The GT300/Fermi Thread - Part 2!

[tajoh111]

tajoh, your up in the clouds man...

Is 20% that much. From the initial hype, people were expecting alot more. I am just saying the 5870 and the gtx 470 matching which isn't expecting that much considering the transister advantage of the gtx 470.

I thought everyone was expecting gtx 480 to be significantly faster than the 5870 until Charlie doom and gloom article.

Is 20% percent really that much considering the low clocks on the gtx 295?

576 mhz is how fast the gtx 295, even Charlies estimates are significantly faster than this clock figure.

[flopper]

And if Nvidia proves me wrong, I'll gladly eat my shorts. And post pics.

used or new?

[hurleybird]

yes, but why should software suddenly, magically, catch up? why should there not only be a lot of dx11 games, but good dx11 games, and then not only good dx11 games but good dx 11 games that use compute shaders so much that gf100 has an advantage from it? i just dont see that happening... sure, eventually games will demand a lot more tesselation and compute shader power, but by then we will have second and most likely third or 4th gen dx11 hardware and all this first gen dx11 stuff will be useless.

I think most of us expect Fermi to be a tesselation monster, but what makes anyone think that Fermi will also have an advantage in computer shaders? Even though Fermi should be able to reach a higher percentage of it's theoretical compute speed, it seems unlikely that Fermi will be able to make up for it's massive 1+ TFLOPs performance deficit that way. in doublr percision it will be even worse, where HD 5870 will be several times faster thanks to Nvidia crippling the DP performance of their Geforce line.

It might be interesting in a couple years time looking at how future proof these two architectures were. We might see HD 5870 leading in compute heavy games, and GTX 470/80 leading in tesselation heavy games.

Lastly, there's the driver question. Fermi is a (relatively) new architecture so one would think that Nvidia will be able to get quite a bit more performance from it as they tune their drivers to the architecture. On the other hand, even though Fermi hardware is late that doesen't mean driver development is as well. Nvidia's driver team might have used this extra time to do a lot of performance tuning already.

ATI's driver situation is even more questionable. For some reason performance took a dive clock/clock and unit/unit from RV770, despite supposed enhancements. Recent 10.2 drivers and 10.3 beta drivers show some performance increase, but the question remains whether future drivers can put performance back to, or perhaps even beyond, RV770 levels. Perhaps ATI is intentionaly holding back major performance improvements so that they can attempt to ruin the Fermi launch with a 'magic driver'

[mindfury]

Is 20% that much. From the initial hype, people were expecting alot more. I am just saying the 5870 and the gtx 470 matching which isn't expecting that much considering the transister advantage of the gtx 470.

I thought everyone was expecting gtx 480 to be significantly faster than the 5870 until Charlie doom and gloom article.

Is 20% percent really that much considering the low clocks on the gtx 295?

576 mhz is how fast the gtx 295, even Charlies estimates are significantly faster than this clock figure.

GF100 A3 is hardly clocked over 650Mhz from what I heard.

[tajoh111]

GF100 A3 is hardly clocked over 650Mhz from what I heard.

A2 is already at 625 or more. I can't see them getting anything less than 650 for a3.

[mindfury]

I'm 100% sure GTX480 is 480SP now.

[Behemot]

Tell more...

[annihilat0r]

apparently it's 480sp with 700 something mhz

[informal]

So they cut on the CUDA cores some(480 instead of 512) but they add up some Mhz to offset this? From perf. point of view they are still in the same place...

[Andrea deluxe]

http://www.donanimhaber.com/GeForce_...andi-18522.htm

[Mungri]

Thats one of the special edition 5870 coolers with all the labels take off. At least it looks 100% identical to one of them, think it was either the MSI Lightning or the Asus one.

And I think I can see black spray paint or photochop on the top picture.

[saaya]

Is 20% that much. From the initial hype, people were expecting alot more. I am just saying the 5870 and the gtx 470 matching which isn't expecting that much considering the transister advantage of the gtx 470.

I thought everyone was expecting gtx 480 to be significantly faster than the 5870 until Charlie doom and gloom article.

Is 20% percent really that much considering the low clocks on the gtx 295?

576 mhz is how fast the gtx 295, even Charlies estimates are significantly faster than this clock figure.

i wasnt referring to your perf estimates, i think those are optimistic but possible... with the "up in the clouds" i was referring to your argumentation, looking for some reason why gf100 will perform well...

a1 clocked like this, and a2 like this, so maaaybe a3 clocks like this, and maaaaaaybe everybody had a2 cards so far, so the final cards will clock higher, and maaaaaybe drivers are causing the low perf of early cards too

your words about gf100 being faster than rv870:

should be, supposedly, if, should, if, I think, possible, if, should, likely, should, might, if

your clearly looking for some light at the end of the tunnel to grip and pin your hope to

Charlie writes about NV like they are amateur that don't have a right to make silicon because they are too incompetent

actually he kept giving the nvidia engineers credit for their work and performance.. he has been saying all along that it was management giving the engineers a wrong direction or bad and unrealistic goals and not listening to their engineers and thats what caused the issues... makes sense to me, we all know nvidias engineers are capable of excellent designs, the only problems they had in the past is when they were pushed to do things that just werent possible, at least with the resources and time they got for it. nv30 failed because the top guys wanted 32bit precision for shaders while game devs and ati thought 24 is enough and much faster, chipsets failed because the top guys wanted an intel chipset asap, wanted a flexible fsb/mem bridge, wanted ddr3 support etc without giving them the resources to get it all working properly...

A2 is already at 625 or more. I can't see them getting anything less than 650 for a3.

new rev =! higher clocks.. in some cases yes, but not necessarily...

So they cut on the CUDA cores some(480 instead of 512) but they add up some Mhz to offset this? From perf. point of view they are still in the same place...

not really, their rumored target was 750/1500 with 512cores, if they are at 480cores with 650mhz clocks they have 7% less cores and 15% lower clocks.

512 cores with 750/1500 clocks might actually be close enough to a 5970 to compete with it, so that would be a 40% perf advantage... and im sure nvidia will EVENTUALLY get there... so their pr guys never actually lied when they claimed 40%+ above 5870

but again, who really cares if its 15% faster or slower?
if the rumors about 10K total cards until summer is true, then the price will either skyrocket and only rich nvidia fans will be able to get them, or there will be small pockets of availability here and there every few weeks with people crashing etailers websites trying to get one

http://www.donanimhaber.com/GeForce_...andi-18522.htm

looks nice... direct contact heatpipe, if done right, can work really well to lower even idle temps enough to reduce idle fanspeeds... and 1.8A means the fan has enough power to remove quite some heat at full load...

[NaMcO]

I'm 100% sure GTX480 is 480SP now.

If this is true, i'll have to pick up my template on this graphic and change the video card inside the bin

[spursindonesia]

Comparing GTX 470 cooling solution vs HD 5870 one, which one is relatively more beefy ? Does HD 5870 HSF posses 4 heatpipes ?

[NaMcO]

The 5870 cooler seems to be bigger and it also features 4 heatpipes:



They are not in direct contact with the GPU though, and that might make a difference...

[tajoh111]

Perhaps Saaya. I am not sure I really want to buy the card. I actually want the product to come out and it to perform reasonably well so we can get all this harping against NV done with(atleast some of it) and we can have competition again.

At this point those all we have is alot of rumors(the NDA is solid on this one) and all we can do is shoot in the dark. I am just not being lead by Charlie because honestly, one of the biggest reason I want NV to succeed is I want Charlie to look like an ass. He already does for his biased writing but I want him to be wrong so he can learn to think before he writes.

Its also exciting when truly new stuff comes out rather than bumped up specs of older architectures.

[Andrea deluxe]

I'm 100% sure GTX480 is 480SP now.

on gtx480 are 12bga/128mb and 32bit

12*128mb=1536mb video memory
12*32bit=384bit
384bit=48rop
48rop=512sp

[saaya]

Comparing GTX 470 cooling solution vs HD 5870 one, which one is relatively more beefy ? Does HD 5870 HSF posses 4 heatpipes ?

http://www.techpowerup.com/reviews/A...HD_5870/4.html
they look very similar... id say that heatsink has thicker fins and is a bit shorter than the 5870 heatsink though...
that heatsink they show looks like a better design, it has different hight levels for the heatpipes penetrating the fins, no big differece, but it looks better...

[saaya]

on gtx480 are 12bga/128mb and 32bit

12*128mb=1536mb video memory
12*32bit=384bit
384bit=48rop
48rop=512sp

ciao andrea, are you sure the blocks are related?
i thought they can kill some sps and rops WITHOUT losing imcs on fermi?
so they could even make a 240sp fermi with 512bit imc...

Perhaps Saaya. I am not sure I really want to buy the card. I actually want the product to come out and it to perform reasonably well so we can get all this harping against NV done with(atleast some of it) and we can have competition again.

At this point those all we have is alot of rumors(the NDA is solid on this one) and all we can do is shoot in the dark. I am just not being lead by Charlie because honestly, one of the biggest reason I want NV to succeed is I want Charlie to look like an ass. He already does for his biased writing but I want him to be wrong so he can learn to think before he writes.

Its also exciting when truly new stuff comes out rather than bumped up specs of older architectures.

i dont think the 470 will be a good buy, unless your into ln2 benching, then maybe... the 480 will be great for ln2 im sure, and it should be a good card for people with 30" displays as well. but i highly doubt that it will affect rv870 pricing... at all... and even if you could find one for the msrp, which i doubt, it would be priced so high it really doesnt make sense if you dont have a 30" screen or want to bench on ln2...

about charlie... well you know, if hes actually wrong and perf is better than 5% above 5870 for the gtx480, then people will see that as a success and he will actually have helped nvidia to make their cards look good
the more people he convinces that fermi sucks, the more people nvidia can impress easily with fermi being "ok-good"

charlie is a cool guy, really... dont hold any grief against him... he may be wrong, he may be right... for what its worth, he made the waiting a lot more interesting and dropped a lot of hints, most of which turned out to be true. im pretty sure he didnt guess, but even if he did, he was pretty much spot on so far, so... so i dont get why people get so emotional when it comes to charlie

[mindfury]

on gtx480 are 12bga/128mb and 32bit

12*128mb=1536mb video memory
12*32bit=384bit
384bit=48rop
48rop=512sp

ROP's amount is not related to SP's amount.

5850 only has 1440SP,but full 32 rops.

GTX260-216 increased 24SP,but stayed with the same ROP's amount as GTX260-192.

[ajaidev]

GTX 470's cooler seems smaller than the 5870's but the fins seem ticker...

Does not make much difference both seem about equal in mass. The voltage for the fan seems high *1.8v* how much was 5870's fan voltage?

If GTX 480 does have 480 SP's that means it will only have 32 more SP's and if we presume that GTX 470 is around 650Mhz and GTX 480 is 700Mhz how much of a speed bump will 32 SP and 50Mhz more give GTX 480 because according to most data GTX 470 ~ 5870 "stock".

A 5870 Matrix with 2GB's and 850Mhz may be a big problem for GTX 480's and not to mention 5870 eyefinity 6 2GB which is rumored to be cheaper than $500...

[mindfury]

GTX 470's cooler seems smaller than the 5870's but the fins seem ticker...

Does not make much difference both seem about equal in mass. The voltage for the fan seems high *1.8v* how much was 5870's fan voltage?

12V,0.8A

[SKYMTL]

ROP's amount is not related to SP's amount.

This. The ROP / Cache / Memory controllers scale together but separately from the CUDA cores.



However, at this point I won't be commenting in this thread other than to interject factual information simply because we keep going around the same circle:

Step 1: "I hear" or "rumor has it" or "supposedly" bandied around like people have a clue what is going on. They don't.

Step 2: Laughter by people who bought ATI cards and are desperately hoping they made the right decision and that prices won't go south if there is a strong GF100 launch

Step 3: Said ATI buyers or posters who want to start some controversy invariably quoting Charlie's work

Step 4: Someone actually posting some A+ info that has a firm foundation in reality.

Step 5: Repeat Step 2


Rinse and repeat.

[Dimitriman]

Those are numbers from bbs.expreview.com ...posted in table before, not measured ...only copied

PS. Those numbers are legit, but made with pre-production sample with lower clocks!

This card is great for orientation view of performance, because clocks are lower then retail piece. In logical view, when card beat on lower clocks 5780, will not be in final revision more better?

I can actually see this. The test cards everyone been seeing is A2. Regardless of how things went, A3 should be better than A2 in clocks even if just barely. A1 was supposedly 500mhz, A2 was 600mhz according to charlie. A3 is unknown at this point. Charlie is saying things didn't get better but if rumors are both given equal weight, about clocks flying around, the A2 is already at speeds of 625-650. Even with a garbage respin a3, we should be seeing clocks very close if not at 700mhz and shaderclocks into the 1400 range.

A2 is already at 625 or more. I can't see them getting anything less than 650 for a3

Man.. some people are joining a hype wagon going vertical with no seat belts and no brakes... Will be painful on 26th.


When was the last time that several leaked benchmarks 2 weeks short of launch date were completely wrong?
Drivers will improve performance, but claiming this is A2 and A3 will be pwning in rage makes me wonder if people think that from going from A2 to A3 takes as long as changing clothes.
A2 has been been discarded for months now and A3 is THE final revision, any sample that you see working and benched is A3. Just a warning before you get all your hopes too high for 470 doing 20% better than 5870.

My assessment: With tuned drivers, they will be close to identical perf. and you will be able to decide between oranges and oranges based solely on price. Not more, not less.

[Mungri]

The 5870 cooler seems to be bigger and it also features 4 heatpipes:



They are not in direct contact with the GPU though, and that might make a difference...

I were talking about the special version ones, not the stock one. It is exactly the same as the black one posted above, I cant find it, but will look for it.

Here it is, the 2GB Asus one:

I got some new pics of Asus ROG Matrix 5870: PCB and cooling solution:



12 phase PWM (8 for ultra-low leakage Cypress GPU, 2 for IMC and 2 for GDDR5)



2GB of memory, total 16 chips



5 heatpipes and improved fin-area



Heatpipes directly in contact with Cypress GPU



Thicker fan (21,5 mm)

Source & pics of AMD's 5870 reference cooling: Muropaketti.com

Notice the similarity?

http://www.donanimhaber.com/GeForce_...andi-18522.htm

Oh, the black one has 4 heatpipes, the Asus has 5 .... But they look so much similar still :x. Could be photoshopped.