Silly question regarding GF100/Fermi

[Yossarian22]

I've been to Nvidia's website and I heard about the advantages of doing floating point arithmetic, GigaThreads, advantages in linear algebra, DataCache (assignable 64k L1 for each set or something?), ECC memory, sorting and searching algorithms (bubble, quick, binary...), etc...

My question is: will it render graphics? Like, games?

 

[kuyaglen]

My question is: will it render graphics? Like, games?

Of course not.

 

[PlatinumPC]

bait anyone?

 

[Yossarian22]

bait anyone?

Well, they've mentioned ray tracing and workstation machines, but not game performance. Will this general purpose CPU be suited towards the workstation or is that Tesla or are they both the same thing?

 

[Tamlin_WSGF]

I've been to Nvidia's website and I heard about the advantages of doing floating point arithmetic, GigaThreads, advantages in linear algebra, DataCache (assignable 64k L1 for each set or something?), ECC memory, sorting and searching algorithms (bubble, quick, binary...), etc...

My question is: will it render graphics? Like, games?

The Fermi announced was about the HPC capabilities of the card, aimed towards users that earlier have used Tesla. Of course it will render graphics. Thats what GPU does.

Specs for consumer cards have not been released yet, but considering that Nvidia have spend much money into gaming oriented software and hardware (like 3d vision and physx), it would be safe to assume that it will perform also for games.

 

[RevMen]

Let me tell you what bothers me about Nvidia and Fermi. They keep saying Fermi like "fur me." They named the thing after Enrico Fermi but they don't bother to say his name correctly. It's Italian and you say it "fair me." Maddening.

 

[kuyaglen]

It's Italian and you say it "fair me." Maddening.

Its America, we say it like how we want to.

 

[RevMen]

Its America, we say it like how we want to.

Fermi and his family emigrated to the US and he worked on the Manhattan Project. Fermilab (pronounced "fair me lab") is near Chicago. Fermi was about as American as you can get, and the correct way to say his name, in America or elsewhere, is "fair me."

 

[Direfox]

I've been to Nvidia's website and I heard about the advantages of doing floating point arithmetic, GigaThreads, advantages in linear algebra, DataCache (assignable 64k L1 for each set or something?), ECC memory, sorting and searching algorithms (bubble, quick, binary...), etc...

My question is: will it render graphics? Like, games?

Its not a stupid question, just not worded quite right. He is asking if this GPU core is going to be used as a gaming card or a workstation card.

 

[Elledan]

NVidia has always used the same core for workstation, gaming and dedicated CUDA/OpenCL cards. There is no indication they'll do anything else this time around.

Next question please.

 

[sirmonkey1985]

Its America, we say it like how we want to.

exactly.. thats why americans say aluminum as a-lum-in-um instead of ale-min-eum like it should be.. i mean hell we created american football because we didnt like the rules of european soccer.. its the american way to say "eff you to the world, we are america and we say/do what we want"..

 

[BioSehnsucht]

Aluminum's aluminum. Obviously, if someone was talking about this other zany material aluminium they'd spell it that way. Hell, firefox is telling me I'm spelling it wrong even!

 

[needmorecarnitine]

exactly.. thats why americans say aluminum as a-lum-in-um instead of ale-min-eum like it should be...

are you high, drunk or both?

\ə-ˈlü-mə-nəm\

 

[Atech]

Well, they've mentioned ray tracing and workstation machines, but not game performance. Will this general purpose CPU be suited towards the workstation or is that Tesla or are they both the same thing?

That is because those are the target-markets of the audience at NVIDIA's (GP)GPU conference.
If you had bothered to read the homepage, you would know this:
http://www.nvidia.com/object/gpu_tech_conf_agenda.html

It wasn't for gamers or gamedevelopers, it was for a different breed....hence the lack of gaming focus.
But since "Fermi" has over two times the ASIC for gaming than the GT200b...what do you think?
(The GTX285 has 240 streaming processors, "Fermi" has 512 CUDA cores)

 

[sirmonkey1985]

are you high, drunk or both?

\ə-ˈlü-mə-nəm\

lol havent been to england then i take it.. the way you put it there is how we say it in the US.. but its not how they pronounce it in england..


but back on topic..

 

[Ryan711]

lol havent been to england then i take it.. the way you put it there is how we say it in the US.. but its not how they pronounce it in england..


but back on topic..

then they're saying it wrong. Too lazy to disguish alumiNUM from all of the other iums.

 

[Camberwell]

then they're saying it wrong. Too lazy to disguish alumiNUM from all of the other iums.

Right, so a pronunciation that has been used in the English (not American) language from before Columbus even discovered America is both lazy and wrong....facepalm....

 

[MoparMadness]

That is because those are the target-markets of the audience at NVIDIA's (GP)GPU conference.
If you had bothered to read the homepage, you would know this:
http://www.nvidia.com/object/gpu_tech_conf_agenda.html

It wasn't for gamers or gamedevelopers, it was for a different breed....hence the lack of gaming focus.
But since "Fermi" has over two times the ASIC for gaming than the GT200b...what do you think?
(The GTX285 has 240 streaming processors, "Fermi" has 512 CUDA cores)

Oooh Yossarian22, sounds like you hit a nerve with someone! How dare you ask questions!

 

[RevMen]

Right, so a pronunciation that has been used in the English (not American) language from before Columbus even discovered America is both lazy and wrong....facepalm....

Folks, it's actually spelled differently in the UK and the US. It's not a matter of how it's pronounced.

And saying Fermi like furme isn't an American thing. The guy lived in America for a long time and Americans always pronounced his name correctly. It's just the people at Nvidia who somehow never heard the guy's name spoken. It makes them sound ignorant, like someone in the marketing department just googled "famous physicists" and that was all the research they needed to do.

 

[Kowan]

Silly question regarding GF100/Fermi
GT300 is codenamed Fermi
Quote:

I do think nVidia are shifting more focus to workstation and HPC, might even see a decline in TWIMTBP as the engineers are reassigned.

I am sorry but thats not gonna happen. TWIMP primary core group is based out of Russia. And most of its software engineers are based out of Russia.. And they work with developers there. They are not hardware engineers. And this is their specialty. I know this because I was to be the American contact for Nvidia's TWIMP had I not turned down the job.

These people are not gonna ship off to Nvidia's American offices to work on different projects. Theres no reason for Nvidia right now to be cutting back. If anything TWIMP and its influence has been growing over the years.
Quote:

I do think nVidia are shifting more focus to workstation and HPC

I can tell you for a fact the Fermi has more gaming ASIC than the Gt200. Nvidia is not stupid. I dont care what anandtech says about what they think Nvidia's plans are for the future. Nvidia is not going to abandon the gaming market because they are trying to spread their horizons and use their GPU architecture to do more than just gaming
__________________
Nvidia User Group Member and SLIZONE administrator:

Quote:

(I notice you compare it GT200 and not RV870...

I compare it to the GT200 Solely because people are saying that Nvidia has invested very little into games with the Fermi. When the Fermi "Entire GPU" details become clear. People will see just how untrue that is. Like I said. The Fermi has more gaming ASIC than the GT200. ((And the improvement is larger than the GT200 to G80 was)). Also even if I wanted to compare it to the HD5870 we have two totally different architectures. And its impossible to compare them directly on a "Number to Number" basis of ASIC. With the Gt200 its quite a bit more simple.

All I said was that nVidia are going to go deeper into the workstation and HPC market; I speculated they might use some of their existing resources to do this (makes more sense than hiring and training new engineers or over-extending the existing teams, for example).

I have no problem with speculation that Nvidia is working their way into other markets. I am saying that the belief that Nvidia is moving away from gaming market is unfounded and based on the speculation of people who dont understand Nvidia's business model or don't understand why Nvidia announced its GPU Compute shader potential at the GPU Computing teleconference.

Originally Posted by acroig
Only reason why that could be is that gemers (read: us) wanted details that we don't have yet and nV talked about GPU Computing and not gaming. I hope more gaming related info will release this year.

I understand that, But the conference they had was about GPU Computing, CUDA, and OpenCL, So thats all they talked about. I would not expect Nvidia to release more of its architecture "Gaming GPU" details until closer to its actual launch.

But like I said. I think speculation that Nvidia is moving away from the gaming market is unfounded right now. And like I said. I can't specify further until that time. But the Fermi is not handicapped in the gaming department and is well expanded beyond previous gen in dedicated circuitry and feature support and things about how Nvidia handles tesselation will be fully understood by then.

I just hope you guys understand theres only so much I can actually talk about. Technically even saying that Nvidia's gaming ASIC has improved is me breaking NDA. But I grow frustrated with some of the posts I see here. And I do understand user frustration with wanting to know more. But I just wouldn't expect much more information till closer to launch.

 

[RamonGTP]

Let me tell you what bothers me about Nvidia and Fermi. They keep saying Fermi like "fur me." They named the thing after Enrico Fermi but they don't bother to say his name correctly. It's Italian and you say it "fair me." Maddening.

Who cares? I mean really, besides a few uptight people with nothing better to worry about, who really cares?

 

[Yossarian22]

Oooh Yoassarian22, sounds like you hit a nerve with someone! How dare you ask questions!

I guess so.

That is because those are the target-markets of the audience at NVIDIA's (GP)GPU conference.
If you had bothered to read the homepage, you would know this:
http://www.nvidia.com/object/gpu_tech_conf_agenda.html

I did read it back when Kyle and co. were going to the conferences and yesterday. I watched the videos and stuff, not a dicky bird about real time graphical rendering/game performance, which is why I was confused (still am).

It wasn't for gamers or gamedevelopers, it was for a different breed....hence the lack of gaming focus.

I know, that's what worries me and makes me confused. Until now, if a card wasn't for games it was for rendering graphics at a workstation machine in like Blender, 3D Studio Max, etc... not sure where they're trying to go with this one . Super computing?

But since "Fermi" has over two times the ASIC for gaming than the GT200b...what do you think?
(The GTX285 has 240 streaming processors, "Fermi" has 512 CUDA cores)

I think this card will be prohibitively expensive if they're trying to make it a super computer when all I want to do is play games (granted, it might be good if college/work bought these cards, I suppose). What exactly is a CUDA core? Is having 512 CUDA cores similar to my processor having 512 ALUs or 512 FPUs?

Also, do they plan on making a model that just isn't Fermi? Like, what if I don't want a "super computer GPU" or whatever, if I just want to play games and the like? Does that make sense? Like, I don't need those features, I just want something cheaper, my case to be cool, and good gaming performance. Does that make any sense at all?

 

[jeremyshaw]

CUDA core is just nvidia's marketing lingo for a shader processor.

They used to call them shader processors, after larrabee, just because... dunno. Intel called them 'cores'?

They are just shaders. They compute.

Then they send the data to output.

That's about it. Be it graphics, or otherwise.

 

[nobody_here]

I guess so.


I did read it back when Kyle and co. were going to the conferences and yesterday. I watched the videos and stuff, not a dicky bird about real time graphical rendering/game performance, which is why I was confused (still am).



I know, that's what worries me and makes me confused. Until now, if a card wasn't for games it was for rendering graphics at a workstation machine in like Blender, 3D Studio Max, etc... not sure where they're trying to go with this one . Super computing?



I think this card will be prohibitively expensive if they're trying to make it a super computer when all I want to do is play games (granted, it might be good if college/work bought these cards, I suppose). What exactly is a CUDA core? Is having 512 CUDA cores similar to my processor having 512 ALUs or 512 FPUs?

Also, do they plan on making a model that just isn't Fermi? Like, what if I don't want a "super computer GPU" or whatever, if I just want to play games and the like? Does that make sense? Like, I don't need those features, I just want something cheaper, my case to be cool, and good gaming performance. Does that make any sense at all?

i think people are reading too much into all the marketing lingo.....its going to be another progression in graphics processing, they are going to call different aspects of it different things in order to appear to cater to a specific market, you could take the older cards and apply some of the same lingo to them and suddenly they dont sound like a traditional graphics card anymore...

i wouldnt worry about it honestly

 

[Yossarian22]

CUDA core is just nvidia's marketing lingo for a shader processor.

Oh. Oh, okay.

i think people are reading too much into all the marketing lingo.....its going to be another progression in graphics processing, they are going to call different aspects of it different things in order to appear to cater to a specific market, you could take the older cards and apply some of the same lingo to them and suddenly they dont sound like a traditional graphics card anymore...

i wouldnt worry about it honestly

I do worry about it. There are two things that worry me:

1. If they tout this device to be specialized in doing 'super computer' style tasks (floating point arithmetic), what does that indicate about it's graphical rendering or game performance? Have they nothing to brag about that? Potentially by catering to a field never really tapped into before and ignoring the gaming segment, it is alienating. It is a super computer and can play games? A jack of all trades is a master of none. Do one thing, do it well.
2. Since this is a super computer you can stick into your PCIe slot, will there be a cheaper version that caters to the "mainstream" market with reduced or removed functionality? What if I don't need to calculate millions of digits of Pi or do any of the scientific stuff, I just want to kill pixels in games. I imagine this card will be prohibitively expensive, so will there be a slim version? (e.g: 5870 is to 5770 as Fermi is to what?) Like, can they just forgo all the CUDA and scientific calculation stuff and give me a rock solid video card that runs games at light speed?

I don't know, but by the way they are marketing this card, they make it sound like they've lost focus of their audience they have had for the past decade and a half; I mean if what you're telling me is true and so is the marketing.

 

[Atech]

Oh. Oh, okay.



I do worry about it. There are two things that worry me:

1. If they tout this device to be specialized in doing 'super computer' style tasks (floating point arithmetic), what does that indicate about it's graphical rendering or game performance? Have they nothing to brag about that? Potentially by catering to a field never really tapped into before and ignoring the gaming segment, it is alienating. It is a super computer and can play games? A jack of all trades is a master of none. Do one thing, do it well.
2. Since this is a super computer you can stick into your PCIe slot, will there be a cheaper version that caters to the "mainstream" market with reduced or removed functionality? What if I don't need to calculate millions of digits of Pi or do any of the scientific stuff, I just want to kill pixels in games. I imagine this card will be prohibitively expensive, so will there be a slim version? (e.g: 5870 is to 5770 as Fermi is to what?) Like, can they just forgo all the CUDA and scientific calculation stuff and give me a rock solid video card that runs games at light speed?

I don't know, but by the way they are marketing this card, they make it sound like they've lost focus of their audience they have had for the past decade and a half; I mean if what you're telling me is true and so is the marketing.

The G80 has gaming and GPGPU ASIC mixed just fine.
The G200 has gaming and GPGPU ASIC mixed just fine.
Fermi has more than twice the AISC for gaming than the GTX285...and it has more than twice the AISC for GPGPU...what is the problem?

That NVIDIA didn't talk gaming at their GPGPU Conference?
Like Intel dosn't talk 3DMark numbers on their Xeon CPU's?
Come on...

 

[PrincessFrosty]

Nvidia should have done all the games related conferences first and sold it as a GPU first and foremost and then touted its super computing power afterwards, what they've done just has gamers confused, obvious from all the posts in this thread.

 

[Yossarian22]

The G80 has gaming and GPGPU ASIC mixed just fine.
The G200 has gaming and GPGPU ASIC mixed just fine.
Fermi has more than twice the AISC for gaming than the GTX285...and it has more than twice the AISC for GPGPU...what is the problem?

That NVIDIA didn't talk gaming at their GPGPU Conference?
Like Intel dosn't talk 3DMark numbers on their Xeon CPU's?
Come on...

But what if I don't want GPGPU? I just want a regular GPU? A cheaper one. 5870 is to 5770 as Fermi is to what?

If it has twice the ASICs, then its performance should be somewhere slightly near a GTX 295 or slightly better, right?

It is a video card and if they claim it should be out before the year is over, not even a benchmark or anything? Why are they going to ignore the majority of their audience like that? That's not cool.

 

[sbuckler]

Nvidia should have done all the games related conferences first and sold it as a GPU first and foremost and then touted its super computing power afterwards, what they've done just has gamers confused, obvious from all the posts in this thread.

It was sensible to cover to show it to super computer users first. They take a long time to decide to use something, hence nvidia needed to get out the info on ECC memory, C++, float, etc so the people that build and use these number crunchers will start to consider using Fermi.

Game users need no lead time at all - when it's out if the performance/$ is right they will buy it on day one without hesitation.

 

[RevMen]

Who cares? I mean really, besides a few uptight people with nothing better to worry about, who really cares?

I made the original comment when there wasn't much going on with this thread. It was about a "silly question" so it seemed appropriate.

Fermi was a Nobel Prize winning physicist whose work is unequaled by all but a select few in history. If you're going to pilfer his name for your own marketing purposes, at least show enough respect to say it correctly.

 

[Mr. Pedantic]

are you high, drunk or both?

\ə-ˈlü-mə-nəm\

Its America, we say it like how we want to.

exactly.. thats why americans say aluminum as a-lum-in-um instead of ale-min-eum like it should be.. i mean hell we created american football because we didnt like the rules of european soccer.. its the american way to say "eff you to the world, we are america and we say/do what we want"..

Aluminum's aluminum. Obviously, if someone was talking about this other zany material aluminium they'd spell it that way. Hell, firefox is telling me I'm spelling it wrong even!

The blind jingoism in this thread is almost too much to bear...

But what if I don't want GPGPU? I just want a regular GPU? A cheaper one. 5870 is to 5770 as Fermi is to what?

Presumably, by definition a 5870 is to a 5770 what a top-end GT300 is to a lower-end GT300.

It is a video card and if they claim it should be out before the year is over, not even a benchmark or anything? Why are they going to ignore the majority of their audience like that? That's not cool.

Aww...poor diddums has no benchmarks to flaunt and flame AMD users with...poor you...

 

[Yossarian22]

Presumably, by definition a 5870 is to a 5770 what a top-end GT300 is to a lower-end GT300.

So are they purposely making a lowered powered chip based on the same architecture or doing parts binning (probably since there are like 4B transistors, sounds really complex to make) or what? They haven't said or been specific on the issue.

Estimated pricing? The highest or best Fermi will probably be $1000 USD or more, given all these advanced features (I liked the tech demo at the conference, each mote of dust had it's own gravity and mass data and stuff).

Aww...poor diddums has no benchmarks to flaunt and flame AMD users with...poor you...

Why would I want to flaunt and flame people? It's a video card (or isn't in this case), not a cult. I am curious to see if it is worth waiting to buy one or not, benchmarks are actual tangible and empirical data sets, so they're kind of important.

 

[Kowan]

Did you read the quotes from ChrisRay I posted?
How soon was it info on the 8800GTX leaked prior it it's release?
TSMC having troubles on 40nm isn't helping.

 

[Yossarian22]

Did you read the quotes from ChrisRay I posted?
How soon was it info on the 8800GTX leaked prior it it's release?
TSMC having troubles on 40nm isn't helping.

Who is Chris Ray? Isn't he a pitcher for the Orioles?

 

[Kowan]

Yes the one that works for nVidia.

 

[Atech]

But what if I don't want GPGPU? I just want a regular GPU? A cheaper one. 5870 is to 5770 as Fermi is to what?

Are you for real?

If it has twice the ASICs, then its performance should be somewhere slightly near a GTX 295 or slightly better, right?

A GTX295 != 2xGTX285...it has 2x240 cores, but they only run at GTX260 speeds.

It is a video card and if they claim it should be out before the year is over, not even a benchmark or anything? Why are they going to ignore the majority of their audience like that? That's not cool.

You are for real...have a nice day...if you cannot get what a GPGPU conference is and isn''t then good luck...

 

[Yossarian22]

Are you for real?



A GTX295 != 2xGTX285...it has 2x240 cores, but they only run at GTX260 speeds.



You are for real...have a nice day...if you cannot get what a GPGPU conference is and isn''t then good luck...

Estimated. It will be about the speed of 2xGTX285/GTX295 but priced at around $700+ ish? No thanks.

This is too confusing. Nvidia has gone barmy. What is Kyle running? Kyle, tell me what you're running.

 

[Mr. Pedantic]

So are they purposely making a lowered powered chip based on the same architecture or doing parts binning (probably since there are like 4B transistors, sounds really complex to make) or what? They haven't said or been specific on the issue.

Estimated pricing? The highest or best Fermi will probably be $1000 USD or more, given all these advanced features (I liked the tech demo at the conference, each mote of dust had it's own gravity and mass data and stuff).

Well, that's what normally happens with NVidia, they bring out a super-top-end part and then slowly scale it down for cheaper parts and slower cards.

This is too confusing. Nvidia has gone barmy. What is Kyle running? Kyle, tell me what you're running.

4870X2s, I thought.

Why would I want to flaunt and flame people? It's a video card (or isn't in this case), not a cult. I am curious to see if it is worth waiting to buy one or not, benchmarks are actual tangible and empirical data sets, so they're kind of important.

Sorry. I was being flippant. Never mind.

 

[jeremyshaw]

HD5870, but lets not bring Kyle into this. It's his personal choice, after all.

 

[Kowan]

4870X2s, I thought.

He's running a single HD5870. He stated he was running 2 HD4870x2s before going 3 monitors.