Luminous Landscape Forum
Equipment & Techniques => Computers & Peripherals => Topic started by: John.Murray on November 14, 2011, 10:10:41 pm
-
Finally in the channel:
http://www.anandtech.com/show/5091/intel-core-i7-3960x-sandy-bridge-e-review-keeping-the-high-end-alive
http://www.anandtech.com/show/5089/sandy-bridgee-and-x79-the-asus-p9x79-pro-review
Note that these are *not* Ivy Bridge, there is no native USB 3, or Thunderbolt. I'm awaiting Intel's reference ZX79SI board along with a 3930K CPU.
What's exciting to me is an effective doubling of memory bandwidth over either X58 or Z68/9 chipsets. It will be interesting to see how upcoming PCI-3 devices utilize this...
-
This seems to occupy that last niche where the 990/970 was indicated.
It's always nice to have choices.
-
Exactly!
Parts arrived this afternoon, including 32GB (8x4GB - DDR3/1600, under $200). I'll let you know my impressions compared to my X58, in the next few days....
-
Have you heard of any x79 motherboards with dual LAN ports yet? Or PCIe v3?
32g's.. really interested in how that meshes with your workflow. I'm assuming 2133? With the new BIOS's supporting the fast RAM via a profile and no OCing necessary I see this as an advantage.
Eventually I'll upgrade my own workstation (x58/950).. but probably with Ivy Bridge. It's more fun spending other peoples money.. :)
-
The Intel Board has dual nics - but sadly, no teaming yet. PCIe-3 has actually yet to be finalized, but Intel will support it via firmware update.
For those that don't know, PCIe-3 at 8GT/s will double bus throughput over the existing PCIe-2 (5GT/s) Before you complain about my math - PCIe-2 has an encoding scheme that results in a 20% overhead, PCIe-3 uses a different encoding scheme with far less overhead - hence the effective doubling of bandwidth.
With both the X58 and X79 having on die memory controllers - the interleaved memory access scheme (triple for x58, quad for x79) tends to wash individual memory timing parameters - this allows us to be a bit more conservative with minimal impact on overall performance.
I'm holding out on Ivy Bridge as well, with Thunderbolt and Xeon (ECC memory) support.
In any case, given a doubling of memory throughput and a doubling of bus thoughput - this platform has some serious legs
-
PCIe v3 is important for some of my builds, not so much for the GPU, but because I'm using 2 PCIe RAID cards in some of these systems. When you add those to the GPU's the demands can exceed v2.
From what I can tell x79/-E setups are being built exclusively for gamers which isn't new.. but usually Asus and Gigabyte make some decent workstation MB's to round things out. I haven't seen those yet from the x79 offerings.
The memory and PCI bandwidth increases are going to be great.
-
PCIe v3 is important for some of my builds, not so much for the GPU, but because I'm using 2 PCIe RAID cards in some of these systems. When you add those to the GPU's the demands can exceed v2.
X79 offers additional PCI lanes giving you the best of both. The PCIe spec defines per lane performance
In other words, you can populate the bus with a couple video cards, and still have additional lanes available for disk/io.
-
Bumping this as a hopefully relevant thread to garner input from both John and Steve.
Ivy Bridge - it's close (from what I've seen main boards have been demoed and such). I'm holding out upgrading until it's released and that's prompted a few questions.
I'm almost certainly going to go with a Revodrive solution - is there any talk about PCI-3 versions? Would it be a significant improvement? Mostly, would it be worth waiting for if not available when Ivy Bridge itself comes out?
The proposed CPU line up I've seen for Ivy Bridge only shows Quad core, whereas we have the 2011pin Hex core processors already. Is there any talk about hex core for Ivy Bridge?
Any other comments/thoughts you guys (or anyone) has about upcoming Ivy Bridge related upgrades/updates for other technologies? If there's a good link to get up to date, happy to be directed there, but also interested in your personal thoughts.
-
Bumping this as a hopefully relevant thread to garner input from both John and Steve.
Ivy Bridge - it's close (from what I've seen main boards have been demoed and such). I'm holding out upgrading until it's released and that's prompted a few questions.
I'm almost certainly going to go with a Revodrive solution - is there any talk about PCI-3 versions? Would it be a significant improvement? Mostly, would it be worth waiting for if not available when Ivy Bridge itself comes out?
The proposed CPU line up I've seen for Ivy Bridge only shows Quad core, whereas we have the 2011pin Hex core processors already. Is there any talk about hex core for Ivy Bridge?
Any other comments/thoughts you guys (or anyone) has about upcoming Ivy Bridge related upgrades/updates for other technologies? If there's a good link to get up to date, happy to be directed there, but also interested in your personal thoughts.
This is a bit like trying to guess the specs for the new 5d Mark III.. in that most anything said is nothing more than an educated guess at this time.
If you go by past releases of major CPU's they've always started with the main quad, then released the lower cost options, and then a hex to keep people excited and buying until the next major cycle is on the shelves. I don't blame them for doing this, it keeps people talking and buying their products.
I think where we as consumers make our mistake is in thinking we need the hex core, or the next generation Revo, and we sit and worry over what's going to come 'some day', where we could be using the new stuff that available now and benefiting from the really significant upgrade from the last product cycle. I'm being loose with the figures, but to illustrate I estimate we'll get 80-90% of the capability of that product cycle with the first release products, if we wait till the end maybe we'll benefit another 10-20%.
However, the major upgrade is still from the last cycle products to the new cycle products. As an example, plenty of people are using the very first i7-920 CPU from the last cycle and it works great. The i7-960 came out nearly 20 months later and offers what, a 8-9% increase in CPU power? The 970 and then 980 Gulftown at twice the price offers two more cores which are really only beneficial for those who render a lot of video. And of course their halo product the 980/990 Extreme Hex offers 10-14% above that.
The wise money for the vast majority of us, is that initial i7-920 CPU because we can start benefiting from it's power right away, we really don't need the hex cores, and who wants to wait for 6 then 12 then 18 months of each incremental upgrade to the 920? Sure, if we're building a new system we spend the extra $20 and get the 960's.. but if they were out of them I wouldn't feel I'm giving up a significant bump by going with the older 920/930/950's.. Either one is going to serve us well.
I'm not saying the faster stuff won't be more 'neat' to play with, I enjoy building clients machines all the time so I can play with the latest and greatest. But I'm not running out to buy one for myself because I still have that initial 920 and I'm running a business, I want my full ROI and logically I know I'm not going to get a 'seat of the pants' improvement to make it worth upgrading.
I hope you followed that.. sorry it was so disjointed.
And the same goes for the Revo's. These are OCZ's consumer halo product and they're in the habit of leading the industry, so you can bet they'll be finding a way to take advantage of every new interface and feature they can, as soon as they can, but you're probably not going to feel a 'seat of the pants' increase in performance between the latest and greatest.. and the previous model. You'll be able to measure a difference, but not necessarily feel it.
I'll also be upgrading to Ivy Bridge. This is where I planned my next upgrade along with a new Platinum Seasonic, Revo's (The fastest available revo for system/programs, and hybrid (maybe two depending how it performs) for a work drive), a bunch of RAM, and maybe.. maybe.. I'll upgrade my ATI 5970 dual GPU card as well. But once I do, I'll not be upgrading until the next product cycle. I know I'll be building a lot of machines for clients, but there's no significant performance or financial ROI to drag me into it.
And it's not entirely about money or performance, none of this stuff is so expensive that we can't indulge if we choose. It's just a PIA and the mental addiction part makes want to be the master of my own domain.. ::)
-
Thanks for the feedback, Steve. I agree with what you're saying. I'm in a situation where I've started to feel a lack of performance but I reckon I'm close enough to wait for Ivy Bridge (even though I can obviously get some Sandy Bridge-E goodness for quite a reasonable price!), but I don't want to wait too long. I'd hold off maybe 2 months after release if I knew something particularly nice was coming down the pipe (and I'll wait a month anyway just to see if any serious bugs show up).
With Revo I was wondering whether we're seeing a cap in performance due to the bus speed (and therefore PCI-3 might be a benefit).
Looking at your comments on CPUs and checking now some history, you're spot on and that does make sense. You're right, of course, that I don't need 6 cores but if it were available I'd go with it.
Assuming Gigabyte release something the same as their current X79-UD5 board, I'll go with 32GB RAM because the price is so attractive. If not, I'll be happy with 16GB (8GB sticks seem far more expensive/difficult to obtain easily here). Combined with a Revo I don't think I'll be lacking any speed. Price is not an issue, because this isn't a business machine (and hence vanity can write the cheques) but timing is - as you say, you need to jump in at some point!
-
Phil: Remember that Ivy represents a move from 32nm to a 22nm process - a huge deal.
Starting around 2006 Intel has been on a "Tick Tock" development cycle - The "Tick" being first release of the new process - this is typically taking existing features forward. The "Tock" bring additional new features (often including chipset or even form factor).
Socket 2011 represents the "Tock" that began with Socket 1155 Westmere/Gulftown "Tick", and ended in Sandy Bridge - all on a 32nm die
Ivy bridge will introduce Trigate transistors that, according to Intel will result in a 20% increase in performance over equivalent 32nm dual gate offerings, this with reduced generated heat (read - higher CPU clock Speed / More Cores). I'm not holding my breath for any 6 Core socket 2011 CPU's anytime soon. In fact you may even need to wait until Haswell (the "Tock" of the 22nm process).
Why do I think this? Intel is really being challenged in the Data Center for cost effective (read power consumption) solutions. Big providers (Amazon/MS/Facebook) are seriouslly looking at ARM platform alternatives; Windows 8 will support it... They appear (at least to me) to be addressing this pretty agressively - look for Xeon offerings ahead of desktop during this "Tick" phase.....
As soon as Thunderbolt ships this year -I'll make my jump
-
Thanks, John, for your additional points and info. Thunderbolt does seem to be taking off and I'm hoping it will make its way into mainboards early on. Everything seems to be pointing toward a good time to pull the trigger being 2-3 months after Ivy Bridge release for me.
Yeah, everything that I've read seems to suggest that Ivy Bridge brings less heat, less power and more performance and pricing remains attractive (particularly as my local currency is holding very well against the USD). My mind boggles at where they'll be trying to head after 22nm (given the previous jumps that we've seen). The "3D" transistor architecture looks to be an interesting move with plenty of scope for giving improvements without having to reduce the die (although I'm sure they'll continue to do that).
-
With Revo I was wondering whether we're seeing a cap in performance due to the bus speed (and therefore PCI-3 might be a benefit).
Looking at your comments on CPUs and checking now some history, you're spot on and that does make sense. You're right, of course, that I don't need 6 cores but if it were available I'd go with it.
Assuming Gigabyte release something the same as their current X79-UD5 board, I'll go with 32GB RAM because the price is so attractive. If not, I'll be happy with 16GB (8GB sticks seem far more expensive/difficult to obtain easily here). Combined with a Revo I don't think I'll be lacking any speed. Price is not an issue, because this isn't a business machine (and hence vanity can write the cheques) but timing is - as you say, you need to jump in at some point!
1. Currently no, but I wouldn't be at all surprised if OCZ comes out with a future Revo which maxes out PCIe2 and works faster in PCIe3. Where have we seen this before.. hmm... Oh ya, SATAIII SSD's.. :)
2. Hex cores are nice, but with the current state the software video rendering is the big winner for six cores. For a 100% price increase for the hex and probably a significant wait (6-12 months I'm guessing) after the initial release of Ivy Bridge.. I won't be getting one.
3. You can never have enough quality RAM. And it's currently priced where it's not a big deal in a price build to go 16 or 32..
4. With Ivy, 32gb of RAM, and Revos for system and work drives.. your bottlenecks will probably be settings. Be sure to get someone knowledgeable to setup and configure your system.
Good luck.
-
My mind boggles at where they'll be trying to head after 22nm (given the previous jumps that we've seen). The "3D" transistor architecture looks to be an interesting move with plenty of scope for giving improvements without having to reduce the die (although I'm sure they'll continue to do that).
Upcoming is 14nm (Broadwell - 2014) and 10/11nm (Skymont? - 2016?)
And I totally agree w/Steve (as usuall), what used to be really CPU intensive stuff is now being handled very gracefully by multiple cores or by offloading to a GPU (ie: Premiere / nVidia CUDA). With 8-core CPU's becoming available, having a single process effectively use all available cores is pushing Amdahl's law ( http://en.wikipedia.org/wiki/Amdahl%27s_law#Relation_to_law_of_diminishing_returns ). The bottlenecks remain RAM and your Disk subsystem
-
Steve:
Yeah, SATAIII SSD was in the back of my mind - it's like the gas theory - manufacturers will expand the performance of their products to meet the available bandwidth :-)
That said, other than for benchmarking, just how much difference will it make at the moment? If we start using 200MP cameras, a different story!
Regarding RAM - here in Oz I find it difficult and expensive to find 8GB matched sticks whereas 4GB matched sticks are cheap and easy to get. So 32GB means a mainboard (like the Gigabyte X79-UD5) with a total of 8 slots.
Also, the whole "matched sticks" thing - how important is that really? I've done it in my last couple of builds and certainly haven't had any problems but if I'm talking 8 slots I imagine price goes up to get a pack of 8 matched (assuming I can find such a thing) or at least 2 packs of 4 matched. To be honest, since as I said vanity will be the accountant, if I get an 8 slot board I might even look at 64GB but it boils down to being able to find the right RAM. It might just be too much trouble.
Setup and configuration I'm OK with - I've been building my own for a long time (these days I get someone else to put it together) and I always do my research before hand (like useful posts on Thai websites about how to install a Revo... :-)
John:
Thanks for that info - interesting, but not unexpected, that the jumps are smaller. What sort of technology wall in manufacturing are we hitting here?
To you both - thanks for your generous sharing of knowledge and opinions as always - much appreciated.
-
Also, the whole "matched sticks" thing - how important is that really?
Matching ram is important for a motherboard level on first, and then on a controller level.
You see, even if two brand use the same IC's of DRAM, they build it differently. I'm thinking about Corsair and Kingmax. They both use Elpida IC's, but somehow Kingmaxes rams can be pushed further and have better compatibility with motherboards than Corsairs. I've seen many Corsair memories just randomly crash around here, while others not. Some improve read, other copy, other just try to get the better IC's for faster speeds. Then there is the way they are built, single sided, double sided. Single sided means higher density chips, while DS means lower density. Lower density can OC better, while single is more stable under normal usage.
Then the controller comes into play. There are some controllers that cannot handle many sticks, or they can at higher lantencies or lower clock speeds. And it all boils down on the capacity. The higher the density, the more read/write the controller has to do.
Intel quad channel controller is more relaxed when working with ram, compared with the dual channel ones. But, higher capacity ram means higher latencies, especially if you want speed.
So it's important to know what you want. If you want speed over capacity, get two kits of 8gbs at around 1600 mhz, but lower lantecy. If you want capacity over speed, get 32 GBs kits since they became available.
-
Thanks for the info.
When I say matched, I don't mean from the same vendor - that will definitely be the case - what I mean is that vendors sell matched sets of sticks and I'm curious as to how much of a necessity or advantage that really is, particularly if I start looking at an 8 slot main board.
-
With respect to matching ram;
This is pretty important. Maybe not buying in the same blister pack, but you need to be sure that sets of ram are the same revision. Sticks of ram get updated boards/chips just like any other PCB during their production lifespan. Order from a reputable supplier and tell them you need two sets of the same version to match and all ought to be well.
I had this problem with some ram a while ago. I added an extra set a few months after the initial build, apparently the same spec, but, unbeknown, different board revisions. Everything worked fine for a while, but then problems started that were very difficult to diagnose. Eventually whilst swapping out the sticks to try and isolate the problem I spotted the revision difference, my dealer swapped them out for a matching set and all's been well since.
-
Ah, good info on the memory - thanks for the that - makes a lot of sense.
-
Ah, good info on the memory - thanks for the that - makes a lot of sense.
Agreed. Because I build for others and I don't want them experiencing issues I'm not there to solve, I am very selective about my parts selection. With this in mind, I only use RAM which is on the manufacturers list of tested and approved RAM. Gigabyte and Asus are both really good about publishing these lists on their websites on the motherboards main page. When I build something to play with I'll bend this rule if there's a smoking deal and the RAM otherwise meets specifications, but if for my workstation I won't. RAM can cause all sorts of issues from 'wake from sleep' not working to BSOD's..
Today I ordered the parts to build two day trader PC's.. these guys want to run a lot of monitors and I'm used to either the Eyefinity 6 on the higher end or the Sapphire Flex 5670's which can run 4 monitors, require no external power connections so they run cool and contribute less heat to the thermal flow, and so far they've been ultra reliable. Now I find they've been discontinued and I had to order a half dozen of their replacements the 6770 AND the 6870 Flex HD series.. so now we have a choice to explain to the client.. :) And I've never tested these and we don't know if they have firmware or chipset issues or any of the things a new product often comes out with. I could have tracked down some 5670's but now their price is up because availability is down and would that be serving the client best? Maybe, it's proven reliable and trouble free but it's no longer the best performance value.. Anyway, I ended up ordering the 6770's, not mentioning the pricier 6870's because I know for day trading they don't need more GPU power, and I'll test the heck out of them which will take me an extra few hours.. then I'll feel okay about using them in the future.
RAM is like this too. Lots of small variables that can make big differences. If I can eliminate some risk by buying matched sets listed on the manufacturers approved/tested list.. I will. If I can't, I know it's a potential problem area and I plan for more testing and possibly time for a return/exchange, etc..
Oh.. also trying the new Intel 520 SSD's, Seasonics new Platinum series (over the gold), a new Gen3 motherboard.. so many differences it makes me itch just thinking about it.. :-\
None of this stuff is important to the guy building his own, because he can build his own. He feels comfortable changing out parts and accepts this risk as part of the 'build your own' experience. But more often than not, the guys I build for either can't or don't want to take these risks. They want a solid reliable machine they don't have to worry about for a long time. I can't guarantee this, well.. actually I do.. but I can eliminate or reduce many of the risks.
And others who build go for the cheapest parts without regard to this.. you can save a decent chunk of change this way and they feel it's worth it. Still others by the latest parts and others the fastest parts.. each serves his personal priorities. So you have to decide what your priorities are.
Personally, I want my workstation to be reliable over all other priorities. I use a case with exceptional thermal flow, high end power supplies, drives, and other components when possible.. and I go the extra mile by using the biggest possible CPU coolers and I use Arctic Cooling GPU coolers. I'm convinced that bringing down the temp 20 degrees makes my system last 2-3 years more. So I go for the best parts, the best cooling, and very careful build practices. The mileage of others will vary.
-
Steve - other thread in this section about scratch disks. As you've used the Revos a bit - thoughts on using a smaller one (perhaps 80GB to keep price down) for scratch disk compared to RAIDing a few regular (or SSD) drives?
-
Steve - other thread in this section about scratch disks. As you've used the Revos a bit - thoughts on using a smaller one (perhaps 80GB to keep price down) for scratch disk compared to RAIDing a few regular (or SSD) drives?
This really depends on the size of your work, RAM, machine configuration, desired speed, and available space on other fast storage devices.
For instance, if you're just processing 2-3 images at a time, say limited to 3-4 layers each, and you have 16gb of RAM and an extra 30gb on your system SSD.. most people would be very happy with the results.
If you're rendering a lot of video and need a larger scratch disk, then a dedicated SSD becomes practical.
I'm on record saying I'll be spoiling myself and using a Revo 1tb hybrid for my LR indexes and previews.. and I'll also put my scratch disk on the same drive. The hybrid is perfect for LR where you'll be accessing the same galleries/collections as you work, and a scratch disk which is active.
Others will be happy to move their scratch disks to an archive disk rarely used. You don't really need a separate disk, just a disk not being used in the course of your normal work flow. Of if it is used, if it's very fast such as a fast SSD it will be fine.
I didn't respond to that thread because it seemed like they were looking for a one size fits all answer.. and while I can give you a generic answer if money and PCIe slots are unlimited, for most of us they are, and the answer depends on your machine configuration, your work flow, RAM, and other storage devices..
It's one of those areas where you have to think your way through it based on your individual factors.
I'll bet John has some good thoughts on this too.
-
Thanks, Steve. Totally understand the issue with "one size fits all". I guess I was wondering about performance of that style of access and the real world speed benefits over, say, multiple HDDs in RAID. To me, unless there's a technical reason not to use a Revo, then it's a good device for this sort of thing and would outperform a traditional HDD RAID (particularly if using software/mainboard RAID) and probably at a similar or even better price point given only one device would be needed.
-
Thanks, Steve. Totally understand the issue with "one size fits all". I guess I was wondering about performance of that style of access and the real world speed benefits over, say, multiple HDDs in RAID. To me, unless there's a technical reason not to use a Revo, then it's a good device for this sort of thing and would outperform a traditional HDD RAID (particularly if using software/mainboard RAID) and probably at a similar or even better price point given only one device would be needed.
That's the thing "real world difference" depends on work flow. Sure, a RAID is going to be faster than a stand alone like HDD, but will the work flow allow the user to realize the difference. And then consider, most people I've seen doing these types of RAID's with HDD's are using on board (on motherboard) RAID controllers which aren't great performers to begin with and second, they can take a lot off the top of CPU/RAM performance to run one. I'm sure there are some high end HBA's when paired with really fast hard drives like Raptors will outperform some of the lower end SSD's when used as a cache. But in today, February 2012, you'd have to be crazy to want to put a 4 HDD RAID array in your box and suffer the heat, noise, and power draw increases for a tiny bit of performance increase when you can get a mid-higher end SSD which will outperform it. And the Revo changes the game completely. Personally I think only the highest end workstations with a heavy video rendering work flow justifies a standalone Revo as a scratch.
Keep in mind that you can easily dual task an SSD. For instance, I stated that I was going to get a 1tb hybrid Revo for my LR indexes/thumbnails. It's a 1tb drive, of which 400-500gb will be used for that purpose. I can easily set aside 100gb of very fast storage for my scratch disc and I'm not at all concerned about performance because I won't be accessing both taskings at the same time because of my work flow. This is what I mean when I saw work flow is crucial to this decision.
On the easy end of the scale, there are many mid-grade SSD's out there for under $150.. some under $100.. that just blow the doors off a mechanical hard drive and even 2 HDD's RAID's. Most people who would blink at $150 don't have a professional video work flow which would require more than that. And for the guy doing 2-3 still images at a time and who already has a system SSD with some space.. don't hesitate to assign that space to a scratch disk with a second scratch going to your next best drive.
When you sit down with someone and they tell you budget is an issue, and they want the most bang for their buck, I ask them to detail their standard work flow. Not the high draw work flow they occasionally task.. but the work flow they use the majority of the time. Even for professionals you'd be surprised with how little they really require in the way of computer resources. Most are just caught up in the industry and forum hype and somehow think they need what some guy with more money than sense is bragging about. Not to mention all the people out there buying high-end computer systems not because they need them, but because they don't really know what they need and they try to cover all their bases by buying the top gear.
Many look at building their own systems as a way to save money, why give it to the IT guy when he's just going to make money off you. But this is like photography workshops, someone who understands photography and knows how to teach, is going to give you great value for your workshop dollar. Same with IT. An IT guy who's doing his job will very often save you money overall by matching your work flow/needs with the right equipment.. and you usually even end up with a better performing if not more expensive system because it's set up properly. You can't go around tilting windmills on every forum you see with some guy making claims you know aren't complete or accurate.. you just have to let the majority of those go.
-
Great commentary again, Steve - thanks for your time. I hope this helps the guys on the Scratch disk thread, too.
-
Great commentary again, Steve - thanks for your time. I hope this helps the guys on the Scratch disk thread, too.
When you consider the Revo's are essentially RAID 0 arrays (with an on board controller), all I'm really saying is what benefit you derive from certain hardware configurations is dependent on your work flow/load. Nothing new.
But, I do caution those considering a HDD RAID using a motherboard based RAID controller for the purpose of a scratch disk to take a good look at what SSD's are available and compare their performance. It should be obvious a low power draw cool running single 2.5 inch SSD is way preferable to an equal or lower performance 4-8 HDD array with all it's power and heat dissipating requirements in addition to a greater noise level.
There are a ton of choices, from RAID arrays made of 40-60gb inexpensive SSD's, to the same with larger capacity faster SSD's, to the Revo's.. Single SSD or HDD solutions.. all kinds of choices. Know your requirements and your work flow, and then choose wisely.
I wish you all luck with this.