anandtech: The Sandy Bridge Preview

[James]

Indeed. Looks like I'll finally commit to a new laptop.

[Nater]

I don't see why people are whining about new sockets, especially at the high-end. The mid-ranged parts could maybe have been done on LGA1156, we won't know until we know more about them.

The high-end, LGA2011 parts definitely would not work with LGA1366. The high-end Sandy Bridge B2/EN/EP processors have a quad channel memory controller while obviously Nehalem/Westmere chips used a three channel one. That not only requires more pins for the extra bandwidth but it also means you have the wrong number of DIMM slots on X58 motherboards. Excepting some of the early, cheap models made with four DIMM slots. Not to mention, that by the time the high-end Sandy Bridge processors launch the middle of next year LGA1366 will be nearly three years old. Get a grip, three years is a long time for a socket.

Sandy Bridge is a tock. that means new sockets. The high-end, LGA2011 processors look like they're going to be monsters. Eight full cores, sixteen threads with better per clock performance than Nehalem/Westmere, plus AVX instructions, with likely higher clock speeds with a now mature 32nm node.

[nukefault]

Nater - If you'd just spent $250 on an X58 mobo last year you'd probably want to get a bit more use out of it. Unless you had enough money to burn that you're routinely buying ultra-expensive mobos that generally only pay for themselves with a $1000 processor (Lynnfield >= Nehalem for just about everything except absurdly memory-intensive tasks below the Extreme CPUs), in which case blowing $250 every year or two might not be an issue...

Still, the problem isn't so much that there's a new socket this year (or two of them, even) as that there are new sockets all the friggin time. AMD's somehow managed to keep rolling on AM3 with massive backward compatibility for so long it should at least be counted as an advantage of the way they do business. Would you rather buy an overpriced sadgasm G45 mobo for that E6320 or would you rather drop a nice new 785G/880GX on your old X2 6000+?

Of course, in my desktop days I sold and upgrade mobo and CPU together, and now that I'm using a lappie I don't really care anymore... but I can still grumble

[richardginn]

Nater - If you'd just spent $250 on an X58 mobo last year you'd probably want to get a bit more use out of it. Unless you had enough money to burn that you're routinely buying ultra-expensive mobos that generally only pay for themselves with a $1000 processor (Lynnfield >= Nehalem for just about everything except absurdly memory-intensive tasks below the Extreme CPUs), in which case blowing $250 every year or two might not be an issue...

Still, the problem isn't so much that there's a new socket this year (or two of them, even) as that there are new sockets all the friggin time. AMD's somehow managed to keep rolling on AM3 with massive backward compatibility for so long it should at least be counted as an advantage of the way they do business. Would you rather buy an overpriced sadgasm G45 mobo for that E6320 or would you rather drop a nice new 785G/880GX on your old X2 6000+?

Of course, in my desktop days I sold and upgrade mobo and CPU together, and now that I'm using a lappie I don't really care anymore... but I can still grumble

SOOOOOOOOOOOOO Correct.

[Nater]

Not really. If you'd been watching intel's roadmaps for a couple years you'd know that you get a new socket with a new microarchitecture, which works with the next node. That's it. I don't feel sorry for anyone pissed because LGA1366 isn't going to work with Sandy Bridge B2. I do feel sorry for the LGA1156 guys, but again, when Lynnfield launched everyone assumed that Sandy Bridge DT processors weren't going to slot into LGA1156.

Everyone that asked in 2009 I told not to get on LGA1366 unless they absolutely needed it for a single socket workstation. People buying 920s and a single GPU with a $300 motherboard would have been a lot better suited buying 750s and cheaper boards. Tylersburg's advantages over Ibex Peak weren't going to be realized by the average power user. If you're going to need to attach PCIe SSDs, SAN cards, RAID controllers, and/or GPU co-processors for engineering applications, yes LGA1366 did and still does make sense. For 99% of people out there, they're not going to really use it's advantages over Ibex. PCIe bandwidth and memory capacity/bandwidth.

AMD manages it because they have too on one hand and on the other, they don't innovate as much. Bulldozer is their first completely revamped design in nearly eight years. Intel is banging out new chips at a far higher rate (Conroe, Nehalem, Sandy Bridge).

AMD also has to commit precious engineering resources to ensure that new processors work with old sockets and likely compromises are involved. Intel isn't going to do that, their processors performance speaks for itself. They also want the best electromechanical performance for each individual microarchitecture. AMD has to offer other advantages to people because of their lower performance processors.

[nukefault]

Well, that and Intel really likes selling us new mobos every year or two...

[richardginn]

Sandy Bridge Graphics Update

http://www.anandtech.com/show/3885/s...raphics-update

Anand is still trying to confirm if that is GT2 graphics in the CPU he previewed.

[nukefault]

Will be interesting to see if they're GT2 or not. IMHO if the current rumored roadmap where GT2 on the desktop is only in high-end parts is correct... well, that's pretty effing dumb. You want better IGPs on the cheap processors to handle light tasks on a low budget/small space. You want low-power graphics on the high end chips just as a backup in case the graphics card goes belly up or you don't need 3D horsepower. Hopefully the rumor is just wrong.

[Nater]

Well, that and Intel really likes selling us new mobos every year or two...

I'm sure they do, but in this case conspiracy theories fall to technical requirements. DDR3 doesn't offer much performance advantage over DDR2 and your processors need more memory performance. Then you have to add more memory channels and more memory channels means a new motherboard.

Intel is going to double (or more) per socket performance of heavily threaded workloads over the original Nehalem-EP with Sandy Bridge-EP. The market that matters (servers) doesn't care that they'll need new sockets because they don't upgrade CPUs to begin with.

As for the high-end desktop chips, everything I've seen so far suggests that they don't have on-die graphics. Considering they're meant for servers, workstations, and very high-end desktops where either graphics aren't important or discrete chips are going to be used anyway...it makes sense not to waste die space on something that isn't going to be used.

[nukefault]

Eh? I already agreed that there's a need for 1366 and probably its quad-channel successor. It's just put in a bad light by Intel's general lack of interoperability. Does the tick tock actually require a new socket every 2 years or so? Maybe. But in this case "conspiracy theories" (read: good marketing) coincide with possible technical requirements rather than somehow falling to them. AMD's been doing just fine transitioning to DDR3 with backward-compatible sockets and Intel could certainly do the same if it wanted to.

Of course, it doesn't want to, because it makes more money selling new mobo with new chips. That isn't a conspiracy theory; that's business 101. AMD's B101 consists of making a more budget-conscious crowd happy with long-term compatibility.

As to high-end Sandy, I still like having some form of IGP even if it's a weak one just in case the graphics card dies - or so you can save some money/time/complexity by just dropping in a minimalist IGP. They could probably make an absolutely tiny 2-3 EU part for their high end chips as a fallback if they needed. What the market thinks of that I don't know, but it makes sense to me.

[Nater]

AMD only goes for interoperability when they're loosing customers hand over foot like they have been the last few years. When their CPU sales were climbing from 2003-2006, you had the S940/S754, S939, and AM2 for what was essentially the same CPU (degraded IMC on the S754 chips).

But now, AMD has to hold on to every single customer they can so it's worth it to them to burn engineering resources on socket backwards compatibility. It's a way to hold on to people when they have no real chance at challenging intel on performance for at least another year.

[richardginn]

Will be interesting to see if they're GT2 or not. IMHO if the current rumored roadmap where GT2 on the desktop is only in high-end parts is correct... well, that's pretty effing dumb. You want better IGPs on the cheap processors to handle light tasks on a low budget/small space. You want low-power graphics on the high end chips just as a backup in case the graphics card goes belly up or you don't need 3D horsepower. Hopefully the rumor is just wrong.

I have to disagree with you a bit....

Intel needs one and only solution for the desktop CPU no matter how expensive the CPU is and that is GT2 even if the benchmarks we saw were from a legit GT1 GPU.

BUUUUUUUUUUUUTTTTTTTTTTT

You may be able top convince me if Intel had a 1 grand CPU that also included for free 300+ bucks worth of GPU on it for a separate solution.

[nukefault]

I'd love to see that happen but Nater has a valid point about die size issues.

Nater - that's a different take than I had on AMD's socket system. The 754/939/AM2 debacle was disappointing but it's the only point I know of where AMD's had socket inflation so I'm inclined to regard it as an outlier rather than a standard way of doing business. On the other hand, it was also one of few periods where AMD was on top so maybe you're right. Meh, I don't know enough to have a strong opinion.

[Nater]

I just realized that desktop Bulldozer parts are not going to be compatible with AM3. They require AM3+. I had only paid attention to the server pieces (which work in G34). I'm not sure if Magny Cours required a new socket, but it doesn't seem unlikely that they went with one anyway to support future processors.

[kpxgq]

Not to mention, my Q6600 is looking a bit long on the tooth. About time for a platform upgrade on the home system. .

really?

im still on my "ancient" core2duo and i still dont find it slow for anything. i play all the latest games at the highest settings with no slowdown. all my work programs run with no lag (i do professional production work with adobe software), id say my SSD gave the biggest performance boost if anything. the only time i want a faster cpu is when im encoding video which about 3-5% of the time.

i think that since the core 1 duos's that we hit a software plateau where we dont need faster hardware anymore... all my previous cpu's (celeron 333, tbird 1ghz, tbred 1.4, venice 3000, x2 4200, etc)... ive steadily felt a need to upgrade but not this time around