NVIDIA GF100 & Friends speculation

[mczak]

The chap here says his GTX480 ES sample board got all the 512 ALUs enabled for the bench.

600Mhz only though. He says that should be about equal to final card, though I'd say final (he claims 675Mhz, I thought everyone agreed it's 700Mhz) should be like 5% faster (7% less alus/tmus, 17% higher clock). Nonetheless I'm not impressed (if those are legit).

 

[Razor1]

Where does it say it has a marginal lead in Dirt 2? For all we know it could be slower.
That said, I don't think you really can extend the "this game prefers AMD over NVIDIA" theme necessarily to GF100. There's probably way more to this, but 2 obvious reasons why a game would run better in theory on radeons compared to geforces relative to other games are 1) the shader code is not too "serially scalar" and 2) the alu:tex ratio is rather high. But GF100 has vastly increased alu:tex ratio hence 2) should no longer be true.

Do people even read the linked articles? The reason the number looks good in the Nvidia material is because it was running the DX9 path. AKA, it is none comparable but nvidia decided to compare it anyways.

Yes I did, and Dirt 2 isn't even part of the nV sales or marketing slides, so what do you think is up with the article after that? Where did nV run these tests must be for a semi-accurate test only.

 

[nv1911xxx]

GTX 480 vs HD 5870 vs GTX 285 SLI
http://img696.imageshack.us/img696/5613/gtx4803dmarkvantage.png
http://pic.xfastest.com/z/NVIDIA/Fermi/GTX_4X0.jpg

Tom's Hardware Benchmark Results: 3DMark Vantage
http://www.tomshardware.com/reviews/radeon-hd-5870,2422-11.html
Processor: Intel Core i7-975 Extreme Overclocked to 4 GHz
Motherboard: Asus Rampage II Extreme
Memory: Corsair Dominator 6GB (3 x 2GB) DDR3-1600
Hard Drive: Intel SSDSA2MH160G2C1 160 GB SATA 3 Gb/s
Graphics Cards: ATI Radeon HD 5870 1GB
http://img440.imageshack.us/img440/5722/3dmarkvantageoverall.png
http://img440.imageshack.us/img440/5722/3dmarkvantageoverall.png

 

[sham63]

Oh boy.

Would somebody refresh me on which bias each hardware site supposedly has?


I doubt that. Most people are preferring to set up some crazy multidisplay rig at home instead of having Physx? I'm not sure "most people" care about either, actually.

I agree with this. For Eyefinity the cost of two more monitors, and the amount of desk space required, will keep a lot of people from using it. Also, it may require the purchase of a new computer desk. As for PhysX, it just does not grab me in any way. Besides it would probably kill my GTS250.

Jim

 

[rpg.314]

I don't think their hand was forced, I think it was in the plan all along to do a shrink and X2. They did it with G71, G92 and GT200 so why not Fermi too?

Dunno about the G71. But IMHO 9800gx2 and gtx295 looked like a gain-hal0-back parts to me rather than something that would be on roadmaps for a long time.

 

[eastmen]

I agree with this. For Eyefinity the cost of two more monitors, and the amount of desk space required, will keep a lot of people from using it. Also, it may require the purchase of a new computer desk. As for PhysX, it just does not grab me in any way. Besides it would probably kill my GTS250.

Jim

I got to use eyeinfinty , it was only on 19 inch panels (But we had acess to tons of them) and I will never go back.

I'm trying to figure out the best way to go with eyeinfinty. I have a 24 and 22. I can get another 22 and run the 24 in the middle and the 22s on the side

http://i644.photobucket.com/albums/uu165/Darthdingo/IMG_0449.jpg?t=1260996737

I'm not sure how much i like physx . IT seems to kill performance and really requires a second card. The features are nice (i've seen batman physx ) but i'm not convinced you can't do that on quad core cpus and the new 6 core cpus that amd / intel are releasing

3D is another thing i'm not convinced of. It cuts performance in half and needs expensive monitors.

3D eyeinfinty with physx is something i'm to affraid to even concinve of how it would perform on todays hardware !!!!

 

[rpg.314]

The chap here says his GTX480 ES sample board got all the 512 ALUs enabled for the bench.

If there is any truth to that then 1GHz clocks and another gig RAM is all 5870 needs to toast 480's ass.

 

[rpg.314]

Semiaccurate finally approves the recent specs , GTX 480 to run at 700/1400MHz/3696MHz, consumes 250 W and achieves 87 C at load (98c @FurMark).

oh and Nvidia obviously tampers with Dirt 2 benchmark !!

http://www.semiaccurate.com/2010/03/23/nvidia-bends-definition-honesty-gtx480-benches/

Is this the beginning of another round of nv30 like benchmark shenanigans?

I sincerely hope it doesn't come to that.

 

[Squilliam]

I doubt the statements that there won't be a B1 spin for Nvidia.

Not because I have any information for or against, but because Nvidia has every incentive to play down any rumours that they will be making a revision to the process in short order.

The expectation I am seeing is that there will be a GTX280 -> 285 style improvement for their Fermi range of cards as theres a perception that the Fermi cards are broken at some funamental level. This is bad for Nvidia because its something which will make people wait on their purchases and lower the initial demand for the GPUs they are releasing now.

Whether its true or not that the B1 spin exists or doesn't exist, if the general consensus is to wait when applicable then people will remain on the market for a GPU for longer which gives ATI more time to sway that individual with special pricing etc and it means if the eventual update does come in a couple of months they may simply wait for the next generation from ATI instead and in addition to this lower demand would put more pricing pressure on their GPUs.

So because they have an incentive to downplay either way you can't trust a statement like that unless its backed up with some evidence.

 

[DavidGraham]

Yup, the T-unit is expanded to 40 bits to handle single-cycle INT32 op's.

Thanks , got it..

No, because the Fermi ALUs can't accept 40-bit wide data - they can accept *either* 32-bit FP or 32-bit int. There's no benefit to supporting a strange FP40 format, since it's not standard and no one could use it, neither under DX11 or OpenCL (or even CUDA).

The Fermi ALUs can accept 32-bit wide datatypes, either 32-bit integers or 32-bit FP. The details of how they've been implemented internally are likely much more complicated than just saying they have a "binary capacity of 40 bits", given that the ALUs are ganged up to operate on larger data types (like FP64).

That was insightful .. thanks ..

The chap here says his GTX480 ES sample board got all the 512 ALUs enabled for the bench.

At xtremesystem they say :

Also worth noting, this guy claims his Engineering Sample card had 512 shaders @ 600mhz, 1200mhz shader, and ONLY 2800 (GDDR5) RAM.

He also said the newer drivers added a couple % performance...

So I wouldn't trust these "leaks" ..

 

[rpg.314]

Also worth noting, this guy claims his Engineering Sample card had 512 shaders @ 600mhz, 1200mhz shader, and ONLY 2800 (GDDR5) RAM.

He also said the newer drivers added a couple % performance...

That's sounds much more reasonable. Let's hope 480 will be able to pull ahead of 5870 a lot more than this.

 

[DavidGraham]

Also with regards to this image :

http://i42.tinypic.com/9vgvt4.jpg [PC Games Hardware: Crysis Warhead bench]


It is fake , look at the catalyst(or catalytic ) driver version .. here is an old legitimate one ..:

http://www.hardwarecanucks.com/foru...rafikkarte-benchmark-crysis-warhead-3-png.png

 

[onethreehill]

NVIDIA GTX480 Disassembly Guide

http://i.haymarket.net.au/Galleries/20100324113521_IMG_0191 copy.jpg
The front of the GTX480 consists of (from right to left) a squirrel-cage fan, glossy black plastic shroud, large radiating surface and finally the expansion slot bracket.

http://i.haymarket.net.au/Galleries/20100324113523_IMG_0197 copy.jpg
Once you've pulled that off, the GTX480 is a little clearer. The aluminium frame that cools the memory chips can be seen at the bottom level, with the main heatsink and fan sitting just above.

http://i.haymarket.net.au/Galleries/20100324113626_IMG_0219 copy.jpg
With the heatsink out of the way, we can see where the core is clearly. The grooves in the aluminium frame provide plenty of surface area for heat to dissipate.

http://cdn.i.haymarket.net.au/Utils.../20100324113530_IMG_0204+copy.jpg&h=450&w=665
With everything removed, you should have three large pieces of kit; one main heatsink, one aluminium frame adorned with thermal pads, and one PCB.

NVIDIA GTX480 Disassembly Guide > Gallery > Atomic

 

[seahawk]

If there is any truth to that then 1GHz clocks and another gig RAM is all 5870 needs to toast 480's ass.

That is already a big fail for NV. With those results it is hard to come up with just one reason to buy a 4X0.

 

[annihilator]

That's sounds much more reasonable. Let's hope 480 will be able to pull ahead of 5870 a lot more than this.

I doubt 480sp vs 512sp would make a lot difference, but the clocks (600 core vs 700, and 700 mem vs 900) sure would make a big difference.

BTW whoever that guy is I'm 100% certain that he was the source for Charlie's "5% perf with 600mhz" remarks.

 

[thop]

With those results it is hard to come up with just one reason to buy a 4X0.

Don't underestimate brand loyalism, to avoid a different word.

 

[rpg.314]

That is already a big fail for NV. With those results it is hard to come up with just one reason to buy a 4X0.

Let's wait for the official reviews, shall we?

 

[ChronoReverse]

@onethreehill

Thanks for those pics. The GTX480's construction is quite a step up =O

http://www.atomicmpc.com.au/Gallery/170368,nvidia-gtx480-disassembly-guide.aspx/22 <== check out the fan

 

[AlNom]

Now where's the pic showing a removed heat spreader?

 

[thop]

The BFB0712HF is a custom fan as only the BFB0712HD is listed on their website. It draws 0.3A and is rated at 4000rpm and 42dBa. And the fan used by GTX480 draws 1.8A current