GTX480 only 5% faster than a 5870?

[Apocalypse23]

Alright I'm not sure how many Anandtechers have read HardwareCanucks's online review and benchmark of the GF100 (Fermi), but they conducted some benchmarks in January on Nvidia's March shipping cards (possible Gtx480)...I just found this link and the results are mind blowing: In Far Cry 2 Fermi dominates by about 20 more fps than a 5870, in Dark Void the difference is roughly 30 fps....

I have a feeling assuming this card was their high end card (GTX480), then it beats the 5870 and either equals or is about 9-5% slower than a 5970....

HardwareCanuck Fermi benchmark preview January 17th

I don't know if Nvidia will blow it, there's a possibility they may have built a single GPU card that outperforms the 5870, but the 5970 still beats everything overall....So then you have SLI competition and nvidia may win that battle? Not sure then again, one thing I am positive about is that the 5870 might not cut it....

Hardwarecanucks reviewers witnessed Nvidia's Fermi bench and also their hardware eye to eye and this is what they reported:
Nvidia fermi test setup:
"PLEASE READ:

As you look at the charts below, you will notice that we added in “simulated” HD 5870 1GB results and there is a story behind this. NVIDIA was very forward with us and revealed the specifications of the test system they were running: a stock Intel i7 960, 6GB of 1600Mhz memory and an ASUS Rampage II Extreme motherboard running Windows 7. While we weren’t able to copy the exact same system, we did our best to replicate it by using the exact same CPU and OS while estimating the memory latencies and using a Gigabyte X58-UD5 motherboard. The result? Our stock GTX 285 performed within 2% of the results NVIDIA showed us so we are confident in the accuracy of our results. "

here is a 5870/5970 Far Cry 2 performance similar to Fermi Benchmark testbed I found off the net (Fudzilla 5970 review):
Test system:
Motherboard: MSI P45D3 Platinum ( Provided by: MSI );
Processor: Intel Core 2 QX9770 Extreme edition at 3.6GHz ( Provided by: Intel );
Memory: Corsair Dominator 12800 7-7-7-24 ( Provided by: Corsair);
HDD: WD VelociRaptor 300G 10,000RPM ( Provided by: SmoothCreation );
Driver: ForceWare 195.50, Catalyst CCC 9.11 Vista 32 SP1

Here is anandtech's 2560x1600 4xAA ULTRA Far Cry 2 bench, compare it to the Fermi bench up top...
AT Testbed:
CPU: Intel Core i7-920 @ 3.33GHz
Motherboard: Intel DX58SO (Intel X58)
Memory: Patriot Viper DDR3-1333 3 x 2GB (7-7-7-20) -- similar to Fermi testbed

All in all it looks as though the 5970 beats Fermi but the 5870 gets clobbered (10-20fps slower) by Fermi's top line card....Unless they tested a GTX470.

 

[Unkle_Tar]

Simulated benchmarks, haha what next.

 

[zerocool84]

Alright I'm not sure how many Anandtechers have read HardwareCanucks's online review and benchmark of the GF100 (Fermi), but they conducted some benchmarks in January on Nvidia's March shipping cards (possible Gtx480)...I just found this link and the results are mind blowing: In Far Cry 2 Fermi dominates by about 20 more fps than a 5870, in Dark Void the difference is roughly 30 fps....

I have a feeling assuming this card was their high end card (GTX480), then it beats the 5870 and either equals or is about 5% slower than a 5970....

I don't know if Nvidia will blow it, there's a possibility they may have built a single GPU card that outperforms the 5870, but the 5970 still beats everything overall....So then you have SLI competition and nvidia may win that battle? Not sure then again, one thing I am positive about is that the 5870 might not cut it....

This has already been discussed to death in many other threads.

 

[Silverforce11]

Apparently, the double performance in unigine heaven is not reflected in current dx11 games.. as charlie claims.

 

[Apocalypse23]

I updated my post...see above

 

[Piotrsama]

Alright I'm not sure how many Anandtechers have read HardwareCanucks's online review and benchmark of the GF100 (Fermi)

Cherry picked bench FTL, as many have said here.... we want fair reviews done by non-bias sites. (not judging that particular site, just saying...)

Apparently, the double performance in unigine heaven is not reflected in current dx11 games.. as charlie claims.

Especially if it's also true that there were graphical glitches all over the place on fermi runs.

 

[Apocalypse23]

Cherry picked bench FTL, as many have said here.... we want fair reviews done by non-bias sites. (not judging that particular site, just saying...)



Especially if it's also true that there were graphical glitches all over the place on fermi runs.

I agree totally, but Far Cry 2 is a game that deserves a benchmark..it gives a good indication of a card's performance...and should hint at other games also..I think the bench is fair...

 

[3DVagabond]

It's funny how everyone that says Charlie is full of it and biased, etc.... don't have a problem in the world believing that Fermi beats Cyprus by 100% in heavy tessellation, just as Charlie says. Everything else he says though, is a lie, or he's stupid, or biased. You can't have it both ways folks. You can't say he's right on the stuff you want to hear but wrong on everything else.

Besides, have you looked at the reported releases? Nothing at all until 04/31. Then we are probably looking at pro cards, not Geforce. Nothing, nothing at all in the mid or lower ranges. Just 480/470, eventually. If you are going to wait for GTX480 to become available at retail, where you can actually get your hands on one, you're going to see ATI's refresh of the 5000 series first. You might see HD6870 first.

http://www.nordichardware.com/en/co...geforce-gtx-480-in-stock-in-may-at-best-.html

 

[dug777]

All in all it looks as though the 5970 beats Fermi but the 5870 gets clobbered (10-20fps slower) by Fermi's top line card....Unless they tested a GTX470.

Comparing the FC 25x16 benchmarks between AT and HWC, the GTX 285 looks comparable (41 vs 37), but the 5870 is way way off...44 vs 58 for the AT one (which puts it pretty close to the GF100 sample at 62 in the HWC benchmarks).

Who do I trust more? AT by a country mile.

 

[MarcVenice]

Charlie is getting more specific, he now mentioned clockspeeds. Write it down, and soon we'll know wether he's full of it, or if he actually has some inside sources.

 

[Apocalypse23]

Comparing the FC 25x16 benchmarks between AT and HWC, the GTX 285 looks comparable (41 vs 37), but the 5870 is way way off...44 vs 58 for the AT one (which puts it pretty close to the GF100 sample at 62 in the HWC benchmarks).

Who do I trust more? AT by a country mile.

noted,

I don't buy the 5870 simulated bench marks by HardwareCanucks...but if you simply look at the fermi scores and compare them to the GTX295 scores on all the charts, they are awfully close to each other...

in other words...Fermi = single GPU GTX295 + tessellation/phsyx...i think Charlie is very precise...

 

[Soleron]

He does seem to have contradicted himself by previously saying the top bin is 448 shaders yet the top card is 512.

Although, maybe the top bin is 448 and they're going to launch a 512 anyway... with consequently (almost) zero cards available to achieve it. They could launch a card with 2000MHz shaders, but if the top bin is only 1250MHz then you'd never see a retail card.

 

[Hauk]

Thanks Apocalypse for putting that info together. Simulated benchmarks are all there is at this point so it's information worthy of study. Hopefully we'll see some real performance numbers in the next couple weeks.

As for Charlie, he's in overdrive. The data's too vague for moles with their hands on actual cards IMO. He's throwing crap against the wall to see what sticks. He's likely hitting close when he talks about clocks and initial supply. We know of TMSC yield issues, and hear of multiple respins.

My guess is it'll be a hard launch next month with some low number, like 10,000 units worldwide, then in short supply for months following. ATI's stocking nicely now, even 5850s. They're over their "ramping up" hurdle. nV will likely have an even greater challenge. But I'm betting there's a launch with limited availability next month. What it can do, we'll have to see..

 

[TemjinGold]

Well now hes just bullshitting, just the other day he said:



Now he says:



Really consistent lol

Not to defend him but that first quote says initial top bin. If nVidia launches the 448 only initially and then later launches the full part, Charlie would not be inconsistent.

 

[evolucion8]

Nvidia would have to screw up really bad for the GTX 480 to be only 5% faster than the 5870. The 5870 is what, 30% faster than the GTX 285? The GTX 480 has over twice the SPs and a slightly slower core clock, so unless Nvidia managed to drop the SP performance/clock by about 50% it should be much faster :hmm:

Wrong, a single HD 5870 is a hair slower than a GTX 295 which is are two GTX 275 GPU's which each GPU is a hair slower than a GTX 285.

Alright I'm not sure how many Anandtechers have read HardwareCanucks's online review and benchmark of the GF100 (Fermi), but they conducted some benchmarks in January on Nvidia's March shipping cards (possible Gtx480)...I just found this link and the results are mind blowing: In Far Cry 2 Fermi dominates by about 20 more fps than a 5870, in Dark Void the difference is roughly 30 fps....

I have a feeling assuming this card was their high end card (GTX480), then it beats the 5870 and either equals or is about 5% slower than a 5970....[/QUOTE]

That's not a reliable benchmark, a simulated HD 5870 doesn't come near of the results of the Anandtech's review which are far more consistent. It even puts the Fermi score too close of the GTX 295.

 

[bryanW1995]

Cherry picked bench FTL, as many have said here.... we want fair reviews done by non-bias sites. (not judging that particular site, just saying...)



Especially if it's also true that there were graphical glitches all over the place on fermi runs.

that is certainly cherry-picked, but I've always found hardware canucks to be extremely unbiased. They're usually one of my first reads when new reviews come out.

 

[tviceman]

Charlie's moles probably had the GTX470 and mistakenly thought it was the GTX480. :/

 

[dzoner]

He does seem to have contradicted himself by previously saying the top bin is 448 shaders yet the top card is 512.

Although, maybe the top bin is 448 and they're going to launch a 512 anyway... with consequently (almost) zero cards available to achieve it. They could launch a card with 2000MHz shaders, but if the top bin is only 1250MHz then you'd never see a retail card.

Following the initial binning process of the A3 wafers coming off the line 448 chips may have been considered the top bin if viable 512 chips were so scarce they would never be commercially viable, say ~100 total of the entire wafer run.

But if extensive testing of the 448 chips revealed it could not be optimised to beat the 5870, even by a little bit, that handful of 512 chips had to be brought into play, commercially viable or not.

Nvidia absolutely, postively, CANNOT release a top end single GPU Fermi card that underperforms the 5870.

If so there will only ever exist a handful of uber-optimised Unobtanium GTX480 cards released to selected review sites and none released into the retail channel.

 

[dzoner]

that is certainly cherry-picked, but I've always found hardware canucks to be extremely unbiased. They're usually one of my first reads when new reviews come out.

Taking Nvidia's 'benchmarks' at face value is where things went awry.

But then that was an earlier, more innocent, age.

Be a different story if Nvidia released them in the present PR enviroment .

 

[akugami]

I don't find Hardware Canucks biased but at the same time, and as many others have pointed out, these are simulated benchmarks.

We also have zero information on what the exact specs of the Fermi card are. So even if we knew everything else and Hardware Canucks was able to replicate the exact system down to the timings of the RAM chips, CPU, exact HD, etc., it still tells us nothing because we the star of the show is hidden behind a veil of secrecy. Basically, the benchmark is useless. Not to mention we should not trust any benchmarks from either ATI or nVidia at face value.

 

[Apocalypse23]

Perhaps everyone is expecting too much from Nvidia? I don't think HarwareCanucks would simply post that information up...there's something we can conclude from nvidia's benchmark alone, forget the 5870 bench they conducted...compare the ultra numbers and it does come close to a GTX295, but maybe that's just it???

Remember back in the Geforce 9 end days when the 9800 GX2 was top of the line, it was a DUAL GPU...the next product nvidia produced was a single gpu GTX260 and GTX280, the 9800GX2 was equal to the Single GPU GTX280 back then...nvidia had a brand new architecture but similar performance....


The same is happening now, I already see it coming, the GTX480/470 are single Gpus with the GTX480 possibly beating the GTX295 or coming close to it. Nvidia won in SLI back with the Geforce 200 series, they will do the same now with GF100 chips..Hence there is also talk about mass production, so more likely they will only win in SLI and beat the 5970 hands down..so they are waiting for mass cards to come out around Q2 2010, it already makes sense to me...

We will know on Monday...

Just don't keep your expectations too high, I'm sure nvidia has a good lineup of cards, but just not enough power for a single card yet.

 

[Soleron]

Charlie has responded to the 448/512 thing.

http://www.semiaccurate.com/forums/showpost.php?p=28431&postcount=234

It's as I thought; the 480 is a 512 part but the top bin is still 448. Hence, no 480s at all in the initial allocation to AIBs.

Putting it all together, the only Fermi you'll be able to buy performs less than the 5870.

 

[Udgnim]

Apparently, the double performance in unigine heaven is not reflected in current dx11 games.. as charlie claims.

I think the reason why is that the DX11 games that have been released don't push tessellation very hard (if at all).

 

[SHAQ]

OMG...the 480 is only the fastest single card by 5%. Is he already hedging the broken and unfixable line?

70C at 2D clocks. LOL

 

[bryanW1995]

I don't find Hardware Canucks biased but at the same time, and as many others have pointed out, these are simulated benchmarks.

We also have zero information on what the exact specs of the Fermi card are. So even if we knew everything else and Hardware Canucks was able to replicate the exact system down to the timings of the RAM chips, CPU, exact HD, etc., it still tells us nothing because we the star of the show is hidden behind a veil of secrecy. Basically, the benchmark is useless. Not to mention we should not trust any benchmarks from either ATI or nVidia at face value.

didn't the original phenom outperform c2d by up to 40% in "simulated benchmarks"?