Got a tip for us?
Let us know
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.
Mac Pro 2011 "Sandy Bridge E" everything we know
- Thread starter Amethyst
- Start date
- Sort by reaction score
http://www.anandtech.com/show/5458/the-radeon-hd-7970-reprise-pcie-bandwidth-overclocking-and-msaa
Thunderbolt is equivalent to PCIe 3.0 x2.
The difference isn't all that big, especially if we talk about 7750 (much slower than 7970). It would still be a lot faster than the IGP, which is enough for most users.
It's not. It's equivalent to PCIe 2.0 x2.
PCIe 3.0 has much faster bandwidth (doubled according to Wikipedia.)
In addition, Thunderbolt itself adds overhead.
These sorts of benchmarks are also misleading because they're not that dependent on throughput. Change it to something like a video editing benchmark, and you're going to see a lot worse numbers.
And as far as replacing the Mac Pro, the 7750 isn't even a high end GPU, it's a mid end, and it's still seeing significant performance degradation in some spots.
Last edited:
I think we can agree on one thing:
TB is NOT a suitable replacement for PCI-E in a pro desktop.
Not only that, if people are running RAID cards, displays, GPUs and god knows what else then TB will be saturated in no time at all.
TB is for external IO connection only. It is no replacement for PCI-E (in it's current state anyway...)
TB is NOT a suitable replacement for PCI-E in a pro desktop.
Not only that, if people are running RAID cards, displays, GPUs and god knows what else then TB will be saturated in no time at all.
TB is for external IO connection only. It is no replacement for PCI-E (in it's current state anyway...)
I was interested in a thin/light notebook with Thunderbolt. I would like some more power to play games and have an open expansion slot vs. a sealed BTO break out. The HD 77xx launch is this week and more information points to even single slot cards in the 60-80W range.
It is going to require supplemental power regardless.
On the other side I am tempted to get a X79 desktop once Windows 8 is out.
You're assuming both Thunderbolt lanes can be used for one GPU. They may have to show to the host as a discrete devices. Basically the host sees 2 10Gb/s PCIe slots, not one 20Gb/s slot. Depending on the Thunderbolt specs, the second lane may even be required to be passthrough.
Really what you're saying is that you can run a GPU off of two 2x PCIe 2.0 connections. Which is... maybe possible? But you're comparing 4x PCIe 2.0 to 2x PCIe 3.0.
Two things to consider:
- It's only 10Gb/s per direction though (1.25GB/s), as where PCIe can push 2GB/s in either (4x Gen 2.0 lanes or 2x Gen 3.0 lanes).
- More importantly, TB adds overhead that lowers it's throughputs per direction to ~800 - 850MB/s under real world conditions (Intel's own data as well as from Promise's published data on the Pegasus R6). So even if both directions are saturated, their combined throughput is less than 2GB/s (1.6 - 1.7GB/s for TB at it's current spec).
It's still the fastest interconnect on a laptop to date, but not for desktops with PCIe slots.
Go and do some research first.
The Xeons have not been released yet.
Go and find me a SB-E E5-* Dell workstation. (Tip: You can't!)
Why can't people read?
Go and do some research first.
The Xeons have not been released yet.
Go and find me a SB-E E5-* Dell workstation. (Tip: You can't!)
Why can't people read?
That's because the Xeons are all Sandy Bridge-EP!
Pedantic silliness aside, Fujitsu's Polish site seems to be the first live page with details and order information for an E5 workstation - the M720. This appears to be a single socket workstation, but with all the E5 processor options available.
The shipping date is 2012-03-08.
Order page:
http://konfigurator.fujitsu-shop.pl/en/celsius/celsius_m720/S26361-K1399-V115/conf.html
System board D3128, Chipset: Intel C600; Theseus system management controller, 800 W wide range power supply, optical USB tilt- wheel-mouse;
prepared for rack mounting (rack mount kit optionally available), DeskView license included.
Slots:
8 DIMM-slots, 4 channels(DDR3 1600Mhz) ;
7 expansion slots: 2x PCIe Gen3 x16, 1x PCIe x4 Gen3 (mechanical PCIe x16), 1x PCIe x4 Gen2 (mechanical PCIe x16), 1x PCIe x4 Gen3 (mechanical PCIe x8), 2x PCI (32-bit / 33 MHz)
Onboard I/O devices: 1x 1Gbit network controller Intel 82579LM , SAS 2.0 controller as an option via UpgradeRom (4 Ports, RAID 0, 1, 5 and 10 ), SATA 3 Controller (6 Ports), HD-Audio 5.1 CX20642
Go and do some research first.
The Xeons have not been released yet.
Go and find me a SB-E E5-* Dell workstation. (Tip: You can't!)
Why can't people read?
I didn't state workstation class processors that the Xeons are; I merely mentioned enthusiast ie i7-3960X/3930K...
Last edited:
Why does a processor which won't be used in the Mac Pro have any affect on whether or not it will get a refresh? Mac Pros have always only used Xeons, and the Sandy Bridge-EP ones aren't out yet.
4x PCIe 2.0 is the same as 2x PCIe 3.0 in terms of bandwidth. We can look at PCIe 2.0 x4 and it won't change the data, but I thought I would choose the latest GPU that should actually show if there is a difference.
http://www.techpowerup.com/reviews/NVIDIA/GTX_480_PCI-Express_Scaling/24.html
My point is: You can run AMD 7750 or any GPU on Thunderbolt when/if it becomes possible. How much the performance suffers is another question but as you can see in the graphs I linked, it's software-dependent. Some games take no hit at all while some take a big hit.
Thunderbolt is bi-directional so it can push 10Gbps in both directions at the same time.
http://www.anandtech.com/show/4489/promise-pegasus-r6-mac-thunderbolt-review
Nope.
Of course it is, you can't magically add bandwidth
How much is the question but I still find it a bit presumptuous to say that Thunderbolt is too slow for a certain GPU. You can run any GPU off of it, heck even SLI/CF, but whether it's worth it over a cheaper GPU is another question.Sure, you can run any PCIe card on Thunderbolt, but the performance is kind of the key, isn't it?
I'm not sure why we're talking about PCIe 2.0 x4. Thunderbolt is PCIe 2.0 x2.
You're assuming you can use both Thunderbolt lanes on a single device, which I don't think has at all been proven, and even so likely wouldn't work without making decent modifications to the card. If you use both lanes, the host is going to see two different devices, not a single device.
One channel is PCIe 2.0 x2. TB has two channels so that's (roughly) PCIe 2.0 x4. If AMD 7970 does fine on PCIe 3.0 x2 slot, then AMD 7750 will be more than fine even on PCIe 3.0 x1 (appears to have about 1/4 of the performance of AMD 7970 when looking at raw figures).
I know I'm assuming. However, Thunderbolt is based on PCIe and there are up to 16-lane PCIe slots. Of course we won't know anything for sure until actual devices start shipping but even if they were limited to single channel, something like AMD 7750 would work there just fine.
It's not like we are talking about workstation replacements anyway. The only real market I can see is Ultrabook owners who are limited to Intel graphics. Otherwise it's better to just buy a model with better dGPU because TB isn't fast enough for a higher-end GPU.
You are assuming here as well. PCIe devices don't show up as multiple devices even if they use more than one lane. I don't think this would be an issue, we have SLI/CF that enable the use of multiple GPUs so something like this shouldn't be a problem. OS X sucks for these kind of things anyway so it would most likely be Windows only.
Monitor PCIe bandwidth in realtime?
As an interesting coincidence my topic fits nicely to the current discussion (don't want to hijack this thread):
Is there a monitor tool to measure the bandwidth usage of PCIe slots in realtime? Or at least a realtime monitor tool to measure the bandwith usage of individual cards currently installed and used (e.g. graphic, RAID etc.)?
The usual suspects (iStat & Co.) offer to measure CPU load, network load and much more, but so far i did not find an option for internal bandwidth usage...
As an interesting coincidence my topic fits nicely to the current discussion (don't want to hijack this thread):
Is there a monitor tool to measure the bandwidth usage of PCIe slots in realtime? Or at least a realtime monitor tool to measure the bandwith usage of individual cards currently installed and used (e.g. graphic, RAID etc.)?
The usual suspects (iStat & Co.) offer to measure CPU load, network load and much more, but so far i did not find an option for internal bandwidth usage...
Who said anything about them being used in MacPro's? I just said I can't see it happening...
----------
Besides....
If a refresh occurred this year, would this be Apple's first new machine since Steve's unfortunate departure? I wonder what the team will do... I'd love to upgrade my current Mac, she's getting on past retirement
I know that.
Keep in mind however, it can only run bi-directionally under certain conditions:
- Is the end-point technology it's connected to capable of bi-directional communication simultaneously?
- How many other devices are connected at the same time?
- How are they being accessed (% of read v. write)?
Now if you've multiple storage devices (or other types of products attached to TB and being used simultaneously), there is the possibility that you'd be able to utilize TB's bi-directional capability simultaneously. But it will depend on the specifics every time a TB device or group of TB devices are being used.
But statistically speaking, most will likely only benefit from one direction at a time, which makes the 10Gb/s throughput limit of a single direction more important than it's combined throughput.
This is why I suspect Intel has published the 10Gb/s figure rather than advertising 20Gb/s (rather surprised at this actually, as marketing depts. would usually latch onto the higher value). Which lends me to think that someone in management was paying attention, and didn't want such a "stretching of the truth" to cost them in the long run.
You're forgetting the effect of cache.
The testing I saw was based on the drives they shipped in the unit, which were Hitachi
They're certainly not shoving SSD's into it for the base MSRP.
As per the details of the Anandtech article, we can go into that if you want, but here's a hint:
Each 10GB/s lane, even though is listed as bi-directional, can only run in a single direction at once, just as is the case with many other bi-directional technologies. Strapping dedicated lanes is required, and is also the case with TB.
For example, if you're running a single TB port bi-directionally, one of the 10Gb/s lanes will be used for up traffic, and the other for the down traffic. They could have made it more robust in regard to how it handles traffic, but that would have added complexity, and therefore cost to a product that was already over budget, and cut the optical portion of the spec, primarily to keep component costs within a range that vendors/board makers will bite.
Maybe I didn't understand your original statement then:
nanofrog said:
To me, it sounded like you suggested that TB is one-directional while PCIe is bi-directional, which is not true. Your above storage example should apply to PCIe as well, in that case. I know bi-directional is all that important in real world and it most certainly doesn't double the bandwidth.
You're forgetting the effect of cache.
The testing I saw was based on the drives they shipped in the unit, which were HitachiDeathstarsDeskstars at that time.
They're certainly not shoving SSD's into it for the base MSRP.
I only tried to show that TB is capable of more than 800-850MBps which is what you claimed. We should be talking about the limits of TB, not about the limits of Deathstars
It's not a surprise that you aren't getting more than ~800MBps from RAID'ed HDs, especially Hitachis. The actual interface is capable of more, though (about 1GBps).
http://www.techpowerup.com/160456/EVGA-SR-X-Dual-Socket-LGA-2011-Motherboard-Teased-Some-More.html
I really want to see the block diagram for this baby.
I really want to see the block diagram for this baby.
Attachments
I wouldn't wait for Apple ... MacPro update isn't going to happen. Sales numbers have nothing to do with Apple's decision to drop the MacPro ... Apple make good profits from the MacPro line, even more profit from additional components they sell for the MacPro. Profit is profit regardless on quantity sold ... in the case of the MacPro Apple are making some great profit margins.
There have been new CPUs running at higher speeds, but Apple has just ignored them still stuck at 2.93 Ghz.
There is nothing special about a Xeon or any other "i" series Intel processors -- they all run the same instruction set and Apple OS does NOT need to be changed. Apple requirements to upgrade MacPro is really only a matter of case design (minor), power consumption (minor), heat (minor), chipset, and an EFI update (minor). Apple have been doing Intel CPUs/Motherboards for a long enough time to understand the release of new hardware ... it's not a difficult task (look how quickly they got Thunderbolt out on iMacs) and much easier than the regular releases of iPhones and iPads (whicih is a lot more difficult than any Mac computer release).
Remember it used to be Apple Computer Inc. -- now it's just Apple -- the "computer" aspect of the company got dropped a long time ago.
As far as waiting for an official Apple announcement that the MacPro is dead, why would they do that? They're under NO requirement to announce the death of a product line ... they'll just continue to not upgrade it and flush out their remaining stock. You can't find a MacPro "in stock" anywhere.
Apple is placing everything into Mobility.
And if you still want to "believe" there will be another MacPro, ask yourself this ... do you believe a company that has surpassed both Google and Microsoft financially would NOT have the resources to produce regular updates to the MacPro line? Apple have let go, I think you folks need to start to accept that and don't wait for some "official" news from Apple ... just not going to happen.
The EVGA board supporting a Core i7/ E5-1600 Xeon if only using one socket thing is interesting - if accurate.
----------
Just because they could have made minor updates to the line doesn't mean it makes financial sense to do so outside of a major platform revision. Apple's development costs for a new system are surely higher than other workstation vendors due to the software and driver side of development and not only are their sales lower anyway, but nothing they could have done would really have driven sales a lot.
----------
Just because they could have made minor updates to the line doesn't mean it makes financial sense to do so outside of a major platform revision. Apple's development costs for a new system are surely higher than other workstation vendors due to the software and driver side of development and not only are their sales lower anyway, but nothing they could have done would really have driven sales a lot.
Back when the Power Mac line was their bread and butter they didn't regularly update.
Seriously, take a time machine back 13 years and you'll find people complaining about how Apple still hasn't updated the Power Mac G3 to come with a Radeon instead of a Rage 128.
Not doing minor spec bumps is normal for Apple.