Radeon HD 5870 Six with 6 display outputs, prices, performance

[zanzabar]

Good, more cards for the rest of us

i doubt that there will be cards with display port as the norm

[informal]

Maybe it's 20% faster than a GTX295 and 40% faster than a 4870X2?

Yeah i feel your sarcasm,but there is no need for it . The Cypress GPU has double the specs of the RV770/790 so one doesn't have to be a design team member to roughly guess what its performance will be. There is a question will the GDDR5 used on it be a bottleneck or not,but the 5870 IS going to outperform 4870X2 or match it in some worst case. Also there is some talk that 5870 trails a bit or roughly matches the GTX295 ,which is roughly (surprise surprise) 25-30% faster than 4870X2

[xXxDieselxXx]

In case you didn't see this. I looks like fresh from the oven:

http://alienbabeltech.com/main/?p=11135

Some Highlights
According to our own sources, the new HD 5870 offers over 1600 Stream processors. Amazingly AMD doubled the number of SIMD units from 10 to 20. Still every SIMD unit contains 16 5D units and a Quad-TMU. Overall: 1600 stream processors and 80 TMUs. We are talking about a videocard whose core is at 850 MHz and whose 256-bit GDDR5 runs at 1200 MHz – all for the suggested retail of $399! AMD is expecting HD 5870 to come close to the performance of a HD 4870-X2 or GTX 295.

There is a new ERM eyefinity which means 3 LCDs can be simultaneously supported at 2560×1920, with options for future cards to support six LCDS!!

The die size is 330 MM2 and packed with over 2.1 billion transistors. This translates into one beast of a card with just over 150 GB/sec. It is very likely that the 8X + 8X PCIe CrossFire slots of the new p55 motherboards for Core i5 will become saturated.

What is outstanding is that we are hearing that the HD 5870 will perform at just over 26 watts at idle and peak below 190 watts maximum!! That is quite a challenge for Nvidia to meet or beat in their own upcoming GT 300 series.

[Frontl1ne]

Yeah i feel your sarcasm,but there is no need for it . The Cypress GPU has double the specs of the RV770/790 so one doesn't have to be a design team member to roughly guess what its performance will be. There is a question will the GDDR5 used on it be a bottleneck or not,but the 5870 IS going to outperform 4870X2 or match it in some worst case. Also there is some talk that 5870 trails a bit or roughly matches the GTX295 ,which is roughly (surprise surprise) 25-30% faster than 4870X2

Haha, there was no sarcasm in my post ! Just wishful thinking

I am certain ATi would not leave such an obvious bottleneck unattended on their cards? After all, I'm sure the majority of their development teams have had much more experience designing video cards than most people on the forums...

[lockee]

The holes are actually there just for aesthetics. PCIe plugs are pointing to the same direction as the CF fingers.

I hope its for air intake. Drawing in extra air could help keep the fan speed down making things quieter.

[labs23]

Now on waiting mode for the benchies and the actual date for the hard launch. Which I think will happen also today, no?

[Solus Corvus]

I hope its for air intake. Drawing in extra air could help keep the fan speed down making things quieter.

They aren't air intakes. Air intake flows down from the top of the fan and gets pushed radially away from the axis of rotation. If those were air intakes the air would have to flow against the push of the fan.

Surely they are for decoration, or air outlets from the VRM section.

Speculating about the performance from this article is kinda pointless since he didn't mention what card in the current gen it is being compared to.

[Chumbucket843]

Haha, there was no sarcasm in my post ! Just wishful thinking

I am certain ATi would not leave such an obvious bottleneck unattended on their cards? After all, I'm sure the majority of their development teams have had much more experience designing video cards than most people on the forums...

i am certain doubling the shader count theoretically doubles processing power therefore requiring double the bandwidth. 120GB/s on the 4870 and now 150GB/s on the 5870. we will have to wait for faster gddr5 to see this card reach its maximum potential.

[zerazax]

i am certain doubling the shader count theoretically doubles processing power therefore requiring double the bandwidth. 120GB/s on the 4870 and now 150GB/s on the 5870. we will have to wait for faster gddr5 to see this card reach its maximum potential.

No that's not true at all. The bandwidth might be more than is necessary for the card. Excessive bandwidth won't make the card much faster.

Look at the reviews for overclocking cards out there such as the 8800GTX. Increasing memory speed gives far less performance increase than increasing the core or shaders.

The fact that ATI was able to use the same core across GDDR3 (4850) and GDDR5 (4870) suggests that the actual bandwidth needed for the processing power probably lay somewhere in between what the 4850 and 4870's bandwidths were. Doubling the processing power means it should still be fine with the memory bandwidth given

[zerazax]

Oh and...

The biggest piece of evidence for me is that if the 5850 = $299, and 5870 =$399 is actually true, then that's a pretty big indicator of performance in relation to the current generation and the next generation. If a 5870 matches or beats the GTX295 at $399, that would be the perfect price to compete at and sell cards at the same time

[Jowy Atreides]

In case you didn't see this. I looks like fresh from the oven:

http://alienbabeltech.com/main/?p=11135

Some Highlights
According to our own sources, the new HD 5870 offers over 1600 Stream processors. Amazingly AMD doubled the number of SIMD units from 10 to 20. Still every SIMD unit contains 16 5D units and a Quad-TMU. Overall: 1600 stream processors and 80 TMUs. We are talking about a videocard whose core is at 850 MHz and whose 256-bit GDDR5 runs at 1200 MHz – all for the suggested retail of $399! AMD is expecting HD 5870 to come close to the performance of a HD 4870-X2 or GTX 295.

There is a new ERM eyefinity which means 3 LCDs can be simultaneously supported at 2560×1920, with options for future cards to support six LCDS!!

The die size is 330 MM2 and packed with over 2.1 billion transistors. This translates into one beast of a card with just over 150 GB/sec. It is very likely that the 8X + 8X PCIe CrossFire slots of the new p55 motherboards for Core i5 will become saturated.

What is outstanding is that we are hearing that the HD 5870 will perform at just over 26 watts at idle and peak below 190 watts maximum!! That is quite a challenge for Nvidia to meet or beat in their own upcoming GT 300 series.

Sounds beastly.

[Chumbucket843]

No that's not true at all. The bandwidth might be more than is necessary for the card. Excessive bandwidth won't make the card much faster.

Look at the reviews for overclocking cards out there such as the 8800GTX. Increasing memory speed gives far less performance increase than increasing the core or shaders.

The fact that ATI was able to use the same core across GDDR3 (4850) and GDDR5 (4870) suggests that the actual bandwidth needed for the processing power probably lay somewhere in between what the 4850 and 4870's bandwidths were. Doubling the processing power means it should still be fine with the memory bandwidth given

lol. did you see the word theoretical. it depends on what you are doing but more bandwidth is generally better.

the reason you dont see a huge increase from ocing memory is because memory doesnt oc well. try cutting your memory speed in half and run a benchmark.

your 4850/4870 argument doesnt make any sense considering how much slower the 4850 is.

[kadozer]

Hoping for some leaked benchmarks sometime tonight.

[labs23]

Hoping for some leaked benchmarks sometime tonight.

There's a guy at Anand saying he's got the card(5870) already, but points out the lack of drivers as of this time. http://forums.anandtech.com/messagev...IEWTMP=Linear&

[xlink]

your 4850/4870 argument doesnt make any sense considering how much slower the 4850 is.

umm, when the cores are at identical speed... it's not THAT much slower.



WOW around 5%(when both cards have the same core speed).... WOWOWOWOWOWOWOWOW...

admittedly the gddr5 is of higher latency but the difference isn't night and day....

and cutting that laggy gddr5 in half bandwidth wise, is only around a 25% hit... and it'd be less if latency wasn't so bad, probably around half since when comparing identically clocked cards the latency hit is around 13%....

[largon]

im really disappointed i hate display port and will not be buying anything with it

[JohnZS]

The HD5870 sounds epic, especially if it's performance is comparable with that of a GTX295! I am not a fan of MultiGPU solutions at all, however if these new single GPU cards are as fast as the current top crop MultiGPU cards then we are talking huge performance increases here folks.
The power consumption (if 190W Max is true) is also most impressive.
Unless nVidia have a true Ace of Spades up their green sleeved suits I can see ATi winning this round.
Good to finally see display port being implemented in Graphics Cards... it would be interesting to see how this works on those new fancy Dell U series IPS screens
Roll on with the Reviews already!
John

[SoulsCollective]

Urgh. DisplayPort is not a good thing, folks. No quality improvements over HDMI or plain old DVI, some theoretical advantages in some areas that currently nothing takes advantage of, infested with DRM crap.

That being said, it's just a little bit silly to declare that you won't buy any cards with DP connections.

[blindbox]

I don't get the hate for DisplayPort. Isn't it an open standard? There's no DRM, no royalty fee to pay for HDMI and has higher bandwidth compared to HDMI (and just a bit higher compared to DVI dual link).

Someone mentioned about latency, but googling doesn't help me find its relation to latency.

[SoulsCollective]

I don't get the hate for DisplayPort. Isn't it an open standard? There's no DRM, no royalty fee to pay for HDMI and has higher bandwidth compared to HDMI (and just a bit higher compared to DVI dual link).

Someone mentioned about latency, but googling doesn't help me find its relation to latency.

HDMI has more than adequate "bandwidth" to carry digital data at resolutions even higher than what we consider "extreme" now (ie. 2560), and I didn't think anyone would call the ability to carry full 7.1 audio data over the same cable "useless" (DisplayPort has the same functionality anyway - would you call it useless in that context?). Furthermore, HDMI supports xvYCC, which we should all be using anyway.

But this isn't about HDMI><DP, this is about DP itself - the supporters are advocating it as "HDMI for your PC", which sounds neat, but the problem is the DRM - the number of consumer-level displays which are HDCP certified is tiny compared to the number of otherwise perfectly good displays that will be needlessly and pointlessly rendered useless if DP gains wide acceptance. If Big Content (R) and Big Beige Boxes (TM) get together and enforce this standard on us, we all lose.

The point is, it isn't needed, it isn't even desirable when ordinary DVI works just fine for even the highest resolutions available today, and really it has no purpose other than to restrict consumer rights and choice.

See Post #38 onwards for context.

[jaredpace]

really it has no purpose other than to restrict consumer rights and choice.

Exactly.

[AKM]

In case you didn't see this. I looks like fresh from the oven:

http://alienbabeltech.com/main/?p=11135

Some Highlights
According to our own sources, the new HD 5870 offers over 1600 Stream processors. Amazingly AMD doubled the number of SIMD units from 10 to 20. Still every SIMD unit contains 16 5D units and a Quad-TMU. Overall: 1600 stream processors and 80 TMUs. We are talking about a videocard whose core is at 850 MHz and whose 256-bit GDDR5 runs at 1200 MHz – all for the suggested retail of $399! AMD is expecting HD 5870 to come close to the performance of a HD 4870-X2 or GTX 295.

There is a new ERM eyefinity which means 3 LCDs can be simultaneously supported at 2560×1920, with options for future cards to support six LCDS!!

The die size is 330 MM2 and packed with over 2.1 billion transistors. This translates into one beast of a card with just over 150 GB/sec. It is very likely that the 8X + 8X PCIe CrossFire slots of the new p55 motherboards for Core i5 will become saturated.

What is outstanding is that we are hearing that the HD 5870 will perform at just over 26 watts at idle and peak below 190 watts maximum!! That is quite a challenge for Nvidia to meet or beat in their own upcoming GT 300 series.

It's good to see that AMD makes efforts to improve load power consumption on it's high performance video cards. A big thumbs up for AMD

[570091D]

The new HD 5850 will launch on September 23 also. We hear it is priced below $299. HD 5850 will sport 1440 stream processors and it will have lower clockspeeds than its big brother. We are hearing somewhere around 700/1000 MHz and it can also display simultaneously on three LCDs at up to 2560×1600 resolution.

here's the weenie.

[FischOderAal]

Well, how much damage does 160 SP less do? If it still is near HD-4870X2 (keep Crossfire limitations in mind) it will be a great card! 100 Dollar have to come from somehwere, otherwise noone would buy HD-5870.

http://www.xtremesystems.org/forums/...d.php?t=234155
Thefud has speaken. 1440 SP clocked at 725 MHz. The HD-5870 is at 850 MHz, so HD-5850 is clocked app 15 percent lower.

[jaredpace]

First rumored benchmark is Crysis:

HD5870 Crysis Benchmark Score
CPU: AMD Phenom II X4 955BE
Win 7 RTM
VGA: HD5870 1GB

Crysis 1900x1200 4AA +16 AF DX10 Very High
min: 30 .**
avg: 43 .**
max: 54 .**




http://74.125.159.132/translate_c?hl...ivsFsnZPqEd56Q


TPU 4890 Xfire review:
CPU: Intel Core i7 920 @ 3.8 GHz
Software: Windows Vista SP1
Drivers: NVIDIA: ForceWare 181.20, GTS 250: 182.06, GTX 275: 185.63
ATI: Catalyst 9.1, HD 4890: 8.592.1


(dont know how w1zzard gets his ati products to run so smoothly ;P)

http://www.techpowerup.com/reviews/A...ossFire/6.html

Well looks like it is faster than 4890 Crossfire and GTX295