HIS Radeon HD 4870 CrossFire Video Card Review - GDDR5

[Apoptosis]

HIS Radeon HD 4870 CrossFire Video Card Review - GDDR5 Arrives

Last week, AMD was forced to release the ATI Radeon HD 4850 graphics cards early since some retailers had already started selling the card to the public before it was launched. This week AMD has released the the Radeon HD 4870, which is priced at $299. This is the big brother to the Radeon HD 4850 and features higher clock speeds and newer cutting edge technology like GDDR5 memory ICs for improved memory bandwidth.




Article Title: HIS Radeon HD 4870 CrossFire Video Card Review - GDDR5 Arriv
Article URL: http://legitreviews.com/article/734/1/
Pricing At Time of Print: $299

[DX]

The link on the review points to a jpg of the card vs to this thread.

OH and a great info. I wish that AMD had gone with faster GDDR5 instead of the entry level. I wonder just how much of an impact on performance that would have if they had gone full speed there.

[Apoptosis]

The link on the review points to a jpg of the card vs to this thread.

OH and a great info. I wish that AMD had gone with faster GDDR5 instead of the entry level. I wonder just how much of an impact on performance that would have if they had gone full speed there.

Fixed the link... Anything before 7am and my first glass of tea will be buggy ;)

Not too much as you can see from out 21% overclock of the memory on the memory page... You also have to keep in mind that the 4 Gbps GDDR5 is far cheaper than the 5 Gbps GDDR5, so if they went with higher binned parts the price would be over $299 and as you can tell it wasn't worth it. AMD needs a clock boost for higher performance as overclocking was very core clock limited.

Memory is very interesting as NVIDIA used top binned and expensive GDDR3 with a 512-bit bus. The bus alone is expensive as it adds PCB layers. ATI on the other hand when with a smaller bus and expensive new GDDR5 IC's, but they use the slowest bin. AMD claims memory performance is better, more advanced and more cost effective. Time will tell!

[gwolfman]

Hey, thanks for the review. Very interesting to see how it beat the 9800 GTX+ in the particular test even when PhysX was enabled. I'm think I'm still debating between buying and overclocking a 4850 or the 9800 GTX+. Especially with regards to power consumption and heat, I'm leaning towards the nVidia.

[NAiLs]

Wow, very impressive card. When I saw the power consumption and temperatures, my jaw dropped. My 8800 GTX is already too hot for my liking, running at 60*C @ idle and 68*C under load. This card will definitely not be in my books of consideration just based on these two factors.

[MadHatter604]

Hey, nice review
Just want to add that GDDR5 memory is quadpumped, so when overclocking from 900 to 1090MHz, the memory goes from 3600MHz to 4360MHz (and not from 1800MHz to 2180MHz as is stated in the review)...


http://www.qimonda-news.com/download/Qi ... epaper.pdf

Look where it says Signaling and Clocking, 2 write clocks, and 2 bytes assigned to each clock, so x4.

[martini161]

Hey, nice review
Just want to add that GDDR5 memory is quadpumped, so when overclocking from 900 to 1090MHz, the memory goes from 3600MHz to 4360MHz (and not from 1800MHz to 2180MHz as is stated in the review)...

not calling you a lier or anything, and welcome to the forums, but where did you get that info?
nice review btw nate, seems like the 4850 is a great buy especially compared to this

[Apoptosis]

Hey, nice review
Just want to add that GDDR5 memory is quadpumped, so when overclocking from 900 to 1090MHz, the memory goes from 3600MHz to 4360MHz (and not from 1800MHz to 2180MHz as is stated in the review)...

Thanks for pointing that out, the percentage overclock is the same, but the frequency is quad-pumped... So used to just doubling the numbers I didn't think twice when writing that.

[skier]

Thanks for pointing that out, the percentage overclock is the same, but the frequency is quad-pumped... So used to just doubling the numbers I didn't think twice when writing that.

really?! thats new

good read Nate, i wonder if it runs so hot because ATI is making the PCB's such that they can handle higher and higher temps with each release, so in reality, they are comparibly cooler than before, i sure know that after every new technology, everything runs hotter and hotter and takes more power(compare a Pentium 1 to 65nm Intels, or an old 4MB PCI video card to a GTX280, the latter would burn on contact, the first wouldn't even be warm to the touch respectively)

[aircommando13]

sweet comparison, wish the GTX 260 was in there to make things a bit more interesting though

[Apoptosis]

sweet comparison, wish the GTX 260 was in there to make things a bit more interesting though

next week... it takes a solid 4 hours per card to benchmark them, so this article I added these cards

Radeon HD 4850 CrossFire
Radeon HD 4870
Radeon HD 4870 CrossFire
Radeon HD 3870 with 8.6 Drivers
Radeon HD 3870 X2 updated with 8.7b drivers

I put in 20 solid hours of benchmark time this week just on those cards... I have GTX 260's, but I also have a stack of about 40 products to review also. I'll bump the GTX 260 ahead, but right now I'm GPU'd out. Last week I logged the hours I worked on the site and it was 107, so Nate needs a break here soon.

[Methious]

That's amazing Nate, your like the Energizer Bunny of Review sites.

[DMB2000uk]

That's like every waking hour in a normal week!

Thank you very much for the effort, appreciate it.

Take a break, you deserve it.
And you can send those products waiting to review my way while you put your feet up

Dan

[gwolfman]

That's amazing Nate, your like the Energizer Bunny of Review sites.

lol just keep the good reviews coming and we won't complain... too much :p

[Zertz]

Very nice review, nice to see ATI is really back in the game now!

Hey, nice review
Just want to add that GDDR5 memory is quadpumped, so when overclocking from 900 to 1090MHz, the memory goes from 3600MHz to 4360MHz (and not from 1800MHz to 2180MHz as is stated in the review)...

Thanks for pointing that out, the percentage overclock is the same, but the frequency is quad-pumped... So used to just doubling the numbers I didn't think twice when writing that.

I'm pretty sure he got it right the first time. The 3600MHz number really is 3.6Gb/s (bit, not Byte), the memory still works at 1800MHz. Just like AMD HyperTransport is sometimes advertised as 2000MHz, in reality it's 1000MHz and 2000MT/s

[aircommando13]

great effort so far this kind of thing is always appreciated even if people don't say anything. Keep up the good work, not many people can stare at a computer screen that long I'll give you credit for that.

[gwolfman]

great effort so far this kind of thing is always appreciated even if people don't say anything. Keep up the good work, not many people can stare at a computer screen that long I'll give you credit for that.

lol

[w00fd06]

great review! keep em comin!

[squngy]

Thanks for pointing that out, the percentage overclock is the same, but the frequency is quad-pumped... So used to just doubling the numbers I didn't think twice when writing that.

really?! thats new

good read Nate, i wonder if it runs so hot because ATI is making the PCB's such that they can handle higher and higher temps with each release, so in reality, they are comparibly cooler than before, i sure know that after every new technology, everything runs hotter and hotter and takes more power(compare a Pentium 1 to 65nm Intels, or an old 4MB PCI video card to a GTX280, the latter would burn on contact, the first wouldn't even be warm to the touch respectively)

That would be a terrible strategy (as demonstrated by pentium 4),

+ more heat means more wasted energy = bigger power consumption

you might also check what the power requierments were for pentium 1 and old 4MB PCI video cards,
if you ren the sort of power todays components use through them, you wouldn't want to touch them, trust me...

besides: any overclocker can tell you that hot chips arn't wery stable, if you wanted to make a chip run at higher temperatures you would inevitably have to secrifice some performence (doesn't matter if its MHz or space/$ which could be used othervise).