Anand Sandy Bridge performance preview is up

[ilkhan]

Intel internal vs external used to make a lot of sense, now its just spaghetti to me. Tying one roadmap to another, completely lost. I blame lynnfield/clarkdale.

(I think thats what you meant, anyway)
Yeah I really don't know how they seemed to fuck that one up so bad. Lynnfield should've been released at the same time as bloomfield along with a 45nm dual instead of getting lynnfield 10 months later without a dual. Clarkdale/Arrandale/gulftown/32-quad should've been out last november but they were all late (or none-existent). What a fucking mess. Guess the chips just weren't ready.

 

[OCGuy]

(I think thats what you meant, anyway)
Yeah I really don't know how they seemed to fuck that one up so bad. Lynnfield should've been released at the same time as bloomfield along with a 45nm dual instead of getting lynnfield 10 months later without a dual. Clarkdale/Arrandale/gulftown/32-quad should've been out last november but they were all late (or none-existent). What a fucking mess. Guess the chips just weren't ready.

Or some guy in an expensive jacket, who knows nothing about CPU architecture (or CPUs in general,) decided that there were too many of the previous generation in the pipeline...or maybe the lack of competition in that segment warranted a delay in a launch to avoid cannibalization of what is out there already.

 

[StrangerGuy]

From what I seen so far, 2011 is going to be one helluva year to buy a laptop/netbook.

 

[ydnas7]

Assuming there are native a 4C (CPU) / 12EU(GPU) and a 2C 6 EU parts, it looks like Sandybridge might be quite a compelling upgrade.

 

[ydnas7]

1st box, generally where sandybridge slots in, sandybridge will be as a CPU about 50% better value than lynnfield/clarkdale, if you can use the IGP, then value raises to 75% better than lynnfield/clarkdale (ie for notebook users)

2nd box, currently the i3-2390 is missing from the rumors, it would if it existed, be a very nice dual core.

3rd box, for similar price points, the 4C4T i5-2400 is so much better value than the i5-650 that it is silly.

notes, I've calculated the average IPCC improvement from AT article as 13%, for the 2600 i boosted it to 16% due to the extra cache, the IPCC for the 'dales is penalized by not having a IMC, so that is why I boosted the the Sandybridge IPCC from 13% to be 25%. value of AVX is arbitrary put as 10%. who here would want avx turned off for a 10% discount? (is it at least 50% of users - i doubt it.)

 

[Gryz]

So if I understand correctly, the defining properties of SB are:
1) transitorsize went from 45nm to 32nm.
2) new architecture.
3) integrated GPU.

Does this mean that all new Intel CPU from now on will have an integrated GPU ?

I am a gamer. I will always put a high-end videocard in my PC. I will never use an integrated GPU. So for me it's a waste of transistors, waste of watts, waste of heat, waste of money. I rather have a CPU without integrated GPU. Even at the same price. Will there ever be new Intel CPUs like that ? I can't find anything about this issue.

 

[IntelUser2000]

(I think thats what you meant, anyway)
Yeah I really don't know how they seemed to fuck that one up so bad. Lynnfield should've been released at the same time as bloomfield along with a 45nm dual instead of getting lynnfield 10 months later without a dual. Clarkdale/Arrandale/gulftown/32-quad should've been out last november but they were all late (or none-existent). What a fucking mess. Guess the chips just weren't ready.

IMO the reason it happened is because they wanted to keep to the "Tick/Tock" schedule. Although Intel is generally regarded as uncontested leaders in process technology, only small portion of the products have that lead.

Servers were what they were lacking badly, so they must have decided Bloomfield arrive as fast as possible. The dual cores could have arrived with Lynnfield but they were delayed and the 32nm parts were pulled instead(search for Havendale). Otherwise, the first 32nm parts might have been March/April with Gulftown instead of January with Clarkdale.

 

[ydnas7]

So if I understand correctly, the defining properties of SB are:
1) transitorsize went from 45nm to 32nm.
2) new architecture.
3) integrated GPU.

I would add another property, SB is where the uncore and the 'chipset' is raised to the same level of importance as the CPU core. SB was developed by Intel Haifa who did their first fusion X86 CPU/GPU 10 years ago (Timna). This chip wasn't released to market because it relied on a memory controller (from elsewhere in Intel) that failed. SB is about weaving the core and uncore together, not some quick and dirty stitching like the 'dales or the expected simple stitching like on the LLano, or the throw graphics into the core like Larrabee.

 

[gevorg]

So if I understand correctly, the defining properties of SB are:
I am a gamer. I will always put a high-end videocard in my PC. I will never use an integrated GPU. So for me it's a waste of transistors, waste of watts, waste of heat, waste of money. I rather have a CPU without integrated GPU. Even at the same price. Will there ever be new Intel CPUs like that ? I can't find anything about this issue.

Its just a matter of time when integrated GPUs will be faster than videocards. A couple of leap steps like Sandy Bridge and its over. Until then, yes, its a bit of a waste, but Intel rather serve 99% of the market than some <1% hardcore gamers.

 

[Idontcare]

Its just a matter of time when integrated GPUs will be faster than videocards. A couple of leap steps like Sandy Bridge and its over. Until then, yes, its a bit of a waste, but Intel rather serve 99% of the market than some <1% hardcore gamers.

Yeah remember what happened to that sram slot on those pentium mobos? Along came the K6-III and Celeron 300A and bye-bye went the entire market segment that was stand-alone sram dimms.

I agree, we could see such happen to the GPU market as well. Neither Nvida nor AMD can afford the R&D for sustaining leading edge GPU designs if their market opportunity declines to merely those who can afford $500 video cards.

 

[IntelUser2000]

AMD might be fine because they will sell the processor graphics parts, while Nvidia has nothing.

This chip wasn't released to market because it relied on a memory controller (from elsewhere in Intel) that failed. SB is about weaving the core and uncore together, not some quick and dirty stitching like the 'dales or the expected simple stitching like on the LLano, or the throw graphics into the core like Larrabee.

That's a bit simplistic to say don't you think?

 

[Edrick]

Well, we should find out more info starting tomorrow from the IDF.

 

[ehume]

Well, we should find out more info starting tomorrow from the IDF.

Israeli Defense Force?

 

[Edrick]

Israeli Defense Force?

http://www.intel.com/idf/

 

[ehume]

http://www.intel.com/idf/

Sorry. It was a joke. Most of the places I read, the IDF is understood as only referring to that military organization.

 

[Gryz]

I would still appreciate it to get an answer to my question: will there be no more new cpus without an integrated gpu ?

Its just a matter of time when integrated GPUs will be faster than videocards. A couple of leap steps like Sandy Bridge and its over.

I wouldn't be so sure. The question totally depends on a) how much performance is needed, and b) how different is the task of a gpu compared to a regular cpu.

I used to work in the networking business. People used to say: dedicated hardware for network devices won't be necessary soon. Routers and switches will be built with regular PC components. Once cpus are so fast that they can do X million packets/sec, then we don't need specially crafted hardware anymore. Well, that was a mistake. A general purpose CPU can do very fast lookups. But it lacks the bandwidth to forward huge amounts of packets. The only way to build terrabit routers is by having special hardware.

Yes, I realize this is a small market segment. But it trickles down to other markets. If the technology is there, and production costs of special hardware is not so expensive, but resulting performance is a lot better, then companies will build special hardware. Gaming is a huge market, I don't see videocards go away just like that. Not in the next 10 years.

 

[DrMrLordX]

I would still appreciate it to get an answer to my question: will there be no more new cpus without an integrated gpu ?

On socket 1155, maybe not. Socket 2011 is another story entirely.

 

[ydnas7]

That's a bit simplistic to say don't you think?

yes its all a bit simplistic, a point is,
the design team for Sandybridge vividly remembers that a CPU design can fail if its chipset/memory that it relies on fails, this second attempt at fusion places equal emphasis on the uncore as on the core, and weaves the two together.

 

[ydnas7]

for the 'dales, the cpu part was elevated as higher importance, that's why it was in 32nm process whereas the northbridge uncore was 'only' in 45nm. it was a sensible design during ramp up for the 32nm process, but the 2 parts are not coupled closely like on sandybridge. On Sandybridge, at the worst its through a shared L3 cache, it could just as likely be directly through the ring bus.

Llano reminds me of an ATI/AMD version of clarkdale, except on one die and using better GPU, we are yet to see how synergistically its cores communicate with the GPU.

 

[IntelUser2000]

Looking at IDF 2010 San Franscisco overview.

Sandy Bridge graphics to be 25x faster than the 2006 graphics. To put that into perspective, that's 4x faster than the HD Graphics on the first generation Core i series CPU.

 

[Martimus]

I would still appreciate it to get an answer to my question: will there be no more new cpus without an integrated gpu ?

There is no integrated GPU in the first generation Bulldozer design. Although I expect to see one in the second generation.

EDIT: I didn't read your first post where you asked specifically about Intel processors. In that case, I do not know.

 

[Vette73]

Looking at IDF 2010 San Franscisco overview.

Sandy Bridge graphics to be 25x faster than the 2006 graphics. To put that into perspective, that's 4x faster than the HD Graphics on the first generation Core i series CPU.

Thats not saying much. Intel over states and under performs when it comes to graphics.

 

[IntelUser2000]

Nah. They showed Starcraft 2 demo on a mobile Sandy Bridge which ran at least medium settings. Geforce G210m only gets 17 fps on that.

 

[Vette73]

Nah. They showed Starcraft 2 demo on a mobile Sandy Bridge which ran at least medium settings. Geforce G210m only gets 17 fps on that.

And how many FPS did the SB get?
That and I hope you did not base the whole 25x on 1 item?

 

[amenx]

Its just a matter of time when integrated GPUs will be faster than videocards. A couple of leap steps like Sandy Bridge and its over. Until then, yes, its a bit of a waste, but Intel rather serve 99% of the market than some <1% hardcore gamers.

Thats assuming discrete GPUs dont advance. Videocards will always have an advantage over IGPs due to one thing: size. Much more circuitry, features, technology can be packed into discrete GPUs than can ever fit into a small CPU. There will always be 2 market segments imo, the high performance one which discrete GPUs will cater to and the lower end that IGPs can fill. Dont see it coming to an end anytime soon.