Sandy Bridge-E processor details leaked

[Umbongo]

According to this leaked slide it seems that there won't be 8-core uni-processor CPUs until Ivy Bridge, but it gives us more of an idea what the next single processor Mac Pro might look like. There will be 8-core Xeons for dual-processor models, though it may be a situation where these have clock speeds a lot lower than 6-core models making them less suitable for workstations and more for servers. So a 16-core Mac Pro, from Apple at least, might be some time away.

Although this slide is for the consumer/enthusiast market the UP Xeon processors have mimicked the XE, P2 and P1 categories with just the addition of ECC memory support for many years. The family name for The UP Xeons is E5-1600 and E5-2600 for the DP Xeons.

All these processors also have four memory controllers to provide quad-channel memory performance up to 1600MHz. It'll probably be 1333MHz on the lower end and 1600Mhz on the higher, like the current 1066/1333 split you see in today's Mac Pros.

Also of note is another refresh of LGA 1366 Xeons coming in Q3. While Mac Pros ship with 2.80x4, 3.20x4 and 3.33x6, the current line up on these three price points is 3.20x4, 3.20x6 and 3.46x6. In Q3 these will be 3.33x4, 3.33x6 and 3.60x6 providing lots of upgrade options for 2010 owners for the future.

 

[bearcatrp]

Thanks for the update. Looks like I'll keep my westmere PC around until ivy bridge comes out. Been looking for some Hexes to replace my quads so hopefully the new ones lower what's out there now.

 

[nanofrog]

As per past comparisons, it's complicated this time around as there's 3 sockets for Sandy Bridge total (LGA1155 parts, with LGA1356 and LGA2011). Only the LGA2011 are intended to have quad memory channels (1356 will still use 3 memory channels).

For example, there's been information that suggests there will be an Enthusiast Desktop part on the LGA2011 socket (which is capable of 8 cores on a single die). I'd expect there to be such a chip based on LGA1356 as well, and there's been some indications there will be 8 core variants of those as well (will wait to see if that one actually pans out, as I'd think it would be more effective from a marketing standpoint to limit those to 6 cores max.; helps push LGA2011 and makes the LGA1356 cheaper as a result of higher yields per wafer).

So far, Wiki's Sandy Bridge page has been rather decent in pulling in all the various tidbits of information. But ultimately, it looks like we'll have to wait for more definitive information, as it's still a bit sketchy IMO.

 

[Umbongo]

As per past comparisons, it's complicated this time around as there's 3 sockets for Sandy Bridge total (LGA1155 parts, with LGA1356 and LGA2011). Only the LGA2011 are intended to have quad memory channels (1356 will still use 3 memory channels).

For example, there's been information that suggests there will be an Enthusiast Desktop part on the LGA2011 socket (which is capable of 8 cores on a single die). I'd expect there to be such a chip based on LGA1356 as well, and there's been some indications there will be 8 core variants of those as well (will wait to see if that one actually pans out, as I'd think it would be more effective from a marketing standpoint to limit those to 6 cores max.; helps push LGA2011 and makes the LGA1356 cheaper as a result of higher yields per wafer).

LGA 1356 is intended for entry level servers. My guess would be that the processors will be lacking features, maybe things like hyper-threading, virtualisation options, turbo-boost etc. In addition to having lower L3 cache and clock speeds. We know the chipset has less features so it is all reducing cost. My impression is this is the new approach at separating EN and EP server markets as they add more features to the premier chipsets that aren't needed for many uses.

 

[nanofrog]

LGA 1356 is intended for entry level servers. My guess would be that the processors will be lacking features, maybe things like hyper-threading, virtualisation options, turbo-boost etc. In addition to having lower L3 cache and clock speeds. We know the chipset has less features so it is all reducing cost. My impression is this is the new approach at separating EN and EP server markets as they add more features to the premier chipsets that aren't needed for many uses.

LGA1356 supports HyperThreading according to Wiki, and it makes sense from my POV (servers that use software written for HT, could better utilize the cores as a result). I've seen nothing for sure on VM or TB, but expect those will be there as well.

Currently, I've not seen an indication they're going to make a desktop part out of LGA1356, but given the costs of the LGA2011 parts, I think there's spot for a couple/few (just disable ECC as is currently done and will be done for the LGA2011 enthusiast parts). I see it as a Performance Desktop part, and the LGA2011 as the Enthusiast socket.

May just be me though.

 

[CaoCao]

That's a lot of cache, since K-series goes so high increasing cache on higher end parts is the logical step. Maybe the fully unlocked Enthusiast chips will not have a max multiplier. Also the Enthusiast chips might lack IGP to allow higher clock rates.

 

[Chris7]

Very Basic Questions

LGA1356 supports HyperThreading according to Wiki, and it makes sense from my POV (servers that use software written for HT, could better utilize the cores as a result). I've seen nothing for sure on VM or TB, but expect those will be there as well.

Currently, I've not seen an indication they're going to make a desktop part out of LGA1356, but given the costs of the LGA2011 parts, I think there's spot for a couple/few (just disable ECC as is currently done and will be done for the LGA2011 enthusiast parts). I see it as a Performance Desktop part, and the LGA2011 as the Enthusiast socket.

May just be me though.

UmBongo gave me a link to this thread from a post I just made (thanks)...

Need a new Mac soon, as my MBP's graphic's card is now ancient. I don't want another laptop nor a cute, glossy, and impractical iMac. Will wait until the Sandy Bridge chip hits the MP Q4 this year or Q1 2012.

If the good people here could bear with me, I have some very basic questions...

1) The $300 Core i7 2600 is probably about to hit the iMac. This chip is 4-core, 3.4 Ghz with 3.8 Ghz Turbo.

Can Apple really release a single processor 4-core MP after this? (I guess this would be the LGA 1155 Xeon E3, yes?).

2) Or will the base model MP have to be a 6 core DP (LGA 2011, Xeon E5, 26xx)?

3) Which is likely to be clocked faster for the price/per core: The 6 core DP (LGA 2011, Xeon E5, 26xx) or the 4 core DP (LGA 1356 Xeon E5, 24xx)?

Please pardon any typos -- I'm into graphics software, which has led me to pay attention to the new chips (software and hardware is just about to catch up with HD video, IMO). But I only have a superficial understanding of exactly what is novel and what is not about these new chips, how they might be configured, and Apple's insanely amoral marketing approach.

-Chris

BTW, I'm embarrassed to say I bought an iPad today, straight from an Apple Store, complete with the hip sales kids. I needed something for work, and I just can't seem to properly maintain PC's, so I use Mac. But somehow I just don't feel cool enough for the bubbly little thing...

 

[nanofrog]

1) The $300 Core i7 2600 is probably about to hit the iMac. This chip is 4-core, 3.4 Ghz with 3.8 Ghz Turbo.

Can Apple really release a single processor 4-core MP after this? (I guess this would be the LGA 1155 Xeon E3, yes?).

The new iMacs were just announced, so take a look if you're interested.

As per the MP, they're not going to use any LGA1155 part, as those are consumer grade. The SB enterprise parts will be on 2 different sockets; the LGA1356 and LGA2011 (which is the one the MP will most likely use).

2) Or will the base model MP have to be a 6 core DP (LGA 2011, Xeon E5, 26xx)?

It appears there will be a single Quad core version of an LGA2011 part that runs at 3.6GHz. Hard to say if it or a 6 core unit will be chosen, as there's no pricing out yet (quantity pricing may be similar due to the higher clock speed on the Quad vs. the lowest Hex).

3) Which is likely to be clocked faster for the price/per core: The 6 core DP (LGA 2011, Xeon E5, 26xx) or the 4 core DP (LGA 1356 Xeon E5, 24xx)?

Hard to say just yet, as there's less information out on the LGA1356 parts than any of the others (not seen any frequencies for these yet).

BTW, Wiki's Sandy Bridge page is actually a decent source of information.

 

[Chris7]

...As per the MP, they're not going to use any LGA1155 part, as those are consumer grade. The SB enterprise parts will be on 2 different sockets; the LGA1356 and LGA2011 (which is the one the MP will most likely use).

It appears there will be a single Quad core version of an LGA2011 part that runs at 3.6GHz. Hard to say if it or a 6 core unit will be chosen, as there's no pricing out yet (quantity pricing may be similar due to the higher clock speed on the Quad vs. the lowest Hex)...

Could you please catch me up regarding a little history?

IIRC, before the Nehalem, the single 4 core Server and dual 4 core (8 core) Server chips were clocked similarly. When the Nehalem chips hit the Mac Pro, the single 4 core version had a much higher clock speed than the dual 4 core (8 core). As you know, currently the base 4 core Mac Pro uses the Xeon W3530 2.8 GHz "UP Server," whereas the base 8 core Mac Pro uses dual Xeon E5620 2.4 GhHz "DP Server" (I think the base 8 core was something like 2.33 GHz then).

1) I'm assuming this is because Intel could not get the same clock speeds with the DP Server chips as the UP Server chips. What happened that the clock speeds with the DP processors tanked on the Nehalem?

2) This is relevant to me because it looks like all of the Sandy Bridge LGA2011 Xeon E5 Chips are "1-2 P Server" (which I assume means "DP Server," but please correct me if I am wrong). This would seem to mean the server chips for single processor and dual processor applications would all be clocked similarly, as there is no DP/SP Server split this time around. Is there anything to this assumption?

3) According to Wik, Sandy Bridge "Desktop" LGA2011 starts at 3.2/3.8GHz standard/turbo. Any reason to think the server chips for single and dual processor Mac Pros will will be clocked lower?

4) On another thread Umbongo said (I think) that the Ivy Bridge probably would not hit the Mac Pro until about a year after the Sandy Bridge (so around Q4 2011 or Q1 2012). Anyone differ?

Again please kindly look past my typos...

Thank you for your time,

Chris

 

[Umbongo]

1) I'm assuming this is because Intel could not get the same clock speeds with the DP Server chips as the UP Server chips. What happened that the clock speeds with the DP processors tanked on the Nehalem?

The issue wasn't with Intel's engineering. Apple either had a deal with Intel for big discounts on Xeon processors run out (for transitioning to them) or they wanted more margin from the Mac Pro. They went from using two $700-$800 processors in their 2006 and 2008 base models, to using a $300 processor; or on the DP model, which now cost $500 more, two $375 processors.

It made less sense to use the UP processors before, there were far more differences in the LGA 775 and LGA 771 platforms and chipset features before LGA 1366 came along.

2) This is relevant to me because it looks like all of the Sandy Bridge LGA2011 Xeon E5 Chips are "1-2 P Server" (which I assume means "DP Server," but please correct me if I am wrong). This would seem to mean the server chips for single processor and dual processor applications would all be clocked similarly, as there is no DP/SP Server split this time around. Is there anything to this assumption?

It's carrying on as it is now from everything we've seen. E5-1600 are the replacement for 3500 and 3600 Xeons and are sister processors of the Core i7 line. E5-2600 will be the high-end DP processors.

3) According to Wik, Sandy Bridge "Desktop" LGA2011 starts at 3.2/3.8GHz standard/turbo. Any reason to think the server chips for single and dual processor Mac Pros will will be clocked lower?

I'm sure there will be low clock-speed Xeon E5-2600 processors, servers don't tend to need high clock-speeds and that is where the vast majority of Xeon processors go.

4) On another thread Umbongo said (I think) that the Ivy Bridge probably would not hit the Mac Pro until about a year after the Sandy Bridge (so around Q4 2011 or Q1 2012). Anyone differ?

Just to be clear, I say that because enterprise, manufactures and the likes of Intel, AMD and Nvidia have agreements in place that hardware won't become obsolete within certain time frames. Intel also control their market pretty much so can dictate releases and a year is a good length of time to sell a lot of product without having buyers holdout. Doesn't always make sense to consumers who want the best of the best now, but its how it happens.

 

[deconstruct60]

According to this leaked slide it seems that there won't be 8-core uni-processor CPUs until Ivy Bridge,

Eh? There are 8 core single CPU processor package right now.

http://ark.intel.com/Product.aspx?id=53676&processor=E7-4830&spec-codes=SLC3Q

It is up in the E7 range but these slides are primarily oriented really for the non Xeon labeled processors. Sure the "extreme" versions have same foundation as the Xeon's, but there are multiple archs going on also. The 8C Sandy Bridge Xeons with *no* budget for GPUs is entirely possible and will be in the E5 sequence. Dropping the GPUs give a bigger core and L3 cache transistor budget.

It is less likely though that all 8 cores will pass clocking at the maximum rate. So the ones that pass 6 will get binned into the "extreme desktop" bucket. The transistor count is the same. What is different is what is flipped on/off as being workable. So it will likely be possible to pull a 8 core Xeon normally targeted at DP set up and plop it into a single CPU package socket. Whether they market a single package only model to use all 8 is open question. The problem with the single package models are that more cores tend to be more expensive and that usually drives a perception "need" more GHz also. However, to keep more than 6 cores feed a higher clockspeeds you'll need more than 4 memory channels. That's one reason why the non "desktop" oriented processors models then to couple more cores to multiple package configurations. You need to feed the beast. Otherwise what you have is a long stream of no-ops because pumped through the processor while it waits on many workloads.

Although this slide is for the consumer/enthusiast market the UP Xeon processors have mimicked the XE, P2 and P1 categories with just the addition of ECC memory support for many years. The family name for The UP Xeons is E5-1600 and E5-2600 for the DP Xeons.

http://www.cpu-world.com/news_2011/2011050402_Intel_Xeon_E5-1600_and_E5-2600_processor_details.html

The 1600 and 2600 are both projected by Q4 in this article.

 

[deconstruct60]

4) On another thread Umbongo said (I think) that the Ivy Bridge probably would not hit the Mac Pro until about a year after the Sandy Bridge (so around Q4 2011 or Q1 2012). Anyone differ?

Huh? The Sandy Bridge Xeons aren't even out yet. They are the ones due around Q4 2011 / Q1 2012 . Intel usually releases Xeons on a yearly basis so Ivy Bridge Xeons of class that MP uses wouldn't be due until Q3-Q4 2012. Ivy Bridge follow on to the E3 class would appear around Q1-Q2 2012 timeframe.


Ivy Bridge (for any platform, not the MP specifically) will likely launch in Q4 2011/ Q1 2012. It is unlikely that Apple is going to dump Xeons for desktop processors in the MP models. It would be extremely difficult to maintain the MP price points with processors aimed at significantly cheaper boxes and laptops.

The reality is that GPUs are being integrated into the "CPU" package. That is increasingly going to become a misnomer because much more than the "CPU" is being placed in the package. The introductions that AMD and Intel are going to do will be for these intgrated packages first. Those are then followed by separate deriviations of that same basic architecture that dumps the GPU to make room for more classic computation stuff ( general purpose cores and cache ) and/or greater I/O.

 

[Umbongo]

Eh? There are 8 core single CPU processor package right now.

My statement meant that I don't feel we will see a $999 priced 8-core Sandy Bridge processor intended for single socket systems, as the leaked slide indicates this price point will be initially occupied by a 3.3GHz 6-core. Not that we won't see 8 core processors, Intel have suggested (maybe even said) there will LGA 2011 Sandy Bridge 8-core processors.

 

[deconstruct60]

My statement meant that I don't feel we will see a $999 priced 8-core Sandy Bridge processor intended for single socket systems,

The blocker is more the price point. If folks would pay $1,200-1,300 for a 8 core option it wouldn't have to wait.

 

[Chris7]

The issue wasn't with Intel's engineering. Apple either had a deal with Intel for big discounts on Xeon processors run out (for transitioning to them) or they wanted more margin from the Mac Pro. They went from using two $700-$800 processors in their 2006 and 2008 base models, to using a $300 processor; or on the DP model, which now cost $500 more, two $375 processors.

It made less sense to use the UP processors before, there were far more differences in the LGA 775 and LGA 771 platforms and chipset features before LGA 1366 came along...

I had actually forgot about the cheaper chips and higher base prices on the Mac Pro with the Nehalem. I now remember there were a couple threads on this at the time. I think that's when I started seeing how ruthlessly brilliant Apple's marketing really was.

I wonder if the new iMac will put pressure on Apple for a more reasonably priced Mac Pro, come Sandy Bridge server time. When the Nehalem started shipping on the Mac Pro, I believe Apple had more effectively crippled the iMac -- the 4 core iMac's were not out yet, and there was only one FW800 port, and no ESATA. Now the iMac could work well for some video pros (with an external video monitor and an inexpensive converter, such as a Matrox MXO2 Mini), so long as they do not require a color accurate main screen for serious still graphics work. (For those here who are not into graphics, video pretty much requires TWO fast ports: one for imputing video, and second for looping it onto an external hard drive. One of the two ports is usually much faster than FW800, for use with a RAID drive).

At least the current iMac is more than enough for incredibly easy to run and somewhat underpriced broadcast standard NLE, Final Cut Pro 7 (which, BTW, I believe continues to exist primarily to sell overpriced Mac Pro's, and secondarily to keep Apple's "for creative professionals" marketing image for other markets).

...to be clear, I say that because enterprise, manufactures and the likes of Intel, AMD and Nvidia have agreements in place that hardware won't become obsolete within certain time frames. Intel also control their market pretty much so can dictate releases and a year is a good length of time to sell a lot of product without having buyers holdout....

Makes sense to me. Sandy Bridge server chips Q4 2011/Q1 2012, and Ivy Bridge Server ships Q4 2012/Q 2013.

...The problem with the single package models are that more cores tend to be more expensive and that usually drives a perception "need" more GHz also. However, to keep more than 6 cores feed a higher clockspeeds you'll need more than 4 memory channels. That's one reason why the non "desktop" oriented processors models then to couple more cores to multiple package configurations. You need to feed the beast. Otherwise what you have is a long stream of no-ops because pumped through the processor while it waits on many workloads...

I had not realized this before your post -- thanks. I see that on the Wik Sandy Bridge site, the 4, 6, and 8 core server chips will have a max of 4 memory channels. So it makes sense that there would be another dual 4 core somewhere in the line, with a (RAM starved?) dual 8 core probably offered as a more of a novelty upgrade -- too expensive and impractical to sell many, but makes look hefty for offering it.
Makes me wonder if the 6 core will have a RAM bottleneck with After Effects CS5, an application that makes extremely efficient use of all cores and large amounts of RAM (like over 32GB).

At any rate, thanks for everyone's time...

 

[sammyman]

Based on the leaked processor details and likely improvements from the next gen Mac Pro, can anyone guess what kind of improvement we would see as far as performance is concerned?

 

[theSeb]

Based on the leaked processor details and likely improvements from the next gen Mac Pro, can anyone guess what kind of improvement we would see as far as performance is concerned?

I would estimate about 1000 quarks of performance.

 

[deconstruct60]

I wonder if the new iMac will put pressure on Apple for a more reasonably priced Mac Pro, come Sandy Bridge server time.

No, it won't. There is general pressure from the overall PC market for a cheaper mini-tower box. However, that isn't new; that pressure has already been present.

More than a few of those potentially "former" MacPro buys were not really MacPro buyers. They would have been more satisfied spending iMac kind of money ( $1,500-1999) but got pressed into buying a MacPro. There is no need for MacPro to try to chase after those folks buy lowering the price. Some of those folks were bolting into Windows mini-towers right now. The new iMacs will probably capture more of those folks that were not going to buy Macs. That was the bigger hole to plug.


What Apple needs to do is make sure the Mac Pro has more value features than the iMac for those who have the budget to spend $2,500+ on a box. It still can rest in I/O , just not to low end (common 2010 speeds) disk expansion.

1. PCI-e v3.0 on the MacPro. That would put some distance between TB equipped Macs and the MacPro (whether it gets TB or not). TB is stuck on PCI-e v2.0.

A diagram in at least one article suggests that these new server class Xeons will get PCI-e v3.0 In the "Romley Platform" diagram here : http://www.cpu-world.com/news_2011/2011050402_Intel_Xeon_E5-1600_and_E5-2600_processor_details.html there is a indication of "up to 40 PCI-e Gen 3.0 lanes"). That is indicate that really the X58 effectively got absorbed by the Xeon package, not pushed into a another "Southbridge"-like chipset (e.g., X79 rumors ).

http://en.wikipedia.org/wiki/File:X58_Block_Diagram.png

The primary duty of the X58 was to provide PCI-e lanes and to provide DMI conduit to "Southbridge"-like support chip. The new diagram suggest that E5s do that all by themselves ( yet another piece of functionality can squeeze in when don't have to implement a "fast" GPU. )


2. Better Video cards with less drama. The iMac tops out at 2MB of VRAM. People with apps that push lots of data to the VRAM will still get more value out of a MacPro. It would also help if there were a wider variety of cards that worked with MacPro. ( Note: PCI-e 3.0 can possible help here also with either a GPU + TB combo card option or just GPU card with just that more bandwidth. ).

Also can get RAID and Network cards with higher bandwidth.


3. Soak up part of the XServe market: rackable case , lights-out management, etc.

Get more value out of more than 4 cores: more parallelized software (FinalCut X ) and OS support, etc.

More diverse I/O without additional external dongles ( e.g., USB 3.0 )


The Mac Pro can be priced higher if it offers more value. Most people will pay more to get more value. The problem of the overall PC market prices going down is present for the whole Mac line up. Cheaper MacPros won't come until cheaper iMacs do which in turn are held up by cheaper Mac Minis ( and iPads ).

What Apple could do to incrementally help would be to start the Mac Pros off in the $2,100-2,200 range. Since the iMacs standard configs stop at $1,999, MacPros can climb down a small amount without breaking their "no price overlap" law. Right now they provide the 27" i7 option a bigger window.

(For those here who are not into graphics, video pretty much requires TWO fast ports: one for imputing video, and second for looping it onto an external hard drive. One of the two ports is usually much faster than FW800, for use with a RAID drive).

You don't need two. What you need is enough bandwidth. For HD workflows one TB controller will work. And it isn't two ports. Two FW ports that operate off the same controller have the same bandwidth as just one port. For example, the new 21" and 27" iMacs have exactly the same TB bandwidth. There is more connection convenience with the two port 27" set-up but not more bandwidth.

In practice with older MacPros you needed an additional FW/eSATA/etc. in a PCI-e slot to get the incremental step in bandwidth. However, that confusing the implementation (card/port) for the root cause problem (bandwidth).

At least the current iMac is more than enough for incredibly easy to run and somewhat underpriced broadcast standard NLE, Final Cut Pro 7

Depends. If there is centralized storage it still may be easier to use a MacPro than an iMac. For example tapping into legacy XSan or a 10Gb Ethernet set up. Likewise FCP 7 is core choked. The iMacs max out at 4 so not an issue now. In the next generation it will be a larger differentiator.

Sure some users will peel off into iMacs. But it is bigger issue to better compete head-to-head with the other $3,000+ workstations and their workloads.

 

[deconstruct60]

Based on the leaked processor details and likely improvements from the next gen Mac Pro, can anyone guess what kind of improvement we would see as far as performance is concerned?

In the 20% range on average. check the Romley platform graphic here http://www.cpu-world.com/news_2011/2011050402_Intel_Xeon_E5-1600_and_E5-2600_processor_details.html

That's probably an average over a set of different applications. It will probably be application specific. If you are running some 8 year old Pentium optimized binary it will often be lower. If have AES streams that need real time encryption it will likely be higher.

The Westmere-era vs. Sandy Bridge have also seen approximately 20% jumps.

For folks sitting on a 5+ year old MacPro or G5 it isn't as big of a releative jump between current MacPro (if have to buy now) and the upcoming one. On the other hand if trying to rationalize an upgrade of a 1.5 year old machine it is probably worth the wait if you can.

 

[sammyman]

In the 20% range on average. check the Romley platform graphic here http://www.cpu-world.com/news_2011/2011050402_Intel_Xeon_E5-1600_and_E5-2600_processor_details.html

That's probably an average over a set of different applications. It will probably be application specific. If you are running some 8 year old Pentium optimized binary it will often be lower. If have AES streams that need real time encryption it will likely be higher.

The Westmere-era vs. Sandy Bridge have also seen approximately 20% jumps.

For folks sitting on a 5+ year old MacPro or G5 it isn't as big of a releative jump between current MacPro (if have to buy now) and the upcoming one. On the other hand if trying to rationalize an upgrade of a 1.5 year old machine it is probably worth the wait if you can.

Thanks that helps. I will be upgrading from a Mac Mini, but I think I can wait for a few more months. I just hope I am not waiting until January or later like some people are suggesting in this forum.