So this is Larrabee?

O

octoberasian

2[H]4U
Joined
Oct 13, 2007
Messages
4,081
Says it all here: http://en.expreview.com/2008/07/07/larrabee-unleashes-2-tflops-capacity/

http://en.expreview.com/img/2008/07/07/larrabee-pci.gif

"Meanwhile, Gelsinger said the Larrabee will be a 45nm product featuring SIMD technique, 64-bit address. Besides, 32 of cores runing at 2.00 GHz will unleash 2 TFLOPS capacity, twice as much as the RV770XT. And what’s more, this monster requires a 300W TDP, which also frustrates NVIDIA’s biggest monster."

Interesting.

- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
Mod Edit - Please don't hotlink images - BlindedByScience
 
If it will come out in the end of 2009 or early 2010 then, meh. nVidia has just got kicked in the nuts by AMD so they will definitely come out with something better by the end of 2009. AMD has already reached 2.4 TFlops on a single card with their HD 4870 X2 and they only need a die shrink to lower the power usage.

Btw to the OP, did you pay for their bandwidth?

Edit: 2450 mm^2 die size?
 
Sounds great on paper, but we will see. A lot can happen in the next year or two.
 
For a big part it's about having the right product at the right time.
As GPUs keep getting more programmable and more generic, they enter Intel's area of expertise.
G80 already was a very convincing demonstration of how executing programs efficiently is far more important than just having the raw processing power.

So if Intel times this right, they may ride on the wave of the new-and-improved programmability of DirectX 11, and they might implement it more efficiently than nVidia and AMD, which will give them a competitive edge (they're only aiming for mid-end performance anyway, with this first generation... 2 TFLOPS seems to be very comfortable, as I expect today's 1 TFLOP cards to still be reasonable mid-end performers in 2009).
Ofcourse they could also tank royally... Later this year they'll give the first demonstrations of their Larrabee technology, that should give us a rough idea of how well they will be doing.
Intel only *just* needs to get its foot in the door... if they see that there's a possibility to catch up with AMD/nVidia, that should be enough incentive to keep investing and developing... and Intel has plenty of resources for that.
 
I'm rooting for intel to succeed. More competition means faster cards and lower prices.
 
I'm rooting for intel to succeed. More competition means faster cards and lower prices.
Click to expand...

Right now Intel's biggest fear is that AMD will die (which fiscally, it is in SERIOUS trouble) and it becomes subject to major anti-trust/monopoly crap. atm, Intel is valued at 115B while AMD is at 2.7B (and dropping like a rock). Yeah, you heard that right. Intel is 40 times the size of AMD/ATI on paper. Short term, this may be good. Long term we may have a serious monopoly on our hands.
 
Intel could always invest in AMD.
It's not entirely uncommon for a strong company to invest in a weaker competitor in order to maintain the illusion of competition.
 
Scali2 said:
Intel could always invest in AMD.
It's not entirely uncommon for a strong company to invest in a weaker competitor in order to maintain the illusion of competition.
Click to expand...

Intel grosses the Market capital of AMD in ~26 days, invest? Why not buy :p

Seriously though, I think Intel is scared shitless over anti-trust enough atm. Investing/buying AMD would get them hammered.
 
unless they come out with a single processor solution. Ie GPU+CPU in one :D then if everyone else follows suit (NVIDIA and Intel) then its a 3 way horse race by then. Better prices for us
 
Trying to steer this post back out of wall st. - not because it's not that relevant, just because I hate to talk about money and finances :p.

I think the really super crucial part here is for Intel to make drivers that don't turn this thing into a waste of space...and if that gif is to be believed A LOT of space. I don't trust that gif personally, maybe at some archaic process like 90nm it would turn out that large just as a proof of concept or something. Production samples would have to be much much smaller to have any sort of chance of profitability and good yields.

Intel's driver department has been far less than impressive with their graphics products in the past. Partly because all they've had to do is make sure they worked for the desktop and a few media apps and everything else was extraneous, they simply weren't meant to be used for gaming, etc. That's a pretty simple thing to do in comparison.

When gamers get their hands on these things they will be doing a lot more than making powerpoint animations. AMD and NVidia have been doing this type of thing for years now, it's old hat. I hope Intel has been hiring some top notch driver developers with their oceans of cash, they're going to need them. Otherwise they'll be bounced out of the club before they get their foot near the door.

I also question their real reasoning for developing these chips. They say they are to be used for "graphics". Really when I see them I do indeed see "graphics", but instead of videogames I'm thinking pre-rendered scene creation. Dreamworks has already jumped onboard the Larrabee wagon, others will soon follow. I think these chips might end up as a huge boon for Intel in the HPC/CGI market, but may offer little to the average gamer out there. All TFLOPs are not created equal.

Game engine developers are used to having to deal with graphics APIs. A change to x86 would be the same as asking them to learn a new API, no matter what Intel's marketing tries to spin its x86 compatibility as. Beyond future development, what about backwards compatibility? It comes back to drivers, how well are they going to interface with the API calls of yore and translate them into x86 instructions that the Larrabee can use? How much of that 2 TFLOPs will be used solely for translation?

Color me unconvinced that Intel actually wants to enter the same market as AMD's graphics division (their CPU division is a whole 'nother story) and NVidia right now.
 
More competition, the better.

Shall be interesting to have another competitor in this market though.

Wish them the best on taking on two heavy champs.
 
Silent.Sin said:
Game engine developers are used to having to deal with graphics APIs. A change to x86 would be the same as asking them to learn a new API, no matter what Intel's marketing tries to spin its x86 compatibility as. Beyond future development, what about backwards compatibility? It comes back to drivers, how well are they going to interface with the API calls of yore and translate them into x86 instructions that the Larrabee can use? How much of that 2 TFLOPs will be used solely for translation
Click to expand...

They have already stated that it will still be the same ol' DirectX and OpenGL. The x86 part is really only for GPGPU stuff (like CUDA), which really doesn't matter either, as there aren't really any x86 programs that are massively parallel using light threads that would be suited for GPGPU execution.

The "larrabee is x86!" thing is more a marketing gimmick than any sort of real technical advantage. That said there isn't really any reason *not* to use x86 - so its really neither a pro or a con. Intel only went with x86 because they know x86.
 
Silent.Sin said:
Game engine developers are used to having to deal with graphics APIs. A change to x86 would be the same as asking them to learn a new API, no matter what Intel's marketing tries to spin its x86 compatibility as. Beyond future development, what about backwards compatibility? It comes back to drivers, how well are they going to interface with the API calls of yore and translate them into x86 instructions that the Larrabee can use? How much of that 2 TFLOPs will be used solely for translation?
Click to expand...

An API is not the same as an instructionset.
An API is just a set of functions with a well-defined input and output.
nVidia and ATi both use their own instructionsets (and these instructionsets generally only last a generation of hardware/Direct3D), x86 will be more of the same. All you need is a compiler that converts the Direct3D/OpenGL shaders to native format. And Intel is good at building compilers.
 
kllrnohj said:
They have already stated that it will still be the same ol' DirectX and OpenGL. The x86 part is really only for GPGPU stuff (like CUDA), which really doesn't matter either, as there aren't really any x86 programs that are massively parallel using light threads that would be suited for GPGPU execution.

The "larrabee is x86!" thing is more a marketing gimmick than any sort of real technical advantage. That said there isn't really any reason *not* to use x86 - so its really neither a pro or a con. Intel only went with x86 because they know x86.
Click to expand...

Right, it would have to be an OpenGL/DirectX part in order to gain any traction with developers. That was my mistake for thinking they would try and force x86 down the developers throats.

However, they will still have to deal with an abstraction layer somewhere that is able to make these API calls work with those x86 cores; the same way AMD and NVidia have to provide the functionality to make generalized API calls work with their own specific and very different SPs, etc. I'm just not convinced that Intel can do that so well on their first try. This is like a 10 year old Michael Jordan going up against an NBA All-Star team. They've got the potential and the resources to get there eventually, but this is an unfair fight out of the gate and they might stumble quite a few times before they get it right. Are they going to have the resilience to stick around?

After the API calls are dealt with, then the question becomes how useful these general purpose cores are for something as specific as gaming graphics. Will they really be able to compete with the computing units developed by AMD and NVidia very specifically for GPU tasks? Will 32 CPUs with their ALUs and FPUs be able to compete with 800+ stream processors?

It seems logical now that GPUs can be used for certain types of CPU-type tasks (folding, encoding/decoding, etc) that's just the nature of what the SPUs are designed to do: crazy-fast floating point ops. That statement, up until this point, cannot be reversed, CPUs are not very good at GPU-type tasks. That's the reason graphics-acceleration was created in the first place.
 
Larrabee really isn't all that terribly different from ATI/Nvidia's cards. Yes, it uses a Pentium architecture, but with a shit ton of floating point units bolted on (sound familiar?). The x86 part of larrabee is completely irrelevant - it has no meaning whatsoever (just like no one gives a damn what the instruction set of ATI and nvidia's cards are).

Do not confuse larrabee being x86 with larrabee being CPU-like. It is not. It is much closer to that of a typical GPU than it is a CPU, it just happens to speak x86 instead of AtomBIOS or whatever ATI cards use.

All that said, the card will likely not be too terribly attractive as Intel will most likely drop the ball on the drivers.
 
kllrnohj said:
Larrabee really isn't all that terribly different from ATI/Nvidia's cards. Yes, it uses a Pentium architecture, but with a shit ton of floating point units bolted on (sound familiar?). The x86 part of larrabee is completely irrelevant - it has no meaning whatsoever (just like no one gives a damn what the instruction set of ATI and nvidia's cards are).

Do not confuse larrabee being x86 with larrabee being CPU-like. It is not. It is much closer to that of a typical GPU than it is a CPU, it just happens to speak x86 instead of AtomBIOS or whatever ATI cards use.

All that said, the card will likely not be too terribly attractive as Intel will most likely drop the ball on the drivers.
Click to expand...

Okay so forgetting that the parts are x86 based, I'll try to focus more on the FP aspects of this thing. Each individual piece inside of the Larrabee (each part of it's vector processing array or whatever they call it) is-supposedly-a pentium class CPU, or at least a derivative of it. That seems like it will add a ton of unnecessary extra logic circuitry if it's purpose is to be a GPU. That logic might be good at other tasks, but those tasks aren't of interest to current video cards. AMD or NVidia would make use of that space by filling it with things more pertinent to getting higher framerates. No amount of driver development will make up for sheer lack of horsepower.

Maybe my problem is that I haven't read about what exactly Intel is modifying in these individual cores to make them so much better for Larrabee. Based on some rather vague articles and PRs that I've read here is my stance on what is happening:

So far I can tell the vector processing array will be multi-core, in the 24-48 range, and each of those subunits has 2 FPMACs. Each subunit will also be able to take on multiple threads. The pipeline for each FPMAC is very wide, allowing for multiple single-precision FP ops per cycle. So there is the *potential* for a lot of calculations going on: in a best case you get the full 24-48 x 8 single precision FP ops (or however many it can theoretically do with all threads and FPMACs fully loaded), but then in the worst case you get only the 24-48 x 2, which is a sad figure. This is the same problem that has plagued many GPUs, they have the potential to do a lot if all the stars and planets align, but that case rarely happens, and is usually at the cost of significant driver development time. Thread-scheduling is a beast that AMD and NVidia have been trying to slay for awhile now, and they've made lots of progress but it's taken quite awhile.

Information seems to be scarce, and very conflicting. Some articles say it's purpose is to be more of an add-on board, to help with some tasks that are compute heavy but are too general to be efficient on a GPU (ray-tracing for one), while others say it's squarely aimed to take on the GPU market. Can you point me to articles you've read through that might shed some light?

This discussion seemed like a good starting point, and this article in particular at least has some technical aspects to it. But that discussion suffers from the same problem I stated above, the info is just too conflicting to get a clear picture.
 
Silent.Sin said:
Okay so forgetting that the parts are x86 based, I'll try to focus more on the FP aspects of this thing. Each individual piece inside of the Larrabee (each part of it's vector processing array or whatever they call it) is-supposedly-a pentium class CPU, or at least a derivative of it. That seems like it will add a ton of unnecessary extra logic circuitry if it's purpose is to be a GPU. That logic might be good at other tasks, but those tasks aren't of interest to current video cards. AMD or NVidia would make use of that space by filling it with things more pertinent to getting higher framerates. No amount of driver development will make up for sheer lack of horsepower.
Click to expand...

But then you assume there is a lack of horsepower in the first place.
Don't forget that this is Intel, the largest chip manufacturer in the world.
Intel can probably design these floating point units more efficiently than AMD/nVidia can, because Intel has a technological advantage in that area.
Aside from that, Intel also has the advantage in production. Just look at AMD vs Intel CPUs.
Intel can make CPUs with more cache, higher clockspeed, better performance per clock, and make them smaller, cheaper and more energy-efficient than AMD, simply because Intel has a mature 45 nm process, and AMD is still struggling with 65 nm.
Just think about it, for GPUs, 55 nm is cutting-edge. Intel could have been producing 45 nm GPUs already, had they decided to get in the game sooner... which would give them similar advantages over the competition as they have with CPUs now.

Now, by the time Larrabee hits the market, 45 nm GPUs will probably not be new... but I think there's a good chance that Intel will be the first with a 32 nm GPU.
 
Scali2 said:
But then you assume there is a lack of horsepower in the first place.
Don't forget that this is Intel, the largest chip manufacturer in the world.
Intel can probably design these floating point units more efficiently than AMD/nVidia can, because Intel has a technological advantage in that area.
Aside from that, Intel also has the advantage in production. Just look at AMD vs Intel CPUs.
Intel can make CPUs with more cache, higher clockspeed, better performance per clock, and make them smaller, cheaper and more energy-efficient than AMD, simply because Intel has a mature 45 nm process, and AMD is still struggling with 65 nm.
Just think about it, for GPUs, 55 nm is cutting-edge. Intel could have been producing 45 nm GPUs already, had they decided to get in the game sooner... which would give them similar advantages over the competition as they have with CPUs now.

Now, by the time Larrabee hits the market, 45 nm GPUs will probably not be new... but I think there's a good chance that Intel will be the first with a 32 nm GPU.
Click to expand...

Exactly. Intel would rather spend less time in R&D (by using an old design as a base to modify to suit their needs) and make up for it with their manufacturing advantage. The actual Pentium part of Larrabee will likely go unused by DirectX/OpenGL games.
 
Scali2 said:
Just look at AMD vs Intel CPUs.
Intel can make CPUs with more cache, higher clockspeed, better performance per clock, and make them smaller, cheaper and more energy-efficient than AMD, simply because Intel has a mature 45 nm process, and AMD is still struggling with 65 nm.
Click to expand...

Intel puts more cache on their chips because there is no on die memory controller. Expect that trend to change when Nehalem comes out. Even still, most people will not notice the difference of a bigger cache outside of synthetic benchmarks.
 
Scali2 said:
But then you assume there is a lack of horsepower in the first place.
Don't forget that this is Intel, the largest chip manufacturer in the world.
Intel can probably design these floating point units more efficiently than AMD/nVidia can, because Intel has a technological advantage in that area.
Aside from that, Intel also has the advantage in production. Just look at AMD vs Intel CPUs.
Intel can make CPUs with more cache, higher clockspeed, better performance per clock, and make them smaller, cheaper and more energy-efficient than AMD, simply because Intel has a mature 45 nm process, and AMD is still struggling with 65 nm.
Just think about it, for GPUs, 55 nm is cutting-edge. Intel could have been producing 45 nm GPUs already, had they decided to get in the game sooner... which would give them similar advantages over the competition as they have with CPUs now.

Now, by the time Larrabee hits the market, 45 nm GPUs will probably not be new... but I think there's a good chance that Intel will be the first with a 32 nm GPU.
Click to expand...

I honestly don't believe that Intel can make an FPU meant for a rasterized graphics better than AMD or NVidia. At this point I think they've got that down pretty well, hell AMD can squeeze 800 specialized FPU-like units onto a single die, Intel hasn't done anything close to that. I won't say Intel can't equal their efforts, but they're coming in as rookies here. Ford sure as hell knows how to make engines that work and a ton of people buy them and they're used in tons of different cars/vehicles, that doesn't mean I wouldn't rather use an LS-7 (corvette engine) if I was going to be putting it into a car meant for racing.

Large cache sizes are pretty useless in the current GPU landscape, almost 100% of a modern GPU die is logic circuitry, quite the opposite of the Intel CPU. In fact I think I remember one of Intel's checkmarks for Larrabee was that it would be one of the first to have such large caches in a graphics chip. How they use that cache is certainly to be interesting, but it's another reason I have a suspicion this is geared more as a supplement card with its sights on the HPC market.

You're right, when Larrabee does see daylight it will be in an age of 40nm GPUs from AMD and NVidia, how much the 32nm process saves Intel will have to be seen. That is if it is indeed on 32nm, I could definitely see them come out first on a more mature 45nm node while 32nm is reserved for their bread and butter CPUs next year. A shrink to 32nm will be necessary for them to put lots of power on this chip to compete, but the decision to compete may come from how well the first gen does.
 
I have a feeling that Intel isn't going to compete on the high-end. For gamers and overclockers, etc. I think that market will stick with NV/ATI. Where Intel has a real potential for business is with the rest of the market. Imagine if all those crappy on-board Intel graphics chips could actually run games. Even if we are talking about Crysis on low settings, that is still a huge jump. Even if the first cards suck, they will surely be powerful enough to run some old favorites like WoW, CS:S, UT2004, etc. I think that potential is very exciting and could re-invigorate the PC gaming market. At least it could bring the PC platform to a certain baseline were anybody could at least attempt to play modern games (on low-medium settings). The high-end will likely stick with NV/ATI, but don't forget that is a small percentage of the overall market. I think this could be very promising if Intel doesn't botch up the drivers.
 
Silent.Sin said:
I honestly don't believe that Intel can make an FPU meant for a rasterized graphics better than AMD or NVidia.
Click to expand...
It's hard to say. Hot Chips in late August will have a Larrabee presentation and hopefully some of the details will be revealed. http://www.hotchips.org/hc20/program/conference_day_two.htm

If I had to guess, i'd probably agree that Intel's first higher end outing probably won't compare favorably to AMD and nvidia graphics chips by the time Larrabee comes out. But wouldn't it be a pleasant surprise? ;)
 
cybereality said:
I have a feeling that Intel isn't going to compete on the high-end. For gamers and overclockers, etc. I think that market will stick with NV/ATI. Where Intel has a real potential for business is with the rest of the market. Imagine if all those crappy on-board Intel graphics chips could actually run games. Even if we are talking about Crysis on low settings, that is still a huge jump. Even if the first cards suck, they will surely be powerful enough to run some old favorites like WoW, CS:S, UT2004, etc. I think that potential is very exciting and could re-invigorate the PC gaming market. At least it could bring the PC platform to a certain baseline were anybody could at least attempt to play modern games (on low-medium settings). The high-end will likely stick with NV/ATI, but don't forget that is a small percentage of the overall market. I think this could be very promising if Intel doesn't botch up the drivers.
Click to expand...

This is the defining statement. Drivers are everything... I dont care how powerfull the card is, if it doesn't have good drivers it will suck. In fact the latest intel integrated graphics are powerfull enough to handle older games, but they don't look good. I had a GMA 3100 (I'm pretty sure thats what it was), and while I was waiting for my 3850 to come in the mail I tryed to play half life 2 on it. Even Intels website shows a screenshot of halflife2 advertising that the 3100 could play it. It got decent frame rates at 1024x768 but objects it the game would suddenly dissapear, textures didn't look right, the whole game looked bad. The 3100 is capable, but the drivers were shotty. If intel doesn't get the drivers right, who cares what Larrabee is capable of...
 
kllrnohj said:
The actual Pentium part of Larrabee will likely go unused by DirectX/OpenGL games.
Click to expand...

Perhaps not.
I have a feeling that they will be (ab)using the classic integer pipeline and HyperThreading to simulate certain fixed function operations.
Eg, they can run a rasterizer thread on the classic pipeline while the SIMD pipe renders a few pixelshader threads in parallel. Something like that.
 
Xaijin said:
Intel puts more cache on their chips because there is no on die memory controller. Expect that trend to change when Nehalem comes out. Even still, most people will not notice the difference of a bigger cache outside of synthetic benchmarks.
Click to expand...

No, Intel puts more cache on their chips because they have a larger transistor budget than AMD. The amount of transistors that Intel spends on cache is larger than the amount that AMD spends on cache + memory controller.
Nehalem still has more cache than any AMD processor.
 
Silent.Sin said:
I honestly don't believe that Intel can make an FPU meant for a rasterized graphics better than AMD or NVidia. At this point I think they've got that down pretty well, hell AMD can squeeze 800 specialized FPU-like units onto a single die, Intel hasn't done anything close to that.
Click to expand...

And perhaps Intel doesn't need to.
Firstly, the '800 stream processors' or whatever AMD calls it, is mostly marketing. The actual chip design doesn't contain 800 actual physical units, but rather a smaller number of units that can perform operations in parallel (SIMD).

Secondly, while nVidia's design looks far less spectacular on paper (they claim 240 scalar processors, although these are also SIMD units really), they do actually outperform AMD's offering.

What nVidia does seems to be closer to what Intel will be doing anyway. The scalar approach with 16 threads in parallel seems to be the most efficient way for (GP)GPU operations.
And Intel has announced that Larrabee's cores will have 16 operand wide SIMD units. So all they have to do is execute code the same way nVidia does.

Silent.Sin said:
Large cache sizes are pretty useless in the current GPU landscape, almost 100% of a modern GPU die is logic circuitry, quite the opposite of the Intel CPU. In fact I think I remember one of Intel's checkmarks for Larrabee was that it would be one of the first to have such large caches in a graphics chip. How they use that cache is certainly to be interesting, but it's another reason I have a suspicion this is geared more as a supplement card with its sights on the HPC market.
Click to expand...

As far as we know, Intel will make a sort of 'system-wide' cache, quite similar to what G80 has, which is excellent for GPGPU tasks and things like physics acceleration.
 
Intel is teh BEST, the largest chip manufacturer in the world. They have the most advance technology in CPU manufacturing so they will not have any problem creating GPUs. They can design these floating point units more efficiently than AMD/nVidia can. The prove is their integrated grapchics, it is badass compared to the offerings from both AMD and nVidia :rolleyes:

Last time I checked, Phenom has a higher floating point performance clock for clock than the C2Q.
 
alg7_munif said:
Last time I checked, Phenom has a higher floating point performance clock for clock than the C2Q.
Click to expand...

Ofcourse everyone knows that SpecFP has turned into a memory bandwidth test. Phenom has an advantage because of the integrated memory controller.
You'll see that Nehalem sets this straight, despite not having considerable improvements in the FP units over Penryn.

In pretty much all real-world applications that rely heavily on floating point (eg 3d rendering, video processing etc), Intel is already ahead of AMD clock-for-clock.
 
I was talking about Phenom floating point performance which is a result of the more efficient floating point units, I'm not talking about Phenom performance in general real world application. *sigh* :(

Scali2 said:
But then you assume there is a lack of horsepower in the first place.
Don't forget that this is Intel, the largest chip manufacturer in the world.
Intel can probably design these floating point units more efficiently than AMD/nVidia can, because Intel has a technological advantage in that area.
Aside from that, Intel also has the advantage in production. Just look at AMD vs Intel CPUs.
Intel can make CPUs with more cache, higher clockspeed, better performance per clock, and make them smaller, cheaper and more energy-efficient than AMD, simply because Intel has a mature 45 nm process, and AMD is still struggling with 65 nm.
Just think about it, for GPUs, 55 nm is cutting-edge. Intel could have been producing 45 nm GPUs already, had they decided to get in the game sooner... which would give them similar advantages over the competition as they have with CPUs now.

Now, by the time Larrabee hits the market, 45 nm GPUs will probably not be new... but I think there's a good chance that Intel will be the first with a 32 nm GPU.
Click to expand...

Intel has an advantage in the CPU department, I'm not arguing that but this doesn't make them a better GPU manufacturer too, just look at their IGP, they are hopeless when compared to both offerings from AMD and nVidia.
 
alg7_munif said:
I was talking about Phenom floating point performance which is a result of the more efficient floating point units, I'm not talking about Phenom performance in general real world application. *sigh* :(
Click to expand...

I'm pointing out that Phenom doesn't have more efficient floating point units, you just picked a set of benchmarks that are bandwidth-starved on Intel CPUs currently.
Real-world applications are generally not bandwidth-starved, and then Intel beats AMD on even the most floating-point intensive tasks.
Purely synthetic tests that run entirely from the cache also show Intel the clear winner (eg some of the SiSoft Sandra tests).

alg7_munif said:
Intel has an advantage in the CPU department, I'm not arguing that but this doesn't make them a better GPU manufacturer too, just look at their IGP, they are hopeless when compared to both offerings from AMD and nVidia.
Click to expand...

That's because you're only looking at performance.
Intel's current IGPs are actually a very elegant design, and resemble G80 quite closely. However, they are designed to minimize production costs. Therefore they only have 8 to 10 scalar processors (where eg nVidia will use 16 or more), and on top of that, these scalar processors also run some of the tasks that other IGPs have dedicated hardware for (eg clipping and rasterizing).
Given the amount of transistors/resources that Intel spent on its IGP, it's actually a very good design. It was just never designed to compete with others on performance. It was designed to be cheap and energy-efficient, and that it is.

So Intel simply had different design goals for their IGPs. That doesn't make the IGPs bad.
That's like saying that a GeForce 9600 is a bad card because it's not as fast as a 9800. The point of the 9600 never was to be as fast as a 9800, it was aimed at a different market altogether.
 
Wow saying that Intel IGP is a very good design shows what kind of a person you really are. AMD 780G beats the Intel IGP in every aspect including performance, power consumption, features and sometime price. Even nVidia has a good IGP too, better than Intel..
 
alg7_munif said:
Wow saying that Intel IGP is a very good design shows what kind of a person you really are. AMD 780G beats the Intel IGP in every aspect including performance, power consumption, features and sometime price. Even nVidia has a good IGP too, better than Intel..
Click to expand...

Well, it could be that I'm "some kind of person", but I would say that Intel indeed has the power consumption covered:
http://www.anandtech.com/mobile/showdoc.aspx?i=3117&p=11

(note also that the Intel system has a larger screen and a smaller battery)

As far as features go, Intel has a fully DX10-capable line of IGPs.
 
alg7_munif said:
http://www.anandtech.com/weblog/showpost.aspx?i=462
Click to expand...

It would be nice if there was an apples-to-apples comparison. Same CPU and all that. Else measuring the total system draw doesn't say a whole lot about the IGP.
In this test it's quite clear that AMD clocks its desktop CPUs lower in idle than Intel does.
In the previous test we had laptops, and then the story is different. Even in idle, AMD doesn't have an advantage there... and then we're not even considering the performance differences.
 
Scali2 said:
It would be nice if there was an apples-to-apples comparison. Same CPU and all that. Else measuring the total system draw doesn't say a whole lot about the IGP.
In this test it's quite clear that AMD clocks its desktop CPUs lower in idle than Intel does.
In the previous test we had laptops, and then the story is different. Even in idle, AMD doesn't have an advantage there... and then we're not even considering the performance differences.
Click to expand...

I'm sorry, I edited my last post. I love the numbers from Company of Heroes, less power consumption with twice the fps count. Can we consider the performance difference now? ;)

In the previous test, it was X3100 vs X1250.
 
You must log in or register to reply here.
Back
Top