Retina iMac FLASH storage and firmware update
As reported here (http://www.barefeats.com/imac5k1.html), it seems like the Retina iMac's flash storage is not as fast as that of the nMP or Retina MBP (even though it's still great performance anyway) - my question to the technically-savvy is: would it be possible for Apple to improve it via a firmware update?
Thanks.
Comments
If it's the firmware slowing it down and not the hardware then an update could improve it. I suspect it'll be the hardware. In ifixit's teardown, the 5K iMac had the same Sandisk SSD that was in the 2013 MBP, which is slower:
http://www.barefeats.com/rmbp14.html
The 2014 model uses Samsung and so does the Mac Pro. It may even be the case that some models use different brand SSDs so some 5K iMacs might get Samsung SSDs with higher performance.
You have to play the component lottery again? Ugh, that I hate about Apple.
Could it be though that the higher capacity SSDs are from Samsung and the others the lower ones from Samdisk?
The 5K iMac one tested at the link was 1TB so I don't think the capacities are limited by brand.
The performance difference shouldn't be noticeable in everyday use. If you copied say a 20GB file, the Sandisk would take 27 seconds vs 21 seconds on the Samsung. When it comes to small files, a 20GB folder of images would take 55 seconds on the Sandisk vs 45 seconds on the Samsung.
I just noticed the footnote at barefeats says their iMac has a Samsung drive. They reckon the iMac SSD is running over x2 PCIe vs x4 in the Mac Pro but I don't think so as each PCIe 3 lane has ~1GB/s so both would be x2 plus the MBP runs its SSD at the same speed as the Mac Pro. They might have mistakenly assumed the iMac has the same model SSD as the Pro. No doubt the more people that get them will test them and note the SSD model they have.
You have to play the component lottery again? Ugh, that I hate about Apple.
Could it be though that the higher capacity SSDs are from Samsung and the others the lower ones from Samdisk?
That's why you don't play the component lottery. You determine if the worst possible outcome is acceptable and only buy if that is the case. The mac pro is also a very expensive item, so it would not surprise me to see a higher flash budget there.
I haven't looked at benchmarks between the two. I just stated if there is more than one supplier, you should look at whether the lesser one is acceptable in performance. That way you can be sure of your purchase. Apple plays a lot of games with mini delays. After all this time they finally release a haswell model, and they obviously waited for the retina imac for whatever reason. I find it silly given that the price spread between the two, but anyway I wouldn't expect another update for a year or longer in spite of the asinine delays on this one.
Well the iris monicker is a matter of branding, but I don't see the demand for improved integrated gpus going anywhere. There are various frameworks and things that weren't supported on integrated graphics for years. That is no longer the case. They aren't ideal for gaming, and you aren't going to animate with them, but the general improvements have taken care of quite a few edge cases where you have some openGL reliance without being really computationally heavy.
but as we're seeing with the retina MBP there is not a huge gap between that and the 750M at least to my knowledge.
The gap is still significant. Some of the figures are biased. They pick one test whereas gpus are more difficult to benchmark.
Broadwell's Iris Pro 6200 should top the 770M. The following site says Skylake will have a GT4 version that is 50% faster than Broadwell:
http://www.cpu-world.com/news_2014/2014101501_More_details_on_Skylake_GT1_GT2_and_GT3_graphics.html
So 2015 Iris Pro should fall short of the 850M/950M but 2016 Iris Pro should surpass the 850M. NVidia will only have Pascal in 2016, which is 40% faster than Maxwell so Intel's IGPs and NVidia's GPUs could again have negligible performance difference between them. Once you're at that level, the improvements get less important, most games are playable at very high quality and Iris Pro OpenCL computation will come out on top:
http://www.notebookcheck.net/NVIDIA-GeForce-GTX-850M.107795.0.html
That's why you don't play the component lottery. You determine if the worst possible outcome is acceptable and only buy if that is the case. The mac pro is also a very expensive item, so it would not surprise me to see a higher flash budget there.
But the point here is that even the 2014 MBP (similar price point/target market as the 5K iMac) got speedier SSDs. And if Marvin is correct PCIe-wise, it would really come down to how fast the SSD ITSELF is - this would not be the first time Apple plays such a lottery approach, considering that HDDs have also varied in performance/speed over the years (i.e. Western Digital vs. Seagate and so on).
We'll see as more and more 5K iMacs are REALLY reviewed (by the way, why are detailed reviews SO SLOW to be published this time, apart from BareFeats?).
But the point here is that even the 2014 MBP (similar price point/target market as the 5K iMac) got speedier SSDs. And if Marvin is correct PCIe-wise, it would really come down to how fast the SSD ITSELF is - this would not be the first time Apple plays such a lottery approach, considering that HDDs have also varied in performance/speed over the years (i.e. Western Digital vs. Seagate and so on).
We'll see as more and more 5K iMacs are REALLY reviewed (by the way, why are detailed reviews SO SLOW to be published this time, apart from BareFeats?).
Yeah I'm not that happy with Apple at the moment for another reason, but I was pointing out that validation of components is typically a matter of whether they meet some minimum criteria. It's not always possible to have identical performance between components, and if they decided to advertise performance, it would be prefixed by "up to" (which I really hate).
Yeah I'm not that happy with Apple at the moment for another reason, but I was pointing out that validation of components is typically a matter of whether they meet some minimum criteria. It's not always possible to have identical performance between components, and if they decided to advertise performance, it would be prefixed by "up to" (which I really hate).
Alas, it seems like BareFeats has confirmed that the reason for slower performance is NOT in the SSD, but in the fact that the iMac 5K uses only a 2x PCIe lane for its flash drive - can anyone confirm that the Retina MBP uses four instead (just like the nMP)?
Alas, it seems like BareFeats has confirmed that the reason for slower performance is NOT in the SSD, but in the fact that the iMac 5K uses only a 2x PCIe lane for its flash drive - can anyone confirm that the Retina MBP uses four instead (just like the nMP)?
The chip has 16 lanes total. Bandwidth for thunderbolt, usb3, graphics, drives, etc. all come out of those 16 lanes. I suspect that it's actually over-subscribed somewhat, but they may have allocated whatever they could allocate.
It's important to distinguish between PCIe versions as PCIe3 lanes are 1GB/s each. If it's PCIe2, they are 500MB/s each. There's also overhead so real-world will be lower. The Mac Pro breakdown from Anandtech said it connected the SSD to the PCIe2 lanes that were available:
http://www.anandtech.com/show/7603/mac-pro-review-late-2013/8
The total was 40 PCIe3 lanes from CPU plus 8 PCIe2 lanes from PCH. One interesting note there was that Apple put all 4 USB 3 ports on a single PCIe2 lane so they can't all operate at full speed. All of the PCIe3 lanes are divided between dual GPUs, RAM and Thunderbolt.
Desktop chips only get 16 lanes of PCIe3 from the CPU so they'd probably divide that up by having the GPU in x8 (=x16 PCIe2) and split the rest between RAM and dual Thunderbolt. If they then have 8 lanes of PCIe2 on top, they could split it the same way as the Mac Pro but there is a SATA drive to support and maybe they give full bandwidth to each USB 3 port. The laptops only have 2 USB ports vs 4 on the iMac and no HDD.
Roll on 2016 where they double bandwidth again with PCIe4:
http://www.theregister.co.uk/2014/06/10/cheap_backwardscompatible_pcie_40_on_track_for_del2015del_2016/
SSDs can already get above 3GB/s in enterprise setups so this is a possibility in machines with PCIe4.
It's important to distinguish between PCIe versions as PCIe3 lanes are 1GB/s each. If it's PCIe2, they are 500MB/s each. There's also overhead so real-world will be lower. The Mac Pro breakdown from Anandtech said it connected the SSD to the PCIe2 lanes that were available:
http://www.anandtech.com/show/7603/mac-pro-review-late-2013/8
The total was 40 PCIe3 lanes from CPU plus 8 PCIe2 lanes from PCH. One interesting note there was that Apple put all 4 USB 3 ports on a single PCIe2 lane so they can't all operate at full speed. All of the PCIe3 lanes are divided between dual GPUs, RAM and Thunderbolt.
Desktop chips only get 16 lanes of PCIe3 from the CPU so they'd probably divide that up by having the GPU in x8 (=x16 PCIe2) and split the rest between RAM and dual Thunderbolt. If they then have 8 lanes of PCIe2 on top, they could split it the same way as the Mac Pro but there is a SATA drive to support and maybe they give full bandwidth to each USB 3 port. The laptops only have 2 USB ports vs 4 on the iMac and no HDD.
Roll on 2016 where they double bandwidth again with PCIe4:
http://www.theregister.co.uk/2014/06/10/cheap_backwardscompatible_pcie_40_on_track_for_del2015del_2016/
SSDs can already get above 3GB/s in enterprise setups so this is a possibility in machines with PCIe4.
Which SATA drive are you talking about in a SSD-only configuration for the 5K iMac?
It still has a SATA connector so you can add a drive in. If say the USB ports aren't in use, other parts don't get more lanes assigned to them.