Sidebar Sidebar
Become a MacRumors Supporter for $50/year with no ads, ability to filter front page stories, and private forums.

Is Apple Slowing Down the 16" Intel MBP?

neversoberkid

neversoberkid

macrumors newbie
Original poster
Jul 14, 2020
4
3
So...just got myself a 16" 2 weeks back, and chanced upon this this morning.


TLDW: According to Youtube channel Heads of Technology, Apple has reduced the PCIexpress Lanes allocated to the 5500M from x16 (back in Nov during launch) to x8. You can check this in your System Report-> Graphics->PCIe Lane Width

Signed up an account specifically to raise awareness to this and gather thoughts from fellow 16" owners. I myself have raised a case with Apple as today was my 14-day return due date and I have told them if this is found to be true, I will demand a refund. After 3-4 calls over 2-3 hours today, Apple said they will do some investigation on their end and get back to me tomorrow, so I will be updating this with a reply once I get an answer.

Importantly, have we been gimped once again (iPhone slowdown sheenanigan) to make buying the 5600M more attractive? Who knows.
 
In the beginning of the video it's heavily implied that this change slows things down. Later on he's more nuanced and admits it would require benchmarking x16 vs. x8 to be absolutely sure. I don't think you can draw a meaningful conclusion from just looking at the numbers, except that 8 is less than 16.

Someone in the comment section of the video points to a x16 vs. x8 benchmark using an NVIDIA Titan X. The results of this benchmark are not shocking, although there seems to be a tiny performance hit. This performance hit may translate to the MacBook Pro situation as well, but bear in mind that the NVIDIA Titan X is in an entirely different league than the 5X00m GPU range. To use the highway analogy for PCIe lanes: the Titan X can generate a lot more "traffic" on those PCIe lanes than a 5X00m GPU can. This means it's much more prone to "traffic jams" when the amount of lanes is halved.

In the end a benchmark is needed to conclude whether or not halving the PCIe lanes affects performance. I personally see no reason to return my week old MacBook Pro 16" (5500m 4GB) as I sincerely doubt this is some kind of evil scheme Apple came up with to get us to buy a newer model.
 
Last edited:
  • Like
Reactions: throAU
Hey Mayco,

Where exactly are you seeing a comment about an NVIDIA Titan? I went back to the video after seeing your reply but can’t for the life of me find anyone comparing x16 and x8 benchmarks with NVIDIA cards...
 
I see. Thanks for pointing out. It appears that these benchmarks and the ones I have come across are on the 1080 cards which have a max memory speed of 10Gbps, thereby possibly only a little limited by x8 PCIe limitation experiencing only a 1-4% drop in performance effectively. But as for the 5500M in our MBP’s...the 5500M’s website, memory speed is advertised as 12Gbps. With simple math, this would then effectively translate to a large extent of bottlenecking if PCIe 3.0’s x8 limitations for speeds of around 7.9GB/s is to be respected (In comparison to x16’s 15.74GB/s).

I don’t know at the moment, the whole idea of x16 at launch versus x8 today is still sketchy to say the least. If Apple had determined to limit these MBP’s at x8 from the start it might not be a big deal as part of the engineering decision. To slow it down overtime, even more at 6 months into a product’s lifecycle requires a more upfront explanation if it’s not just a simple bug. It will still suck even if this was taken as a measure to combat all the overheating/noisy fans GPU related complaints that have been all over the Internet as you are then effectively decreasing the performance of a machine to fix certain issues, but still charging the same amount of money from your customers.

Again, I hate to be paranoid, but this coincides very nicely with the recently released 5600M so I hope dearly to be proven wrong with what my guts tell me.
 
  • Like
Reactions: simonmet and Appledoesnotlisten
neversoberkid said:
Again, I hate to be paranoid, but this coincides very nicely with the recently released 5600M so I hope dearly to be proven wrong with what my guts tell me.
Click to expand...
I think that one of the good steps that Apple can take now is to release a bunch of 5600M ones through a refurb store and retailers with a heavy discount. Until they come up with a fix.
 
My 16" MBP was manufactured on 28/10/2019 and according to the System Report the PCIe Lane Width is 8, so if it ever was x 16, then Apple must have changed it soon after releasing it.

PCIe Lane Width: x8
 
neversoberkid said:
I see. Thanks for pointing out. It appears that these benchmarks and the ones I have come across are on the 1080 cards which have a max memory speed of 10Gbps, thereby possibly only a little limited by x8 PCIe limitation experiencing only a 1-4% drop in performance effectively. But as for the 5500M in our MBP’s...the 5500M’s website, memory speed is advertised as 12Gbps. With simple math, this would then effectively translate to a large extent of bottlenecking if PCIe 3.0’s x8 limitations for speeds of around 7.9GB/s is to be respected (In comparison to x16’s 15.74GB/s).
Click to expand...
Please correct me if I'm wrong, but I don't think the GPU's memory speed is related to the PCIe speeds. As far as I know the memory speed (12Gbps) relates to the "internal" connection between the GPU memory and the GPU core, not between the GPU and the CPU (for reference: the 5600m has a memory speed of 1.54Gbps due to its HBM2 memory rather than GDDR6).
 
neversoberkid said:
So...just got myself a 16" 2 weeks back, and chanced upon this this morning.


TLDW: According to Youtube channel Heads of Technology, Apple has reduced the PCIexpress Lanes allocated to the 5500M from x16 (back in Nov during launch) to x8. You can check this in your System Report-> Graphics->PCIe Lane Width

Signed up an account specifically to raise awareness to this and gather thoughts from fellow 16" owners. I myself have raised a case with Apple as today was my 14-day return due date and I have told them if this is found to be true, I will demand a refund. After 3-4 calls over 2-3 hours today, Apple said they will do some investigation on their end and get back to me tomorrow, so I will be updating this with a reply once I get an answer.

Importantly, have we been gimped once again (iPhone slowdown sheenanigan) to make buying the 5600M more attractive? Who knows.
Click to expand...

Howdy neversoberkid,

Intel Mobile CPUs have limited PCIe lanes available, so it is not uncommon for laptops to only configure the GPU to only use x8 vices x16. That leaves more available for things like Thunderbolt ports, NVMe drives, etc.... I am not able to watch the linked video at the moment, but will when I get a chance, and will edit my post.

Here is the report of the CPU in my 16" MacBook Pro: https://ark.intel.com/content/www/u...9880h-processor-16m-cache-up-to-4-80-ghz.html, note that it only has 16 PCIe lanes in total. So if Apple configured the GPU to use all 16, there would be nothing left for the rest of the system. Typically motherboard (or logic board in Apple's case) designers add in extra PCIe lanes via the chipset, but relegate those to less critical devices such as USB ports and extra storage ports. The high speed options are kept on the CPU native ports.

Rich S.
 
  • Like
Reactions: Mayco and Norbert Mikołajczyk
Hey all,

Thanks for the replies and data points.

@Mayco you may be right on that now that I relook at the GPU specs on their sites. Is there anyway to determine then how much bandwidth the 5X00M series in these laptops actually require?

@LinkRS Great explanation and reasoning for why this could be. I’ll accept it if this is the case, just wish someone on Apple’s end was able to explain this (went through like 2 product specialists, 1 Genius and about 3 hours spent total).

Still. Would be interesting to see if Apple responds to this video. But maybe not. After all butterfly keys lasted for a good 3-4 years before a “fix” 😂
 
  • Like
Reactions: Mayco
A radeon 5500 will not push 8x PCIe3.0.

Benchmarks have shown that to get anywhere near saturating 16x, you need a 2080TI, and the 5500 is so far behind that, 8x (or even likely, 4x) will be fine.

The discrete GPU memory speed has zero to do with PCIe bus speed.

Put it this way, 8x PCIe 3.0 is still 2x the bandwidth you'll get to an externally connected GPU via thunderbolt 3.
 
neversoberkid said:
Hey all,

Thanks for the replies and data points.

@Mayco you may be right on that now that I relook at the GPU specs on their sites. Is there anyway to determine then how much bandwidth the 5X00M series in these laptops actually require?
Click to expand...

I expect a theoretical maximum can be calculated, but unfortunately I don't know how. AMD surely knows this value but it's not published anywhere from what I can find.
 
Kung gu said:
AMD Radeon Pro 5300M Specs <-- Mentions only x8 lanes.
Click to expand...
Indeed it does, but it does mention PCIe 4.0 x8, whereas the MacBook Pro uses PCIe 3.0 x8 for the dGPU. According to this source, the bandwidth nearly doubles between PCIe 3.0 and 4.0. Again, I personally don't think going from x16 to x8 affects performance of a 5X00m GPU in a meaningful way, but it would be nice to have the exact numbers.
 
A minor data point. I noticed a slightly lower result in Unigine Valley benchmark than when I ran it earlier this year.

I got just shy of 1900 (ran twice to confirm with identical results) with the 5500M 8 GB now, and when my laptop was new I got just over 2000, so about 5% less. The test I ran now was in cooler months too, whereas I got the laptop in the middle of Summer, adding further weight to a change in policy/allocation and a minor impact on performance.

Unfortunately I don't have evidence of the earlier result. I only recently deleted it during a clean up! I suppose the above finding could be a result of other driver changes or optimisations. But you hope in most cases that changes actually increase performance, not decrease.

I'm curious if anyone else has noticed reduced benchmark results.
 
Last edited:
My 5500M is also connected via 8x PCIe bus. It was the same for the Vega Pro on the 2018 model. I am not sure why this upsets the OP this much? PCIe lanes are a scarce resource on Intel CPUs (it only offer 16 native lanes) and Apple has probably though that it makes more sense to use them where they are actually needed - for the SSD and Thunderbolt.

Anyway, OP is massively overreacting.
[automerge]1594817718[/automerge]
neversoberkid said:
Still. Would be interesting to see if Apple responds to this video. But maybe not. After all butterfly keys lasted for a good 3-4 years before a “fix” 😂
Click to expand...

The “fix” has already been announced - it’s called Apple Silicon ;)
 
Last edited:
You must log in or register to reply here.
Register on MacRumors! This sidebar will go away, and you'll see fewer ads.
Top Bottom
Back