Larrabee. Thoughts?

Well, from what Ive read, it has 32 shrunk down PentiumIII cores I believe, uses x86, and can do 2 TFLOPS at DP. That being said, it looks to be an excellent graphics encoder, audio encoder etc or gpgpu. How will it do in a rasterisation enviroment who knows? Im thinking itll do ok, and will need alot of help from game devs to make it shine

Some serious power draw from that beast - not surprising given the raw power it potentially can unleash.

If it were P4, the pipes would be so long, an entire game would be trapped in Larrabee before we saw anything heheh

300W TDP?

But it says 300W TDP over there... maybe the 32 cores are really hungry

romulus yeah, 300W TDP with current technology and at 2ghz. Heres an interesting little piece of info from my overclocking. A Q6600 @ 3.2ghz requires 133W, at 3.6ghz it can burn through up to 200W. Small dump in clock speed but the CPU wants a lot more power. So lets assume the die shrink doesn't help, they can still scale back to 1.6ghz which would knock a huge amount of power requirements off. Wait this is Intel, their die shrinks always help.

EDIT: Also they state it will support SIMD which allows for multiple cores from a single physical core much like Hyper Threading does. That means it could only actually have 16 cores halving the power requirements.

I was assuming these are going to be modified 'Atom' processors, since a few sites mentioned that these cores were going to process instructions 'inorder'.

Using the Atom itself has advantages but some severe disadvantages too. One of the advanteges would be the in-order architecture, given the planned usage.
One major disadvantage would be the inter-chip communication while hardly a priority for a GPU, it would suffer greatly because of the bus interface used. Looking at the server space and intels 4P and 8P processor scaling makes me wonder how they realize 32 processors on a single PCB. To my knowledge intel will use Quickpath, something the Atom does not have yet. Using Quickpath or a derivative of it is the only logical conclusion if intel intends to make larrabee more than a real basic vector processing unit.
The Atom processor provides some basic SIMD fetures that are similar to the P4 Hyperthreading. I read that Larrabee will provide up to 4 threads per core though.
In the end Larrabee will share some features with the Atom, but will be something of its own.


Here's a snippet from an intel pdf (Source):

Larrabee Architecture for Visual Computing -- With plans for the first demonstrations later this year, the Larrabee architecture will be Intel's next step in evolving the visual computing platform. The Larrabee architecture includes a high-performance, wide SIMD vector processing unit (VPU) along with a new set of vector instructions including integer and floating point arithmetic, vector memory operations and conditional instructions. In addition, Larrabee includes a major new hardware coherent cache design enabling the many-core architecture. The architecture and instructions have been designed to deliver performance, energy efficiency and general purpose programmability to meet the demands of visual computing and other workloads that are inherently parallel in nature. Tools are critical to success and key Intel® Software Products will be enhanced to support the Larrabee architecture and enable unparalleled developer freedom. Industry APIs such as DirectX™ and OpenGL will be supported on Larrabee-based products.

Here are a few more articles with a more in-depth look at it.

Link 1
Link 2

I'm not that impressed by their numbers, since NVIDIA has already stated that by that time, that kind of computational power will have been already achieved (2X4870 > 2TFlops). The real deal for me is the architecture change, and we'll have to see the blue guys dealing with parallelism. I wouldn't bother with this card, if only I didn't know Intel all too well to laugh them off.

could you elaborate why? what have they to gain by doing that.

HAH!!! wow, I cried after that one!

^good point. never thought of it that way.

Besides, isn't Nvidia trying to get into the cpu market? AMD is already in cpu market and gfx.

Intel may lose some cpu market share. Getting into the enthusiast gfx market could balance out that market loss.

Thats just it, Intel doesnt. It looks to be mainly a gpgpu first, gpu second. This doesnt have anything to do with Ray Tracing, so dont even guess that. If youve read various articles, stating that Intel thinks CUDA and AMD's gpu solution for physics/encoding etc will be just a footnote in history, as Intel believes everything thats done on x86 is superior, mainly due to familiarity with the devs. In most apps this is true, and CUDA will really have to take hold fast and furious, and have a great showing as well. As far as this being a gpu, who knows? Theyll have a helluva time with their software for DX connects/communications et al. We need more info, and Im not about to get whimsical about this and deny nVidias and ATIs rightful spot in the gpu hierarchy. Intel knows that nVidia and ATI just released their new cards, and we get this, a few hints, nothing solid, and leaving us all wondering. Bet by the next gen from nVidia and ATI, well see more crap about Larrabee as well

I thought the 4850 used less power than the 9800gtx+?

Hardly.

I think that would see Intel's market share in the gaming and enthusiast area dry up overnight.


It would be absolute suicide.



What do you get more from performance wise (for games - which is the key aspect of said market) - an AMD 3200X2 with a HD3870 or a Intel QX6800 with an intel AGP?

What's the point exactly? The cards could be bigger or smaller pretty irrelevant.
The HD4850/4870 also have about 50% more transistors in the same die size and outperform the GTX and GTX+ per watt, so if you're talking about efficiency ATi's fine this round, after having a terrible problem with the HD2Ks in that respect. So your one-sided statement seems pretty biased, since the efficiency crown goes back and forth.

Well it's not just power consumption it's efficient power consumption. doubling performance without doubling power consumption is agood goal, and does increase greatly the performance per watt figures.
Compare the perf/watt of the GF FX or R300 series versus the HD4K and they are sginificantly different in a short period of time, and mmore so than most areas.

It's far from isnsignificant. It may not be a mjor concern, however without keeping it in check you will run into a hard limit outside the PC.
It's all fine having a 2KW PSU, however if it's still drawing off the typical 15A 110/120V outlet you're going to have trouble.

For a single card it's not a major issue, but it's still means you have 300W of heat to deal with (TDP is about heat not primarily about current draw) and 300W is nothing to sneeze at heat wise, and added to the rest of the case component it's going to be pumping out a whole lota heat into people's rooms.

**Snigger**

The closer we get to an actual launch, and the more details get released, the worse this thing becomes.

1st it is based off Pentium MMX CPU cores.
2nd it uses around 300W at max power. -> http://www.fudzilla.com/index.php? [...] &Itemid=65
3rd it DOESN'T have power saving features. -> http://www.fudzilla.com/index.php? [...] 3&Itemid=1
4th it will not be cheap. -> http://www.fudzilla.com/index.php? [...] 5&Itemid=1

Add in things like memory bandwidth (128 GB/s as of 2006 presentations) already lower than the market leaders and it seems as far as a GPU goes - Larrabee 1st gen is gonna be a dog.

Its only FUD if it is not true.


There are a lot of stories about Larrabee out at the moment - none of them good.

Which, like I said earlier, we wont see until the next release of cards from nVidia or ATI. Intel is just trying to let people know theyre out there, and theyre coming. I dont buy any of this info at all. These measely few breadcrumbs. Wait for the release of the R800 or the G300 before we know anything about Larrabee. This isnt Intel showing off their pretty C2Ds, where they had a poor but well marketed product in the P4 already in the fight. Larrabee will either have to blow the doors off anything thats out, and then we will hear allllllllllllllllll about it, or if we only get these breadcrumbs, itll be just another attempt for a competing product in a well established field

I can't say I agree with you on this. Based on the information I've seen I expect Larrabee to be a tremendous flop for Intel. Most of what I've seen suggests that Larrabee will be capable of 2 teraflops, which would be great if it were released now, but it won't be released until next year. The 4870X2 is expected to be able to exceed 2 teraflops and it will be available next month, not next year. Given that both Nvidia and ATI roughly doubled the processing power of their GPU's over the last 8 months, Larrabee could actually be significantly slower than the competition when it is released.

There is also the fact that Intel has never produced a graphics solution that could be described as being anything better than marginally adequate.

Don't forget also until just before production if things look to be off peak of where they want to be, Intel can start evolving the design and push the existing framework onto the parallel processing market, one which is not inconsiderable (Dreamworks anyone?).

Either way, the graphics market desperately needs a viable 3rd player in it to compete effectively at the mid and higher ends. If nothing else it will help sharpen the minds of the two incumbents.

I thought DP wasnt important in graphics rendering? We desperately need a third party cpu maker thats much larger than Intel, has more money, more resources, more influence etc etc, No we dont need a third gfx maker. Itll be interesting. Needed? No. Does Intel need it? Yes, very much so. They may play down graphics, but in reality, other than business, the vast majority is home buyers who think their cpu is just fine, but would looove to be able to play games and encode videos quickly. Until its here, or close, its nothing. A new approach isnt needed yet, and this may not be it anyways.

Why do think that more choice will be to the detriment of the graphics card sector. Look at the recent history:

Nvidia - FX5xxx fiasco
ATI - 2xxx fiasco
Nvidia - no drivers for months

Add in all the positives surrounding each of these events and the outcome is clear - at each time there really was only one logical choice, and that choice seesaws between one or the other.

Why would we NOT want to be in a position where we could potentially choose from 2 success stories out of 3 manufacturers, or god forbid, all 3. How is that bad?

Also, why isn't a new approach needed? 5 years ago, all GPUs were single cards. Since then we have had multiple GPU solutions, each done differently. If the developer and driver support is there, why not have another new approach? It may fall flat, but it may also change the direction of the industry. Or it may just be another way.

Would you change your mind if Intel were substituted for another firm - is the dislike that institutionalised?

OK.

Intel start getting graphics on x86 programming models.

AMD/ATI can adopt as they have an x86 license... what can Nvidia do?



All of a sudden, we are left with AMD/ATI versus Intel in both CPU and GPU markets.



Given the size of Intel compared to AMD/ATI and the troubles they (AMD) have at the moment, Intel will win that, and go on to dominate both CPU and GPU markets - leading to a slow down of progression in both.

the ATI two series was only a relative flop due to microsoft moving the goal posts. lets not forget, if they hadn't we would have all been praising the 2900 series as the only dx10 card out there.

i personally would want their to be more choice but i do not like it being intel, they do not play nice with other companies. they are not someone who likes standards, they like to control things too much and i think if they did get some success could through brute force try and fracture the market as we have it.

this would not be a good thing for consumers.

i could of course be considered cynical but maybe just realistic.