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.
2013 MacPro CPU and GPU
- Thread starter MacOG728893
- Start date
- Sort by reaction score
synthetic benchmarks focused on CPU? Or benchmarks based on getting workload done for real applications?
They'll be close on synthetics. Both the top end (max BTO config) Haswell iMac and entry SB-EP ( E5 1620 ) will cap out at 4 cores. On simplistic single threaded, purely loaded in L2 cache benchmarks the Haswell would probably win on incremental micro-architectural improvements. Going to a E5 1620 v2 (Ivy Bridge ) probably wouldn't really change that (probably still capped at 4 cores, but likely closer on max Turbo. )
Similar for any code that heavily leverages AVX. The memory bandwidth handicap (about 1/2 that of E5's ) again offset by micro-architecture improvements.
A $3500 Mac Pro isn't the entry level Mac Pro. Crossing $3K threshold there is a core count advantage. Again, if myopically focused on most affordable, maximized single threaded performance buying the Mac Pro is likely non-optimal.
A 2012 iMac with max CPU , Fusion Drive , max GPU is $2,599
A 2013 Mac Pro with E5 1620 , "roll your own Fusion Drive" , min GPU would probably come in about $2,799
Neither one of those are anywhere near $3500.
The latter would likely be much more well rounded performance. The GPU is likely substantially better and for $200 SSD Fusion component would comprise a higher percentage would be accelerated on SSD (e.g., 140-160 GB versus the 128 on iMac).
That would go against their normal cycle and force new board designs for Ivy E5s. You are most likely thinking of E7s (EX) which lacked a Sandy Bridge variant. They'd be re-pinned going from Westmere to Ivy.
I heard Apple sold around a million Mac Pros in 2010? or 2011? That didn't seem too bad. I thought it was more like 50,000
The way people talk...
I heard Apple sold around a million Mac Pros in 2010? or 2011? That didn't seem too bad. I thought it was more like 50,000
No way they sold a million. Maybe you heard the million figure in relation to Apple being able to build a million Macs at a 200 person plant in the US and the Mac Pro being a potential choice for that? A Fortune analyst said "Only the Mac Pro and mini sell fewer than 1 million per year, and the Mac Pro is considerably heavier and more expensive to ship".
my guess is sandy bridge e xeons and radeon 7000 series, regardless of newer technology being around the corner (it always is)...
if it's anything to go on (and though it probably isn't), i think 10.6.6 beta drivers in dec. '10 had radeon 6000 support and then the new mbps w/ 6750m yadda were released in the end of february '11..
so my hope would be the same, we've seen driver support for 7000 series chips in 10.8.3 in Nov. I believe, maybe a new mac pro will arrive in the first few months of the new year.
if it's anything to go on (and though it probably isn't), i think 10.6.6 beta drivers in dec. '10 had radeon 6000 support and then the new mbps w/ 6750m yadda were released in the end of february '11..
so my hope would be the same, we've seen driver support for 7000 series chips in 10.8.3 in Nov. I believe, maybe a new mac pro will arrive in the first few months of the new year.
I don't think that the fact that an iMac has a discrete GPU soldered to it has anything to do with why it's more Thunderbolt capable... I think it's because the desktop processors have iGPUs... The framebuffer is copied from the discrete gpu to the iGPU and then output through the thunderbolt port on the logic board... This is how it's done on the PC thunderbolt motherboards at any rate with Lucid Virtu (there's some videos of JJ of Asus using a Thunderbolt Display with a GTX 680)...Since Xeons lack an iGPU, they have an engineering dilemma and very well could be the hold up. Will be interesting to see how TB is implemented, assuming it even is.
In so far as iGPUs mean there is display port signals being routed on the motherboard already and it isn't hard to do another set is the only causal correlation there.
No. That isn't how it is used in general on PC TB boards, because Lucid Virtu isn't used in general. The virtual GPU solution is used when need to get a discrete card into the mix but that is not part of a Thunderbolt certification exercise.
Intel has been specific since Thunderbolt has been introduced that it is not oriented to PCI-e card solutions. It launched with no PCI-e card components are essential parts of a solution nor where they promised as forthcoming by Intel. PCI-e and display port signals on the same logic board as the TB controller we always part of the essential core requirements.
There is zero engineering dilemma here. Display Port switches are sold off the shelf. Here is a TI one:
http://www.ti.com/product/hd3ss212
Apple doesn't use a virtual GPU like the Lucid Virtu solution. There is a copy when switching all the GPU work from iGPU to dGPU but they don't run in tandeem, nor is the output solely coupled to the iGPU.
Million? No way, no how. The whole workstation market isn't even 4 million. For one 2011 quarter:
" ... All told, workstation vendors shipped about 932 thousand branded workstations, ... "
http://jonpeddie.com/publications/workstation_report/
Apple share of the workstation market isn't substantively different than there overall PC Market placement. Namely, overall in the 4-6% range with location pockets around 10% where there are much higher than average amonts of money spent per unit. About 100K is far more likely in the general ballpark than 1,000K.
Last edited:
another architecture change seems due.
i mean, why have your current stuff work, right?
Rosetta MKII, on the way!
i mean, why have your current stuff work, right?
Rosetta MKII, on the way!
Hey there. You seem to know a little more about the thunderbolt stuff than an average Joe like myself.. so.. may I ask how you think Apple may implement thunderbolt on a new Mac Pro? Would they have the thunderbolt ports on the back with the rest of the ports? If so, would they support outputting video that is sent via the PCIe GPUs? Or, will the GPU's have the Thunderbolt Ports?
The most straightforward way of doing it is to simply just copy what they did on the iMac. This isn't something Apply hasn't already done. Like the 2012 iMac, solder a decent GPU and VRAM right onto the motherboard. There is no need for a video switch since there is no video coming out of the IOHUB support chip ( which gets it from a mainstream Core i CPU package ).
[ there is some speculation Apple may not have hooked up CPU package on the lower iMacs anyway since they are just Intel HD2500 ( not even HD400) packages. In that case it would be exactly like they did on the iMac. ]
The Mac Pro would be different from the iMac but it would still have 3-4 PCI-e slots. (likely 4). In that sense system is different in that the user can add a 2nd video card and use both in parallel. In fact, the standard config might just come that way with two GPUs installed by default.
What folks seem to be getting their underwear in a twist over is getting all 2 or more GPUs to pump video data over Thuderbolt. It isn't necessary now when have 2-4 video cards. Should really be necessary when have just one embedded GPU.
Two ports just like the iMac. [ there may be some other ports on front USB 3.0 and/or Firewire , but not Thunderbolt. Most likely will loose two legacy ports off the back and replaced with TB. Whether those move to front or disappear is a toss-up. ]
No.
No TB ports on PCI-e card. It is theoretically possible but not practical market wise. Lots of design and certification complexity for extremely limited gains for what would likely be a highly proprietary card.
No virtual GPU because I don't think Apple or graphics makers will allocate the resources. Some third party may step in, but not no 3rd party enhanced/solved the embedded+discrete "at the same time" problem on the laptops. I doubt the much smaller Mac Pro market would draw one of those virtual GPUs vendors in.
Nor does Apple put time into supporting SLI/Crossfire. Again complexity that Apple avoids.
A discrete GPU PCI-e card is perfectly capable of hooking to a monitor by itself. There is extremely narrow added benefit of rerouting that output through Thunderbolt into yet another cable. (stuff like locating the Mac Pro 10 yards away from the monitor and a "need" for fiber Thunderbolt is just appealing to tiny corner case contexts. ) For normal cable lengths hooking to a general monitor is cheaper to just use the normal connections already present on both card and monitor.
Last edited: