PowerColor Radeon HD 5870 LCS: The GHz Limit, Broken

[nurgletheunclean]

[citation][nom]Cleeve[/nom]*sigh*It's marketingspeak, nurgle. You're being duped by numbers that they use to sell more cards. There is no relation to real-world performance. You can call it 256-bit or 512-bit but the end result is the same bandwidth. Like I said, you put ten 256-bit cards in a box and you can technically say the box has a 2560-bit interface, but that doesn't mean squat.[/citation]
I am accutely aware of marketing hype. However you stated that AMD claimed the same bandwidth for their 4870x2 as their 4870 which they don't.

Furthermore I never said that reducing the resolution increased bandwidth, you just said that to be an a-hole.

I simply insinuated that the 5870 could be memory bandwidth limited. looking at the last graph on page 6, imagine that a single 4890 is capeable of 17FPS. At 17FPS the 4890 has more than enough memory bandwidth to keep the GPU fed. Now imagine that the 5870 which is mathematically double the 4890 yet has marginaly more memory bandwidth. Now the GPU is fast enough that it has outrun the memory bandwith of the card.

If there was a 2nd 4890 like in the example above. and that one could simply chug along with it's own memory and send another 17FPS to the screen dropping 2 for overhead. Is it unreasonable to think that there's some extra memory bandwidth being utilised? Is AMDs 512bit claim completely baseless? In a dual CPU configuration where each CPU is working only from it's own cache, is it unreasonable to combine the cache throughput levels of each cpu working on SMP enabled code?

I would like to believe that each 4890 can use about 94GB/s (the other 30GB/s being headroom) in that scenario and each one renders 15.8FPS combined under crossfire to produce the 31.6FPS.

mathematically the 5870 with it's 156GB/s produces 25.7FPS which is just about right.
2x94GB/s = 188GB/s / 31.6FPS = 5.95GB/s per FPS
156GB/s / 25.7FPS = 6.07GB/s per FPS

 

[masterjaw]

Nice article. Looks like those LCS-equipped 5870's aren't that much attractive as a better solution to the stock one except for those who are already have WC setup.

Trolls are roaming.

 

[cleeve]

[citation][nom]nurgletheunclean[/nom]I am accutely aware of marketing hype. However you stated that AMD claimed the same bandwidth for their 4870x2 as their 4870 which they don't.[/citation]

They do. Depends where you look. Frankly, I'm apalled that AMD's site lists it as a 512-bit part, although I haven't yet seen AMD's site list it as a 250 GB/s part.

[citation][nom]nurgletheunclean[/nom]Furthermore I never said that reducing the resolution increased bandwidth, you just said that to be an a-hole.I simply insinuated that the 5870 could be memory bandwidth limited.[/citation]

I didn't say that to be an a-hole, I said it to point out that if you're suggesting that two 4890's have double the bandwidth compared to a single card, you're inevitably suggesting that bandwidth is based on resolution. As far as the cards are concerned, they are each rendering half-frames which equates to half resolution, their memory buses are in no way linked to provide more bandwidth per GPU.

Simply put,the 5870 has more usable bandwidth than a pair of 4890s. It is therefore not memory bandwidth limited in comparison.

[citation][nom]nurgletheunclean[/nom]looking at the last graph on page 6, imagine that a single 4890 is capeable of 17FPS... ...I would like to believe that each 4890 can use about 94GB/s (the other 30GB/s being headroom) in that scenario and each one renders 15.8FPS combined under crossfire to produce the 31.6FPS... ... 2x94GB/s = 188GB/s / 31.6FPS = 5.95GB/s per FPS156GB/s / 25.7FPS = 6.07GB/s per FPS[/citation]

Sweet Jesus! Are you actually suggesting that you can calculate frames per second based on memory bandwidth alone? And you 'prove' this by reverse-engineering numbers from a single benchmark, assigning an arbitrary amount of bandwidth for 'headroom'? Then you don't bother applying this calculation to other benchmarks to see if you can prove your theory?

I don't think I've ever seen anyone trying to pass off something like this as plausable. If you honestly think that kind of pseudo-scientific approach works I'm not sure you are capable of understanding how ridiculous it is.

 

[spearhead]

would not be supprised if AMD would launch a 5890 model later on. 1+ GHZ on the core has now been proven to be quite possible. even without drastic messures being taken. Installing some faster memory chips say 5+ ghz effective or perhaps more around 5,4 -5,8 would make the card a good 15-20% faster over stock just like 4890. I dont expact this very soon tough. AMD has no competition right now and they need every chip they can make to sell as 5870 so increasing the clocks isnt something they want to do right now. But i expact a possible 5890 model once AMD is unable to sell there 5870 cards above €200 any longer just like with 4890. It is a good tactic for them. I just hope Nvidia shows off its GTX380 and GTX360 models to compete rather sooner then later. I want a price drop of the 5870 too 200 euro soon 😀. But first these taiwanese guys as TSMC have to ramp up there production it sucks hell that they dont meet demand. 🙁

 

[excalibur1814]

Is it worth the expense for an extra 5fps on Crysis? No, absolutely not. Is it worth it to have a very nice and shiny slice of kit? Yes

 

[zodiacfml]

why is the i7 overclocked only to 3.06 GHz?

 

[cleeve]

[citation][nom]zodiacfml[/nom]why is the i7 overclocked only to 3.06 GHz?[/citation]

Because this article is about overclocking the graphics card and making graphics card comparisons. There is really no need to go further with the CPU.

 

[zodiacfml]

Yes but won't there be a possibility for that cpu speed holding back one of the cards or both? I mean, these cards I think are better and faster than quad sli of the past.

 

[cleeve]

[citation][nom]zodiacfml[/nom]Yes but won't there be a possibility for that cpu speed holding back one of the cards or both? [/citation]

An i7 920 @ 3.06 GHz is no slouch by any standard. It runs a mere 270 MHz slower than the fastest CPU on the planet, the 975. I don't think there's much concern that it's not doing the platform justice.

 

[fball922]

This game is so CPU-limited with these powerful graphic solutions that the results are tied across the board.

Why do they continue to use this as a benchmark then? I have always wondered this... The game is not remarkable, except in the number of characters on the screen. Why continue to ask if it pushed the limits of graphics cards when everyone knows it is CPU limited?

 

[MARSOC_Operator]

All that fancy watercooling ain't worth the dough afer all.

 

[cnyte]

Very nice graphic card, the top graphic card for now.

 

[cnyte]

A awesome graphic card but i wouldnt buy it because its pricey right now and you dont need all that power in that graphic anyways

 

[Guest]

Thanks for the article. I just received two of these 5870 LCS for Xmas. Now ready for crossfire!

 

[leon2006]

I'm holding on to my CF 4890s. I tried an ASUS 5870 and did not get a performance boost compare to my CF 4890s. This benchmark confirmed my observation.

There is no question that the new card consumes less power. I won't recover that $400 in power savings in a year.

 

[Guest]

Yes. once again, I'm not claiming to irst to break a GHz. As mentioned in the article, I was talking about breaking the GHz limit this particular card has in the BIOS.

Cards have been overclocked over a GHz for a long time now.

You probably never read this reply, but I was talking about cards that come stock 1GHZ from AMD out of the FABS.
At the time of writing I was aware of overclocked 4,7Ghz GPU's.
This is not the first card that breached the 1Ghz barrier stock, read the TOM's archived articles!

 

[MarkJohnson]

Which Asus BIOS do you flash to break 1GHz? I see more than 1 model to choose from and I know Asus makes customs boards a lot and want to make 100% sure it's the same board design.