AMD Radeon HD 4890 1GB benchmarked

[Evilsizer]

http://www.hexus.net/content/item.php?item=17770

[Niku-Sama]

well thats not telling much, they compare a 4870 512meg on windows 7 against a 4890 1 gig overclocked and processor over clocked on windows vista.

oooookkkkk

[Rattle]

lol 06 means nothing when it comes to real gaming anway.

i got 17500+ in 06 with a overclocked dual core and overclocked 4870...

[Vengance_01]

I agree, but if we can see 1000MHZ(read it on hard about Asus working on a better PBC to allow these clocks) on the core, most 4870's can barely hit 800. So this alone will push the 4890 way beyond the 4870. Plus any other small tweaks will widen the gap as well.

[vixro]

Very bad benchmark layout. The comparisons are completely biased and on separate types of systems on an old benchmark.

[ChinStrap]

I agree, but if we can see 1000MHZ(read it on hard about Asus working on a better PBC to allow these clocks) on the core, most 4870's can barely hit 800. So this alone will push the 4890 way beyond the 4870. Plus any other small tweaks will widen the gap as well.

this is what grabbed my eye too.

man o man. 1GHz core?!

please please please.... :P

[||Console||]

well thats not telling much, they compare a 4870 512meg on windows 7 against a 4890 1 gig overclocked and processor over clocked on windows vista.

oooookkkkk

i benched 06 vista vs win 7 same score ...

[Niku-Sama]

still not worthy of anything

[Kuroimaho]

Was a vaste of our and their time.

[Kuroimaho]

Legionhardware made a more in depth review with an overclocked 4890 and a 295 with one gpu working only but overclocked to 275 speed. Worth to take a look until we get the real thing. Link (http://www.legionhardware.com/document.php?id=819)

[Farinorco]

Wow. People is really anxious if someone is publishing a "preview" of two new cards trying to emulate the new chips with old ones. It's a enjoyable reading but it gives no real info at all. RV790 is a new chip and we can't be sure that its behaviour is going to be the same than RV770 in a clock per clock basis, even if we can asume that GTX275 can be emulated because all G200 variants are the same chip more or less cut.

Even more, since the day that Legion Hardware published a purposedly faked review of HD3870 made from edited graphs from a previous review (they took another review, changed the HD2900XT to HD3870, and in every game where it had a higher framerate than a 8800GTS 320MB, they exchanged both punctuations too to keep the HD3870 always in the worst place) I don't think they are a trustworthy site in any way ;-)

[flopper]

http://forums.overclockers.com.au/showpost.php?p=10224989&postcount=243

1ghz.

[Badbonji]

http://forums.overclockers.com.au/showpost.php?p=10224989&postcount=243

1ghz.

But the only problem is their shaders are only 1Ghz also, whereas nvidia have much faster ones. I bet that was under LN2 as well.
And from benchmarks the 4890 is a bit dissapointing compared to how much faster it is to a 4870, not much unless heavily overclocked.

[Neuromancer]

LOL A virtual review :P

Phooey

[Craxbax]

Oh goody!!! Another $300+ upper middle tier performance GPU! (Trying to look excited:rolleyes:)

[Dapman02]

But the only problem is their shaders are only 1Ghz also, whereas nvidia have much faster ones. I bet that was under LN2 as well.
And from benchmarks the 4890 is a bit dissapointing compared to how much faster it is to a 4870, not much unless heavily overclocked.

Shader's aren't everything. If they were than nvidia would own ATi in every game (they don't).

[Badbonji]

Shader's aren't everything. If they were than nvidia would own ATi in every game (they don't).

But ATi have so many more SP's which theoretically put it ahead, but it's not. Just seems a little wierd to me.

[dark bishop]

drivers, its all about the drivers.

[ratbuddy]

But ATi have so many more SP's which theoretically put it ahead, but it's not. Just seems a little wierd to me.

ATI measures things differently. What they call a SP is just 1/5th part of what Nvidia does.

[Dapman02]

But ATi have so many more SP's which theoretically put it ahead, but it's not. Just seems a little wierd to me.

There different architectures.
It's like comparing an AMD to an intel. Based on your theory, than a 2.3Ghz Phenom should be as fast as a 2.3Ghz i7. Same thing here.

[Shiggity]

/tap foot for next gen ;)

[rainless]

I've already said my piece about these "Next Gen" cards...

You can take an 8600GT or GTS up to 1000mhz if you just want to see 1000mhz. Hell... you might even get decent fps on Return to Castle Wolfenstein with it!

[Farinorco]

ATI measures things differently. What they call a SP is just 1/5th part of what Nvidia does.

That's not true at all.

They're not measuring the same things in a different way. There's 2 different explanations. The short one:

There different architectures.
It's like comparing an AMD to an intel. Based on your theory, than a 2.3Ghz Phenom should be as fast as a 2.3Ghz i7. Same thing here.

Something more detailed:

NVIDIA's SP are scalar processors. Each one can process a scalar (a single number/magnitude). NVIDIA counts a SP for each of these scalar processors.

ATi's SP are arranged in a very different way. They have vector processors. Each one can process a vector5 (a collection of 5 number/magnitudes). ATi counts a SP for each of the 5 scalar processors contained in the vector processor.

In the best case (from ATi's architecture POV) each of the 5 number/magnitudes in each Vector5 can be used. In that case, you'd need 5 NVIDIA's SP to do the same number of scalar operations.

In the worst case only one of the number/magnitudes can be used. In that case (and only in that case) the vectorial SP from ATi is more or less equivalent to a single NVIDIA's SP.

Of course, additionally to all this each processor works at a different speed (NVIDIA one's work at near 2x speed the ATi ones) and is different so they have different c2c results...

Depending on the code being processed, things can be closer to either the best or the worst case. Shader programs can be very different between them. There are several synthetic benchmarks that test the shader processing capabilities of the hw, and each one is a different program. I've read a shader processing power comparison between RV770 and G200 with several different shader benches, some of them where positive to GTX280, some of them raised HD4870 well above GTX280 (note that I'm talking about a GTX280, not a GTX260).

Which architectural choice is better/more clever depends only on real world applications performance...

[ratbuddy]

That's not true at all.

Yes, it is. It's a very (perhaps over-) simplified explanation but overall, 1:5 is putting them closer to apples to apples.

edit: go read this: http://www.anandtech.com/video/showdoc.aspx?i=3341&p=3

[Farinorco]

Yes, it is. It's a very (perhaps over-) simplified explanation but overall, 1:5 is putting them closer to apples to apples.

edit: go read this: http://www.anandtech.com/video/showdoc.aspx?i=3341&p=3

Nuop. That AnandTech article explains exactly the same that I've done, but giving a lot more of detail (which is what probably got you confused, that's the reason why I've not given this kind of details).

If you do the 1:5 conversion that you're suggesting, you'll be comparing a unit that can process a single scalar at a time, with a composed unit that can process up to 5 scalars at a time.

If you compare the SP count in the way that ATi and NVIDIA gives to us, you'll be comparing units that can process a single scalar at a time each one, that in my book is a much better apples to apples comparison, if such a thing is possible between those so different architectures.

You can't compare a single scalar NVIDIA SP to a group of five scalar SP (ATi's architecture), and that's what you're trying to do...

The thing that probably gets you confused is that in the AnandTech article the author is using the term SP to refer to a whole group of 5 processing units (t,x,y,z,w) in order to compare the clusters (SIMD on ATi, SM on NVIDIA) composition, but even then he is stating that with that naming, an ATi SP can process "5x the number of instructions as NVIDIA's SPs", quoting him textually.

PS: And anyway, you were stating that "ATI measures things differently. What they call a SP is just 1/5th part of what Nvidia does", quoting your own exact words, and that's an absolutely wrong statement, as you can see.

[ratbuddy]

Nuop. That AnandTech article explains exactly the same that I've done, but giving a lot more of detail (which is what probably got you confused, that's the reason why I've not given this kind of details).

If you do the 1:5 conversion that you're suggesting, you'll be comparing a unit that can process a single scalar at a time, with a composed unit that can process up to 5 scalars at a time.

If you compare the SP count in the way that ATi and NVIDIA gives to us, you'll be comparing units that can process a single scalar at a time each one, that in my book is a much better apples to apples comparison, if such a thing is possible between those so different architectures.

You can't compare a single scalar NVIDIA SP to a group of five scalar SP (ATi's architecture), and that's what you're trying to do...

The thing that probably gets you confused is that in the AnandTech article the author is using the term SP to refer to a whole group of 5 processing units (t,x,y,z,w) in order to compare the clusters (SIMD on ATi, SM on NVIDIA) composition, but even then he is stating that with that naming, an ATi SP can process "5x the number of instructions as NVIDIA's SPs", quoting him textually.

I understand just fine, but for a layperson asking 'Why does ATI have 800 SPs and Nvidia only 216/240/whatever? Doesn't that make ATI much better?' it's a perfectly fine explanation ;)

[Farinorco]

I understand just fine, but for a layperson asking 'Why does ATI have 800 SPs and Nvidia only 216/240/whatever? Doesn't that make ATI much better?' it's a perfectly fine explanation ;)

No, it's not. That "perfectly fine explanation" leads to think that a)AMD is cheating with their counting (which is false) and b)a RV770 has less SPs than a G200 (which is false).

If you give that "perfectly fine explanation" to someone, then a new question can arise: "Ahm, and then, if a HD4870 has 160 true SP working at 750MHz, and a GTX280 has 240 SP working at ~1500MHz (I can't remember the exact number, but it doesn't matter), doesn't it make a GTX280 to have 3x the shader processing power?"

Your "perfectly fine explanation" is simply even more flawed than thinking the other way around.

The best explanation to a layperson asking 'Why does ATI have 800 SPs and Nvidia only 216/240/whatever? Doesn't that make ATI much better?' is to tell her/him that they're different architectures, working at different speeds, and that this kind of comparison between different architectures is a nonsense.

Telling lies not even remotely similar to the real thing to the uninformed people is never a "perfectly fine explanation".

[ratbuddy]

Just about done here. Seriously, go read the linked article. The ATI cards, if measuring SPs the same way Nvidia does, would only have 160. Yes, they are different. I know this.

edit: Look at the picture just above where the article says The area in red is actually the SP, but unlike one of NVIDIA's SPs, one of AMD's can handle up to five instructions at the same time. The only restriction here is that all five units have to be working on the same thread.

Yes, it can potentially handle up to 5 instructions at once, but that doesn't make it 5 SPs.

edit again: Not gonna argue this any more. You've clearly already made up your mind.

[Farinorco]

Just about done here. Seriously, go read the linked article. The ATI cards, if measuring SPs the same way Nvidia does, would only have 160. Yes, they are different. I know this.

edit: Look at the picture just above where the article says

Yes, it can potentially handle up to 5 instructions at once, but that doesn't make it 5 SPs.

You're not understanding it.

Each unit YOU and AnandTech are calling SP (not the Stream Processors defined by AMD) comprises 5 units, each one able to process an instruction. The name Stream Processor (SP) is what AMD uses to refer to those units, and it's fine.

Indeed, ATi is measuring SPs "in the same way NVIDIA does", for that matter. Both are counting 1 for each processing unit capable of processing a single instruction at a time.

The fact that ATi units are grouped in groups of 5 (as shown in the Derek's picture you're referring to), with interdependencies and limitations amongst them, and that only 1 inside each group of 5 has the complete functionality, is what makes possible to pack so many processing units in so little space, and what makes them not 1:1 comparable to NVIDIA's.

But if the 1:1 comparison is wrong, a 5:1 comparison is even worst and further from the real world.

PS: I HAVE read the article, but it doesn't say nothing that I didn't know before, thanks ;-)

EDIT to QUOTE:

edit again: Not gonna argue this any more. You've clearly already made up your mind.

Do what you think it's better. But it's not a matter of making up minds, because it's not a subjective matter nor a question of forming an opinion. But if you want to think that to be more happy with yourself, go on.

[Neuromancer]

This raises some questions.

"Best case scenario" = probably does not happen in RW applications.

The current gen of ATI cards perform awesomely, and definitly are equal to any arch in their price bracket. However this scalar SP thing has me wondering.

Since "best case scaneario" uses all 5, and in reality probably not all 5 are being used most of the time. Is there anyway to improve on the utilization of the number of SP in each scalar unit?

Is this merely up to developers to improve game code for ATI cards? Theoretically, 4 times the number of halfspeed SPs should well outperform the faster but less prolific sp's of another video card. Yet, this is not the case. Is it because each SP on the scalar SP is assigned only a specific task, so when lighting is complete this millisecond, it is waitign for texture or physics or something to finish so it can process the next bit of lighting effect?

I would not think that hte parallel processing a GPU would have the same limintaions of current CPU processing, but it is interesting that More slower "cores" in both cases, does not perform as well less faster "cores" in games.

[Rich'[ard]]

what the...
that Legion Hardware benchmark, it shows the 4890 as a very very slow card. it's only marginally faster than a 4870. this better not be true :p

[Farinorco]

NEW HD4890 review found! With several games, at several resolutions, it compares against a GTX285, GTX260, HD4870 1GB, and it shows results with stock clocks and overclocked to 1GHz.

WOW, if this review isn't a fake, nor is otherwise flawed, this is a piece of a beast of card. This is the first more or less complete HD4890 review I see, and it seems it can compete very well against GTX285, even at stock clocks. Even more if we consider the price difference.

German original link: HD4890 review @ GameStar (http://www.gamestar.de/hardware/tests/grafikkarten/pcie/1955014/test_radeon_hd_4890.html)

Google translation: HD4890 review @ GameStar (english) (http://translate.google.com/translate?prev=_t&hl=es&ie=UTF-8&u=http://www.gamestar.de/hardware/tests/grafikkarten/pcie/1955014/test_radeon_hd_4890.html%0D%0A%0D%0A&sl=de&tl=en&history_state0=es|en|colaborar)

This raises some questions.

"Best case scenario" = probably does not happen in RW applications.

The current gen of ATI cards perform awesomely, and definitly are equal to any arch in their price bracket. However this scalar SP thing has me wondering.

Since "best case scaneario" uses all 5, and in reality probably not all 5 are being used most of the time. Is there anyway to improve on the utilization of the number of SP in each scalar unit?

Is this merely up to developers to improve game code for ATI cards? Theoretically, 4 times the number of halfspeed SPs should well outperform the faster but less prolific sp's of another video card. Yet, this is not the case. Is it because each SP on the scalar SP is assigned only a specific task, so when lighting is complete this millisecond, it is waitign for texture or physics or something to finish so it can process the next bit of lighting effect?

I would not think that hte parallel processing a GPU would have the same limintaions of current CPU processing, but it is interesting that More slower "cores" in both cases, does not perform as well less faster "cores" in games.

Only when programming on R700 assembly (making use of the R700 ISA) those questions would affect a developer, and that's something that is never going to happen. Developers work at their lower level when using API calls as DirectX and OpenGL or their shader languages (HLSL, GLSL), often not even that (you can use a higher level rendering engine built over those low level APIs, as I do myself), and drivers are the only responsible to capture those API calls and communicate with the GPU with its own instruction set.

Of course, the code made by developers can give more or less facilities to be converted to GPU instructions, but it uses to be far away from the developer scope... AFAIK.

I've not got a deep knowledge about R700 architecture (and it's the one that I know better :D ) so I can't give anymore than some hints (and probably some inaccurate ones). I'll try: there's 2 types of SP in R700 architecture. From each group of 5, one is a "big fat" complete unit "t", and 4 are "thin" units "x,y,z,w". There are some functionalities only available to the fat unit, so demand of thin units can be less than 100% for starters. On the other hand, I've read that only completely independant instructions can be "packed" in the same vector causing more units to be unused, but I don't figure the whole picture in the paralel computation architecture we are talking about... in a similar way than yourself XDDD.

Anyway, what is clear is that in order to put so many processing units together in so little space, ATi has had to make some sacrifices. And those sacrifices leverage the gains to a greater or lesser extent... which is better choice, that's a matter of real world performance... but if your question is "are developers wasting most of the power of R700 GPUs with their programming practices?", I'd say "no, it's not their responsability to look at that level", even if they could do a more R700 friendly code.

[Kuroimaho]

Hmm so it's like 10% faster than the 4870, the previews weren't that off now were they ? Ok, thx next gen please.

BTW you guys should look into a bit of floating point performances of the cards if you want to compare the archs, but I guess it wasn't accidentally ignored. ;) The streaming processors are restricive enough to unlikely to pull out the theoreticla maximum performance even with hand tweaking not to mention without. The two are closer in real life ATi has the higher maximum output and the lower minimal as well.

[largon]

I think I would be just fine with my G92GTS until real next gen comes - but I believe I'm going to have to grab an RV790 and OC it until burts into flames.

[e6600]

two 4770's would be exceptionally nice

[Mr.Guvernment]

glad i held out with my 4850 for this long and didnt get a 4870 1G yet, either way i win, a cheaper 4870 1G for me or a 4890...

[mastrdrver]

Anyone know if there is any validity to the rumor that the 4890 will move to 40nm process, if all goes well, in the summer?

[Kuroimaho]

With AMD's small die strategy doing shrinks totally makes sense, but no one can tell whether TSMC can pull itself together.

As current 4890 prices look I would threw a bit more at it and get 2 4870.

[Niku-Sama]

glad i held out with my 4850 for this long and didnt get a 4870 1G yet, either way i win, a cheaper 4870 1G for me or a 4890...

yea tell me about it, wheres this damned price drop they were talkin about eeeeh?

either way looking at the ATI/nvidia SP thing i think if ATIs SPs were simplified mabe they would perform a bit better
i mean theyre still good cards but more SPs at the same performance....but then theres the less power thing.

i dont have one yet but i am up there with Guv, price drop or new card. i am sure glad they didnt pull an nvidia and rename an old chip for a new product line.

[Rich'[ard]]

it seems the new 4890 performs as well as 4850 Crossfire. i'm making this on the assumption that 2 4850s perform about the same with the GTX285 (as does the 4890 in Farinorco's link)

so it's the new 4850 X2 :p ...if you take it that way.
hopefully drivers can push it beyond the 285's performance.

[MadMan007]

This review is much better and has a real GTX 275: http://www.techpowerup.com/reviews/Sapphire/HD_4890/ The latter looks pretty badass, GTX 280/285 performance for <$250. Based on that I'd say the GTX 275 is generally a little better card than the HD4890 for around the same price. Pretty high load power draw and temps though compared to the HD4890 but in line with performance.

[Kuroimaho]

If you use aftermarket cooling the TR8800GT fits right on the 4890 while the 275 or 280 which at least here seem to be the same price needs a new cooler which isn't compatible with anything else.
That's one big plus for the ATi cards, with NV likely moving to a new small die arch GTX 200 coolers likely end up in a dumpster while I would be surprised if ATI would change the layout for nextgen.

[Niku-Sama]

thats a good review at techpowerup. interesting that at 55nm its a bigger die....hrmm mabe its got something to do with the bulit in repairability....potental unlockable shaders perhaps?

[Evilsizer]

thats a good review at techpowerup. interesting that at 55nm its a bigger die....hrmm mabe its got something to do with the bulit in repairability....potental unlockable shaders perhaps?

not from what i read in one of the articles. they stated the more transistors is from the DECAP, a bunch of capacitors that are used for filtering. which is how they were able to get the GPU speed up even more.

[Kuroimaho]

That was a rumor which got dispelled rather quickly
.
However, to reach these clock levels, AMD revised the core by adding decoupling capacitors, new timing algorithms, and altered the ASIC power distribution for enhanced operation. These slight changes increased the transistor count from 956M to 959M. Otherwise, the core features/specifications (texture units, ROPs, z/stencil) remain the same as the HD4850/HD4870 series.

[ratbuddy]

That was a rumor which got dispelled rather quickly
.

Got a link? The [H] review I just read says it's basically the same thing as a 4870 but somehow reworked and ended up with 3M more transistors for whatever reason..

[Azetab]

seems like I should get the GTX 275 instead of the 4890, its draws less power and has lower temps and seems more power efficient while having the same performance and price as the 4890

[Kuroimaho]

It has 3M more transistors, I meant the additional backup SPs will be put to use rumor was killed.

[MadMan007]

seems like I should get the GTX 275 instead of the 4890, its draws less power and has lower temps and seems more power efficient while having the same performance and price as the 4890

That was what I thought after reading a few reviews but after reading more things are just all over the place. It's very frustrating when you can't get something resembling a consensus from a bunch of pretty reputable sites.

[Kuroimaho]

Indeed, I felt the same results are all over the place depending on which site I read. Well it means they are pretty close on performance get whichever fills your other priorities best.

[Niku-Sama]

from the things i have seen the 4890 is less of a power user and makes less heat.

this is strange, mabe the companies are shipping out different cards or different bios on the cards or different drivers to the reviewers creating inconsistancies in pre launch reviews.

or every one is rushing the reviews and screwing them up

[Dapman02]

from the things i have seen the 4890 is less of a power user and makes less heat.

this is strange, mabe the companies are shipping out different cards or different bios on the cards or different drivers to the reviewers creating inconsistancies in pre launch reviews.

or every one is rushing the reviews and screwing them up

or it's the minor differences in chipsets, fresh windows install, about a thousand different other variables. I think the 4890 would work best for me because of my dragon platform.

[mastrdrver]

Got to remember there are 4890 reviews and there are 4890 oc reviews. Not everyone has a non oc card in their reviews. Got to check too and see if the other cards are reference clocks or not also.

Reference clock speeds, numbers in parentheses is effect memory speed:

ATI Core/Memory
4850: 625/993 (1986)
4870: 750/900 (3600)
4890: 850/975 (3900)

nVidia Core/Shader/Memory
GTX 260: 576/1242/999 (1998)
GTX 275: 633/1404/1134 (2268)
GTX 280: 602/1296/1107 (2214)
GTX 285: 648/1476/1242 (2484)

[Hardin]

seems like I should get the GTX 275 instead of the 4890, its draws less power and has lower temps and seems more power efficient while having the same performance and price as the 4890

In the benchmarks I've seen the 4890 uses less power and is quieter. Results are all over the place as well.

[Rattle]

someone should tell tech power up that the gtx 275 only has 28 rops not 32

[Firestrider]

Got a link? The [H] review I just read says it's basically the same thing as a 4870 but somehow reworked and ended up with 3M more transistors for whatever reason..

Link (http://www.anandtech.com/video/showdoc.aspx?i=3539&p=3) from Anandtech.

Picture (http://www.firingsquad.com/media/hirez.asp?file=/hardware/ati_radeon_4890_nvidia_geforce_gtx_275/images/07.png) from FiringSquad.

I can't seem to find the original powerpoint from FiringSquad, though.


ATI says they’ve completely reworked the ASIC to enable higher frequencies; some 3 million transistors have been added to RV790, and the chip boasts a slightly larger die than its predecessor. So where did the new transistors go? With RV790, ATI has added decoupling capacitors in order to reduce the signal noise you’ll inevitably encounter as you begin to crank up the clock speeds. A decap ring completely encircles the ASIC. ATI also says they’ve retimed the entire chip to run at the higher frequencies. Power delivery has also been tweaked to the point where RV790 actually consumes slightly less power than RV770 at idle.




The primary logical architecture behind the RV790 remains completely unchanged from that of the RV770. That means the performance per shader per clock will be consistent on the RV790 compared to the Radeon HD 4870. What has changed is the physical layout of the chip (the GPU designers essentially spaced things out to better improve attributes like heat and timing) with what AMD calls the “decap ring.” That ring, added along the outside of the RV790 GPU itself, helps to reduce leakage and voltage fluctuations allowing AMD and its partners to clock the core higher than the RV770 could go.

[ravaneli]

Tom's Hardware has the best comparison review I've seen of these two. The 275 appears to be a hair faster at stock speeds, but the 4890 might end up a couple % faster after max overclock.

I think either way it's obvious both cards are quite even and certainly none can claim to be a better deal. ATI fanboys will get the 4890 and Nvidea fanboys will get the 275.

May I say a special note was made in Tom's hardware that the 275 fan is remarkably quiet.

I personally tend to lean to NVidea because their new product drivers are more reliable. All of the last ATI fast cards had horiffic driver issues in the beginning, and even now not everything is a smooth sail.

[Rinne]

yep, both cards seem to perform pretty well..

GTX 275 seems to run quieter on stock cooling, but 4890 seems to OC like hell, even on stock.

Would like to see how they compare on high end aftermarket cooling.

[Rich'[ard]]

for me, noise means nothing. i could bare the 4890's noise all day long, because my Coolermaster PSU sounds like a vacuum cleaner.

i'm either going to get myself another 4850 or buy the 4890 when i save up enough $$

[rainless]

;6050148"]for me, noise means nothing. i could bare the 4890's noise all day long, because my Coolermaster PSU sounds like a vacuum cleaner.

i'm either going to get myself another 4850 or buy the 4890 when i save up enough $$

Since you already HAVE a 4850 you might as well just get another one for like a hundred bucks.

[Niku-Sama]

i notice alot of the 4890's for $229 after rebate, any one know about how long these will last at this price?

[ghost_recon88]

i notice alot of the 4890's for $229 after rebate, any one know about how long these will last at this price?

Probably a while, I don't think they're gonna jack the price. If anything they'll lower it.

[ratbuddy]

Probably a while, I don't think they're gonna jack the price. If anything they'll lower it.

Indeed. The performance gained over the $160 4870 1GB just isn't that much, especially for another $70. Nvidia gonna have the same problem moving the GTX 275 compared to the 260, if current prices hold. I'll go out on a limb and predict a sub-$200 4890 within a month.