HD 5870 Discussion thread.

[a_ump]

This is just something that i've been wondering since i saw the results of the HD 5870's performance in w1z's review here.

Like the graphs' show the HD 5870 only ever outperforms the HD 4870x2 at 1024x768, which doesn't really matter since no one in their right mind that games at that res would purchase such a card. But i thought i'd heard a couple or few times somewhere that ATI might have trouble getting all their shaders utilized n that they had some trouble with utilizing even the 800 shaders on the HD 4890/70/50. Now i don't own any HD 58XX series card(yet), but I had thought that since the HD 5870 has all 1600 shaders on a single die and double the ROP's that the HD 5870 would outperform the HD 4870x2 by a decent margin esp once AA was enabled but it doesn't. Am i the only person that thinks this? I know we can expect at least 10% more performance with mature drivers, probly up to 20%, but i'm still curious if what i'd read a few times is or could be true with ATI having a difficult time managing all 1600 shaders.

[Mussels]

the fact that a single GPU card even comes close to last gens dual GPU card is rare, and impressive.


expecting it to whup it in every way is reaching - especially since they made it more power efficient too

[Jstn7477]

I believe that the main bottleneck of the card is the RAM. It only has 256bit GDDR5, which doesn't seem to be enough bandwidth for the GPU. When one overclocks the RAM on one of these cards and leaves the GPU core @ stock, the performance of the card increases noticeably. If the RAM were not a bottleneck, then overclocking the RAM would have little or no effect on performance unless the GPU were also overclocked.

[LittleLizard]

agree with jst, altough technically 256bit GDDR5 = 512bit GDDR3 it feels like its a bit slow for such gpu.

[Jstn7477]

Originally Posted by LittleLizard

agree with jst, altough technically 256bit GDDR5 = 512bit GDDR3 it feels like its a bit slow for such gpu.

Yes, 256bit GDDR5 was a tad bit sluggish even for the 4890 @ 1000MHz core, IIRC.

[a_ump]

Originally Posted by Mussels

the fact that a single GPU card even comes close to last gens dual GPU card is rare, and impressive.


expecting it to whup it in every way is reaching - especially since they made it more power efficient too

eh i wouldn't say rare
HD 4870>HD 3870x2
GTX 280>9800GX2
8800GTX>7950GX2

but this time at launch..
HD 5870<HD 4870x2
lol.

well the price for the HD 5850 is great i think so im still getting that asap. Though no one has mentioned yet on whether what i'd heard bout ATI having difficulties getting all the shaders utilized on the HD 5870 or not :/, anyone?

[erocker]

Originally Posted by Jstn7477

Yes, 256bit GDDR5 was a tad bit sluggish even for the 4890 @ 1000MHz core, IIRC.

No, it most definitely is not. If this was the case the 5870 would be bottlenecked to 4890 speeds.

As far as the 5870 goes, it's what I expected I guess. The 5850 seems to be the better card as far as price/performance.

[LittleLizard]

Originally Posted by erocker

No, it most definitely is not. If this was the case the 5870 would be bottlenecked to 4890 speeds.

As far as the 5870 goes, it's what I expected I guess. The 5850 seems to be the better card as far as price/performance.

and performance/watt

[Jstn7477]

Originally Posted by erocker

No, it most definitely is not. If this was the case the 5870 would be bottlenecked to 4890 speeds.

As far as the 5870 goes, it's what I expected I guess. The 5850 seems to be the better card as far as price/performance.

Oh, that is true. Someone said what I said when the 5870s came out, and got me confused lol. Must be those trolls lol jk.

[a_ump]

i wonder how many on TPU in a poll would say that the HD 58XX series met their expectations. I'm satisified with the HD 5850's price/perf for sure, but both let me down some performance.

[erocker]

Originally Posted by a_ump

i wonder how many on TPU in a poll would say that the HD 58XX series met their expectations. I'm satisified with the HD 5850's price/perf for sure, but both let me down some performance.

Maybe you should actually use one first?

[Bo_Fox]

Yeah, the 5870 is only like 25% faster than a GTX 285, which is hugely disappointing.. damn!!!

EDIT: I mean 28%, according to the graph below, if I divide 100% by 78%:





It's all about only 23% greater memory bandwidth than a HD 4890 that already uses roughly 100% of its memory in half of the newer games out there.

That is precisely why a 5870 does not destroy a 4870X2 although it should at least always be equal or better than 2x 4890 Xfire, when only taking the GPU cores into account apart from the memory.

[a_ump]

Originally Posted by erocker

Maybe you should actually use one first?

lol i would have, but my original thought was just about what i'd read on ATI n their shaders, the rest were just things i thought when i typed so the original question was bout that ATI thing but then i revised cause it didn't sound right....poll didn't even cross my mind then

[wolf]

I expected more (performance), but bought one anyway

I'd say it needs more memory bandwidth, but ATi are good at their R&D, if it *needed* more bandwidth I'd imagine they would have given it more.

I think its a pity it doesn't outshine a 4870X2 at high res+AA, with an exact doubling +100mhz I expected more, thus IMO it has to be the memory bandwidth OR drivers holding the 58xx cards back at the mo.

[DRDNA]

Given a little time for the drivers to mature then you will see the 5000 series shine....I doubt there will be disappointment then.

[wolf]

Originally Posted by DRDNA

Given a little time for the drivers to mature then you will see the 5000 series shine....I doubt there will be disappointment then.

disappointment, who said that

The fact that it's sooooo close to my old GTX295 makes me happy, too happy...

[a_ump]

lol i guess if i had one of them i wouldn't be saying a thing just be awed by the performance i'd see, esp jumping from this 7800GTX, though i had a HD 4870 bout a month ago but sold to save for the 50....but money ended up getting spent XD, now saving again lol. I also haven't checked to see if the new official drivers helped any.

[Exeodus]

Originally Posted by DRDNA

Given a little time for the drivers to mature then you will see the 5000 series shine....I doubt there will be disappointment then.

Exactly, it took a good couple driver releases when the 4800 series came out before the cards really started to shine. And I know because I had the 4850, 4870, 4870x2, and the 4890.

[Binge]

Originally Posted by Jstn7477

I believe that the main bottleneck of the card is the RAM. It only has 256bit GDDR5, which doesn't seem to be enough bandwidth for the GPU. When one overclocks the RAM on one of these cards and leaves the GPU core @ stock, the performance of the card increases noticeably. If the RAM were not a bottleneck, then overclocking the RAM would have little or no effect on performance unless the GPU were also overclocked.

Originally Posted by LittleLizard

agree with jst, altough technically 256bit GDDR5 = 512bit GDDR3 it feels like its a bit slow for such gpu.

Originally Posted by Jstn7477

Yes, 256bit GDDR5 was a tad bit sluggish even for the 4890 @ 1000MHz core, IIRC.

There's more of a problem with cache on chip that's slowing down this GPU than the memory bandwidth, but the low bus width does not help matters. There needs to be a room stocked with food and overclockable computers that ATI can keep a few level headed enthusiasts as a design conscience. It seems that more than one of us can see a flaw in design.

[a_ump]

Originally Posted by Binge

There's more of a problem with cache on chip that's slowing down this GPU than the memory bandwidth, but the low bus width does not help matters. There needs to be a room stocked with food and overclockable computers that ATI can keep a few level headed enthusiasts as a design conscience. It seems that more than one of us can see a flaw in design.

lol that would be funny to hear if it were a reality. "I live at ATI in this room with food an graphic cards." though honestly it sounds like a good idea.

[temp02]

I remember when I bought my HD 2600 XT and in most cases it was slower than my previous X700 (in some cases it was even slower then a Radeon 7000 or a Riva TNT, like when rendering OpenGL lines), it took almost 6 months for ATi to sort this (and other) driver issues and make my card deliver the performance it was supposed to at launch times.
I'm not saying that this new generation of ATi cards is slow (they most certainly are not, just look at them scaling at higher resolutions), all I'm saying is that ATi is kinda slow on driver fixes but eventually they will sort it out and make their cards *shine*.

[Steevo]

Good to know so many here have the resources to actually try out their theory, wait.......



W1zz did the test already, and found that the 5870 is NOT bandwidth limited, but core limited to a small degree. I don't know how many times this needs to be reiterated, or if you guys just need to read the white paper on GDDR5. It has TWICE the bandwidth of GDDR3 so it is equal to a 512MB bus, and with shorter trace lengths due to learning it is faster, and at higer speeds......

From the 5770 Review.

"The overclocks of our card are 925 MHz core (9% overclock) and 1500 MHz Memory (25% overclock). The core overclock is ok, but the memory overclock is simply amazing. Remember, on GDDR5, for every MHz real clock you increase effective clock by 4 MHz. "

"The actual 3D performance gained from overclocking is 11.4%."




If the performance increase was larger based off the huge 25% memory overclock then you could confirm the bandwidth starvation, however at a mere 2% more than the core overclock you can determine the minor increase the memory overclock gave was do to a few cache misses and the ability to deliver data at a much higher rate.


Same for 5870.

From the review.

"The overclocks of our card are 890 MHz core (5% overclock) and 1275 MHz Memory (6% overclock). While they are not spectacular, they are decent and promising. Remember, we left the fan running at its default speed. Also we did not tweak any voltages, even though software adjustment is possible on these cards - there is just no software yet to do that."


"The actual 3D performance gained from overclocking is 4.9%."


Again, if the memory was the limiting factor the number would have increased more linear with the memory overclock, however it didn't. It followed the GPU core speed. Perhaps the reason he had a slower amount was the clock ratio at those speeds was out of sync, and the GDDR5 was also still erroring during transmission.

[a_ump]

Originally Posted by Steevo

From the 5770 Review.

"The overclocks of our card are 925 MHz core (9% overclock) and 1500 MHz Memory (25% overclock). The core overclock is ok, but the memory overclock is simply amazing. Remember, on GDDR5, for every MHz real clock you increase effective clock by 4 MHz. "

"The actual 3D performance gained from overclocking is 11.4%."

If the performance increase was larger based off the huge 25% memory overclock then you could confirm the bandwidth starvation, however at a mere 2% more than the core overclock you can determine the minor increase the memory overclock gave was do to a few cache misses and the ability to deliver data at a much higher rate.

i understand what your trying to say but the HD 5770 is hardly part of this argument, it has exactly half the graphics horsepower so it cannot speak for the statements of the HD 5870 and the questioning of it being bandwidth starved or not. The HD 5770 had 96GB/s of bandwidth with that oc, the HD 5870 doubles everything of the HD 5770 except for the memory bandwidth. Now in w1z's performance charts, the HD 5870 with exactly double the processing power, it is only 63% faster overall and 82% faster at 2560x1600, so there's definitely something that isn't as effecient as it should be though i personally think drivers now.

Same for 5870.

From the review.

"The overclocks of our card are 890 MHz core (5% overclock) and 1275 MHz Memory (6% overclock). While they are not spectacular, they are decent and promising. Remember, we left the fan running at its default speed. Also we did not tweak any voltages, even though software adjustment is possible on these cards - there is just no software yet to do that."

"The actual 3D performance gained from overclocking is 4.9%."

Again, if the memory was the limiting factor the number would have increased more linear with the memory overclock, however it didn't. It followed the GPU core speed. Perhaps the reason he had a slower amount was the clock ratio at those speeds was out of sync, and the GDDR5 was also still erroring during transmission.

performance increase isn't going to be linear with overclocks, we all know that, but 4.9% isn't far from 6% at all. I haven't seen any review from w1z that show's him disproving the HD 5870 being bandwidth starved, could you link it for us? And as has been stated a guzillion times, a bandwidth starvation test should be done by decreasing the core clock and checking performance with stock clocks, not increasing memory and core clocks. Only memory or core clocks should be altered if someone were to do a test for the bandwidth suspicion we have, and it should be equal and done by percentages. Example: if you decrease core clocks by 5%, memory at stock, then you should also run through the same benchmarks with core clocks@stock and memory clocks increased by 5%

[erocker]

Not a dissapointment. You know what is dissapointing right now? Nvidia. Very dissapointing. I'm using a 5850 right now, and I am the opposite of dissapointed. I can't even buy a new Nvidia card to play with. We haven't even seen any real proof of a DX11 Nvidia card. Wow, that's dissapointing. My shoelace broke this morning. I'm dissapointed. Here's W1zzards conclusion to his review:

Originally Posted by W1zzard

AMD definitely surprised me, and pretty much everybody else with their new products. I neither expected such a big performance increase, nor many of the other new features that AMD has put in their card. Looking at pure performance numbers, one has to admit that AMD did not claim the performance crown this time, their HD 4870 X2 and the NVIDIA GTX 295 are still faster. However, those are dual GPU cards, AMD's dual HD 5870 dubbed "Hemlock" will be out later this year to shatter all performance records we've seen so far.
Not only performance increased, AMD has also kept a tight lid on power consumption and die size didn't increase that much either, thanks to the 40 nm production process. I find the power consumption improvements very impressive. Especially in idle, the card is one of the most efficient cards you can find on the market today - and you still have all its rendering power just a click away. Under load the power draw is higher, but not nearly as high as performance numbers would suggest when applied to the last generation of cards. Looking at Performance per Watt, the Radeon HD 5870 leads by 23% over the next closest card, the GTX 295. While it is not the only improvement, a large power hog could be tamed by clocking down the GDDR5 memory when it is idle. Past cards had GDDR5 too, but due to the complex clocking mechanisms it wasn't possible to change the clock without an annoying flicker that would confuse users. Now the flicker is so short that it can be hidden between the monitor refresh intervals, enabling downlocking of GDDR5 in idle.
This latest generation of ATI graphics cards also brings support for DirectX 11, which includes a state of the art tesselation engine and support for DirectX Compute. Another area where we saw improvements are the multimedia features. The integrated audio device that is used for HDMI audio is now at HDMI 1.3a which means support for several new audio formats. As far as I know this puts the audio capabilities of the HD 5800 Series on par with anything that any external sound card can provide, making it an ideal choice for a media PC system.
Even though AMD has paid great attention to power, noise levels are not as low as I'd wished for. In idle the card's noise levels are very acceptable, but the idle temperatures are so low that I see no reason why the card couldn't be quieter and run a bit warmer. Under load the card gets fairly loud, but considering the performance, I'd say it's doing alright. Here again, some fan tweaking can give you higher temps and less fan noise if you want.
Finally, let's talk about your hard earned money. While not insanely expensive, those new cards can not claim the price/performance throne. They are high-end cards and that itself comes with a price premium. Being brand-new and having all the latest and greatest features adds a few dollars to the price tag, too. Still, I don't find the cards terribly expensive considering what they offer. If you are looking for the best deal on the market right now, you should look at other products like the HD 4870 which comes at extremely competitive prices at the moment. Also I already heard rumors that AMD's MSRP pricing will not hold for long and we'll soon see two digit $ price drops. Let's hope AMD has enough stock of these cards to supply all the gamers that can't wait to hold one in their hands now.

What I gather from that. Not dissapointing.


Like I said, I think you need to use this card before saying it's dissapointing. It most certainly is not. This thread is somewhat dissapointing.

[Steevo]

So just ignore that a 5770 is half of a 5870? Nnnnk.

Originally Posted by Zubasa

Some Quick Run on 3Dmark Vantage GPU Tests:

First Some using "High" settings:
http://img.techpowerup.org/091004/5870 GPU 850-1200.jpg
http://img.techpowerup.org/091004/5870 GPU 750-1200.jpg
http://img.techpowerup.org/091004/5870 GPU 650-1200.jpg

Custom Setting "Extreme" @1080P and 8xAA 16xAF:
http://img.techpowerup.org/091004/Settings.jpg
http://img.techpowerup.org/091004/58...850-1200 X.jpg
http://img.techpowerup.org/091004/58...750-1200 X.jpg
http://img.techpowerup.org/091004/58...650-1200 X.jpg


Some Results messing with the Memory Clocks:

"High":
http://img.techpowerup.org/091004/5870 GPU 850-1000.jpg

1080P:
http://img.techpowerup.org/091004/58...850-1000 X.png
http://img.techpowerup.org/091004/58... 850-900 X.jpg