on the GTX 275 story, but Nvidia didn't seem to be. The power consumption on the 4890 actually looked like it could be about right, save the idle power, which should have been lower according to ATI. FWIW, though, all of the power-saving settings on the i7--the things that'd be most likely to add variability to the results--were turned off. Chipset is an interesting idea, though, if it's changing power to the link itself.

 

[cangelini]

[citation][nom]rmicro1[/nom]I thought I read somewhere that there were architual changes that would need to be taken advantage of with software updates. So basically games now do not use these advantages. I will try to find where I read this to confirm.[/citation]

This was likely what you read when the X2 launched. There is capacity for more bandwidth between the two RV770 GPUs, but as far as I know nothing has come of that potential, as of yet.

 

[randomizer]

This product is selling for $443+ in Australia, which is about $100 more than it should be with current exchange rates... The 4870 1GB is now down to $310 or so though, so it looks like a good deal.

 

[ira176]

What would be interesting to me is if there are plans on releasing a 40nm part in the near future. I'm sure there would be cooler operating temps. and lower power usage, and higher overclockability.

 

[cangelini]

As you might be aware, ATI's strategy as of late is to push process enhancements at the mainstream. As a result, you can expect to see 40 nm soon, but not at the high-end until later ;-)

 

[LLJones]

Got to love benchmarks and how they can be so different. This site, http://www.hardwarecanucks.com/forum/hardware-canucks-reviews/16365-sapphire-radeon-hd-4890-1gb-review.html , has the 4890 mostly winning against the 275 and pretty much staying beside the 285. They also managed to overclock to: Final Core Overclock: 1Ghz
Final Memory Overclock: 1200Mhz (4800Mhz effective)
I wish all the major sites would get together and design an I.S.O.(International Organization for Standardization) benchmark.

 

[RazberyBandit]

[citation][nom]cangelini[/nom]Both the 2 GB card and the 4850 X2 are exclusive to Sapphire, and neither has been sent over. Nevertheless, we'll be following up with SLI/CrossFire scores in the near future and I'll see if either of those two solutions might be lined up for that story.[/citation]

Sent over? Buy them, then? Heck! Beg, borrow, or steal them if you must. Everyone desires real-world, head-to0head results, but we want them for all the best cards available. That includes the rarer, perhaps even outcast cards. How can you even begin consider all your benchmark comparisons "thorough" when you haven't got all of the players available in the game, competing in a proper (read as, thorough) comparison?

And for what it's worth, why not offer some benchmarks on a more "average" system instead of a brand new (dare I say top-of-the-line?) Core i7 system? For that matter, why are you running all these tests on just one system in particular? If you insist on running on the latest architecture available, why aren't you also doing so on an AMD-based machine as well? Not everyone has an Intel system, you know...

You could at least run these same tests on a new Phenom II AMD system to see how well these same cards perform on AMD's newest architecture, not just Intel's. Who knows... It might just be completely different, and I really think you should consider that possibility. That, or prove such a possibility completely wrong by doing it and never have to worry about it again. Either way, we all win - Everyone gets a much better idea of what these cards would actually run like on their own system.

Consider it... It just makes sense!

 

[randomizer]

[citation][nom]LLJones[/nom]I wish all the major sites would get together and design an I.S.O.(International Organization for Standardization) benchmark.[/citation]
That would be I.O.S.

Note that this would require all of the same hardware, and would defeat the purpose of having all of these sites run benchmarks. Two lots of benchmarks would be sufficient, as you wouldn't expect any real difference if everything was "standard".

 

[LLJones]

Ay Hoi Hoi,

Your right, it should be I.O.S.,but for some reason they call themselves I.S.O. French maybe. The equipment does not need to be standardized as they do that when they build the test bed. I would like to see the settings agreed upon. Example. Are we going to use the in game demo's or run through a certain part of the game, AA on or off, tweaking the game commands, yes or no, 2 or 3 games that favor card A, and 2 or 3 games that favor card B, etc.

I like RazberyBandit's idea of using an AMD set up. I think I saw a review once using AMD but I can't remember where.

Pure speculation on my part, but don't they use the intel setup because all they have to do is change the card?

One last thought. Why is nvidia named after an evil green princess from a 70's anime?

 

[ManusDeus]

"too bad price goes up exponentially for minimal improvements... the 4890 will be about %50 more than the 4870 "

That's why SLI and Crossfire make sense to lengthen the usable life of your graphics pipeline--you can kick up you gfx power when prices drop.

 

[Ryun]

[citation][nom]cangelini[/nom]Ryun, I'll keep this in mind for any update with CrossFire/SLI and see if I can re-test on a different platform. [/citation]

Thanks, that would be awesome if you could. I would do it myself, but I don't have the money or the sources to test the 4870 and the 4890. =)

At any rate here's another site that does not use a Core i7 platform that reports the 4890 (stock) is 15 watts less than a 4870 and pretty close to the GTX 260 when idle: http://www.hardwarezone.com/articles/view.php?id=2849&cid=3&pg=8