The GT300/Fermi Thread - Part 2!

[SKYMTL]

look at what? those are two different ways to illustrate a gpu, of course it looks very different

Tue but the way information flows through the GF100 is quite different from the GT200.

Also, at this point in time everything will be "evolutionary" versus "revolutionary" since the unified shader-based architecture will be with us for some time.

[damha]

Of course everything between sky and earth is a kind evolution or tweak of that , but the "Dedicated L1 Load / Store cache" is closer to a revolutionary step in this round. This gets Fermi much closer to become a supercomputer, because it is the part that support generic C/C++ program (very much like a x86 CPU would do).

Isn't that what ATI has had in their GPUs for the last two generations??

[***Deimos***]

Oh ya one more thing i think that both GTX 480 and GTX 470 uses analog VRM setup and this is a very curious thing to do. The reason why i said curious is that when 5870 was launched with digital VRM i asked the advantages that it will have over analog system and i was told that digital VRM works good handling lower voltage than higher ones but analog VRM works better in handling higher voltage with stability.

no. both digital VRM and traditional step-down buck converter, can output any voltage from 0 - 12V. For both digital, and analog component (ie cap, inductor, diode/MOSFET rectifier) selection determines max current, transient response (ie ripple current) etc.

Caps are high and block heatsinks. AMD digital VRM is better for heatsinks. Multiple-phase solutions are easier to link up. And can better manage switching... ie reduce voltage/clocks/current if overheating.

Its kind of like old school jumper/BIOS overlocking vs using setFSB in Windows - the later obviously easier and more convenient.

i think AMD has made enough profit and will be happy to drop 5870 to 300$ the week of nvidias release.

I've been wrong but I dont think so. TSMC 40nm is still far from "mainstream" and quite expensive. Only reason for AMD to drop down to $300, would be if Fermi launched at $300 - infinitesimally improbable.

[Sam_oslo]

Isn't that what ATI has had in their GPUs for the last two generations??

ATi has a huge potential for GPGPU usage too, but they needs to provide a "working" API for programmers to access the GPU in a common and easy way. It should support generic codings like C/C++ too, and that's where CUDA helps Fermi to get ahead of ATi.

[Dark-Energy]

Does anyone think dual GPU Fermi is possible at all (GTX470, and downed clock speeds like the 5970) without requiring a nuclear power plant and generating as much heat equivalent to the surface of the sun?

[zalbard]

Does anyone think dual GPU Fermi is possible at all (GTX470, and downed clock speeds like the 5970) without requiring a nuclear power plant and generating as much heat equivalent to the surface of the sun?

Definitely.

[trinibwoy]

Isn't that what ATI has had in their GPUs for the last two generations??

No Fermi is the first mass market GPU architecture to have:

#1: Parallel geometry setup
#2: Generalized, coherent read/write caching

Both are huge deals because of the engineering effort required and make a lot of things easier to do. Of course it doesn't mean squat if you just care about the fps that comes out the end.

[damha]

ATi has a huge potential for GPGPU usage too, but they needs to provide a "working" API for programmers to access the GPU in a common and easy way. It should support generic codings like C/C++ too, and that's where CUDA helps Fermi to get ahead of ATi.

My recommendation to ATI would be to adopt/adapt the CUDA api from nvidia and get on with it. Let's be honest, nvidia has the clout to raise a stink big enough for everyone to notice. Their connections run deep.

Unless ATI is planning an API release secretly. I really hope they aren't, waste of resources.

@triniboy: I'll check it out. This gives me something to research

[zalbard]

My recommendation to ATI would be to adopt/adapt the CUDA api from nvidia and get on with it. Let's be honest, nvidia has the clout to raise a stink big enough for everyone to notice. Their connections run deep.

They just need to help some devs with OpenCL, and advertise it some IMO. The potential is there.

[Sam_oslo]

My recommendation to ATI would be to adopt/adapt the CUDA api from nvidia and get on with it. Let's be honest, nvidia has the clout to raise a stink big enough for everyone to notice. Their connections run deep.

Unless ATI is planning an API release secretly. I really hope they aren't, waste of resources.

I believe ATi could adapt CUDA easily, if they want to.

But i agree with you, we need a common API for assessing a GPU (just like AMD and Intel CPU that csn run a sett og common instructions and programs).
Somebody has to provide a common platform to take this further to the next step where everybody can have a personal supercomputer , but these guys (nVidia and ATi) are too busy fighting each other, and Intel will actually see both of the dead and defeated in this area, Because a great GPGPU platform can threaten the Intel dominance in those very suspensive supercomputer marked.

[Levish]

I've been wrong but I dont think so. TSMC 40nm is still far from "mainstream" and quite expensive. Only reason for AMD to drop down to $300, would be if Fermi launched at $300 - infinitesimally improbable.

There are two other reasons to drop the price on the 5870 to $300

- If the 480GTX comes in at close to its current price/performance to deny any signficant marketshare. If the 480GTX comes in at 5-10% higher performance than the 5870 would the average upgrader still choose it if the 5870 was 25% cheaper (probably not)?

- If AMD/ATI counters with a 5875 (or whatever 5870 rev2) at the same price point of the current card (to retain "performance crown"), the "old" 5870 wouldn't sell at all at the current price.

I'm inclined to say that its very likely AMD/ATI would price cut, I'd probably bank on it being closer to $50 though (I would be quite happy if it was $100).

[RaV[666]]

Ati definietely has price flexibility right now.Dont forget that ATI prices have risen above launch msrp.They are cheaper to produce, they have them for months, and have had better yields from the start (6 months ago).
Thing is, if they will not feel threaten by fermi, they probably wont :/.MSRP for 5850 on launch was 259$ ,it stands at 300+ now.

[Sam_oslo]

No Fermi is the first mass market GPU architecture to have:

#1: Parallel geometry setup
#2: Generalized, coherent read/write caching

Both are huge deals because of the engineering effort required and make a lot of things easier to do. Of course it doesn't mean squat if you just care about the fps that comes out the end.

Yep, exactly. These are very big deals in GPU evolution (if not revelation) right now. All GPUs have a HUGE amount of Gflps (compared to a CPU), but how do you control/program the beast to do something more useful than just gaming?

That dedicated L1 cache plays a big role for making the the life much easier to program/control the beast. This makes it possible to have a unified read/write cache, which allows program correctness and is a key feature to support generic C/C++ programs.

[MrMojoZ]

My recommendation to ATI would be to adopt/adapt the CUDA api from nvidia and get on with it. Let's be honest, nvidia has the clout to raise a stink big enough for everyone to notice. Their connections run deep.

Unless ATI is planning an API release secretly. I really hope they aren't, waste of resources.

@triniboy: I'll check it out. This gives me something to research

Horrible idea, CUDA is the next Glide. It will be replaced by something not ruled by a single hardware vendor.

[Iconyu]

There are two other reasons to drop the price on the 5870 to $300

- If the 480GTX comes in at close to its current price/performance to deny any signficant marketshare. If the 480GTX comes in at 5-10% higher performance than the 5870 would the average upgrader still choose it if the 5870 was 25% cheaper (probably not)?

- If AMD/ATI counters with a 5875 (or whatever 5870 rev2) at the same price point of the current card (to retain "performance crown"), the "old" 5870 wouldn't sell at all at the current price.

I'm inclined to say that its very likely AMD/ATI would price cut, I'd probably bank on it being closer to $50 though (I would be quite happy if it was $100).

AMD won't cut the price unless nvidia can meet the demand, if there are only 8,000 480's the ticket price won't mean anything. AMD are already releasing a card with a price of $1,000, it will probably sell.

[jaredpace]

Fermi is a complete arch redesign, like g80 & r350. Focus on cache, compute, & gpgpu programmability. cypress is rv770 with alu's & rops doubled and scheduler & setup redesigned to effectively use new resources.

[Soultaker52]

http://www.nvnews.net/vbulletin/showthread.php?t=148597

http://bbs.expreview.com/attachments...dfef0d34ea.jpg

http://bbs.expreview.com/attachments...be60baceff.jpg

Is it just me or do the roof tiles on the GTX470 shot look odd? I just ran the bench and they looked much sharper and defined on my 5870.
985/1250 gave me 29.8fps, by the way.

[Jowy Atreides]

Is it just me or do the roof tiles on the GTX470 shot look odd? I just ran the bench and they looked much sharper and defined on my 5870.
985/1250 gave me 29.8fps, by the way.

Yes, they look nerfed.

[damha]

Is it just me or do the roof tiles on the GTX470 shot look odd? I just ran the bench and they looked much sharper and defined on my 5870.
985/1250 gave me 29.8fps, by the way.

More than likely running the "quality" texture setting, not "high quality". Which is not all that surprising as MOST review sites keep that setting unchanged, resulting in worse IQ for nv cards but higher performance.

[Soultaker52]

Yes, they look nerfed.

Actually, I just went through it on free roaming camera to make sure, and the tiles look about the same. I guess in that area they just look like

[Chumbucket843]

Fermi is a complete arch redesign, like g80 & r350. Focus on cache, compute, & gpgpu programmability. cypress is rv770 with alu's & rops doubled and scheduler & setup redesigned to effectively use new resources.

i wouldnt call it a complete overhaul. but their are some big changes.

[zerazax]

It's not a G71->G80 jump, but it's not a G92->GT200 one either

[Yvese]

Not sure if this has been posted before. Card is supposedly the GTX470

[zerazax]

Yeah, it's been posted

[insurgent]

It's been posted, but I just checked anandtech's reviews and the 5870 gets 38.5 fps on average I presume, in warhead @19x12 4xAA Enthusiast... granted anandtech's setup has a faster i7 (+130MHz?) but I don't think that's enough for a ~25% increase in frames. I was trying to look for a review with 8x AA but found none atm.

edit:
of course it depends on the part of the game that was ran, but I was thinking they used the built-in benchmarking tool