1
00:00:00,480 --> 00:00:08,880
$7,000. And that's just for the base model of

2
00:00:05,279 --> 00:00:11,599
the new Mac Pro. I mean, in fairness,

3
00:00:08,880 --> 00:00:15,759
Apple has been true to their word here. They have both completed the transition

4
00:00:13,679 --> 00:00:21,680
to Apple Silicon, which they said they would, and they've offered professionals

5
00:00:18,320 --> 00:00:24,400
who are invested in the Mac Pro form

6
00:00:21,680 --> 00:00:29,840
factor some continuity in a product line that has for 20 years been in constant

7
00:00:28,320 --> 00:00:34,960
flux. But I can't help but feel like this is a

8
00:00:32,399 --> 00:00:39,760
classic case of malicious compliance where they're bringing this thing back

9
00:00:36,559 --> 00:00:42,800
to life once again only to prove their

10
00:00:39,760 --> 00:00:45,840
point that it should have stayed dead.

11
00:00:42,800 --> 00:00:48,960
And I'm not the only one. With the same

12
00:00:45,840 --> 00:00:52,000
class of hardware as the Mac Studio that

13
00:00:48,960 --> 00:00:54,879
costs a little over half as much, the

14
00:00:52,000 --> 00:01:01,520
Mac Pro has had a hard time convincing folks that it has any reason to exist.

15
00:00:59,039 --> 00:01:07,280
which won't prevent us from running some interesting experiments on its animated

16
00:01:03,920 --> 00:01:10,080
corpse. For example, Apple said you

17
00:01:07,280 --> 00:01:14,799
can't put a GPU in it, but we tried it anyway just to see what would happen.

18
00:01:12,320 --> 00:01:20,560
And we put it through a battery of tests to search for some

19
00:01:17,680 --> 00:01:24,400
any reason for this computer to have been built. Like the new M2 Ultra

20
00:01:22,560 --> 00:01:29,600
version of the Mac Studio that launched alongside it, the new Mac Pro is

21
00:01:27,040 --> 00:01:33,759
externally essentially unchanged from its predecessor, there's only a couple

22
00:01:31,520 --> 00:01:37,680
of subtle differences around back. Where the previous gen Intel Mac Pro had its

23
00:01:36,000 --> 00:01:42,640
graphics cards mounted at the bottom of the case, nearest the power supply and

24
00:01:39,840 --> 00:01:48,079
the dual 10 gig LAN ports, the HDMI ports for the M2 Ultra Mac Pro are up at

25
00:01:45,520 --> 00:01:53,360
the top, just under the healthy bank of six Thunderbolt 4 ports to go along with

26
00:01:50,479 --> 00:01:57,920
the two on the top. That total of eight is less than the maximum on the old

27
00:01:55,200 --> 00:02:01,600
Intel Mac Pro, but hey, at least you get all eight regardless of the hardware

28
00:01:59,439 --> 00:02:07,360
configuration. Once again, then it's all about what's on the inside. Powering the

29
00:02:04,240 --> 00:02:11,360
new Mac Pro is Apple's top tier M2 Ultra

30
00:02:07,360 --> 00:02:13,360
SOC, which is basically two M2 Max SOC's

31
00:02:11,360 --> 00:02:18,239
glued together with Apple's Ultra Fusion interconnect, which allows them to

32
00:02:14,800 --> 00:02:21,280
operate as one, which is cool, but it

33
00:02:18,239 --> 00:02:25,120
happens to be the exact same chip down

34
00:02:21,280 --> 00:02:27,520
to the maxed out 76 core GPU that's in

35
00:02:25,120 --> 00:02:33,280
our new Mac Studio. The only difference is that unlike our studio, which we set

36
00:02:30,080 --> 00:02:36,879
up with 128 gigs of RAM, our Mac Pro is

37
00:02:33,280 --> 00:02:40,720
equipped with the maximum 192 GB of

38
00:02:36,879 --> 00:02:43,599
unified memory, which sounds like a lot

39
00:02:40,720 --> 00:02:49,040
because it is, but you've got to consider the customer here. One of the

40
00:02:45,920 --> 00:02:51,519
biggest drawbacks to the Intel Mac Pros

41
00:02:49,040 --> 00:02:59,360
was that some CPU configurations were limited to only 768

42
00:02:55,280 --> 00:03:03,519
GB of system memory, which again sounds

43
00:02:59,360 --> 00:03:05,840
like a lot because it is, but was also a

44
00:03:03,519 --> 00:03:11,200
legitimate reason for some professional users to spring for the higherend chips

45
00:03:08,400 --> 00:03:15,360
that could address twice as much. So what I'm trying to say is that when it

46
00:03:13,519 --> 00:03:22,800
comes to folks who are dealing with complex simulations, machine learning or

47
00:03:18,720 --> 00:03:27,280
any large data set, Apple no longer has

48
00:03:22,800 --> 00:03:30,159
any option for them at any price. Now,

49
00:03:27,280 --> 00:03:35,120
this smaller memory pool does have the benefit of being unified, meaning that

50
00:03:32,159 --> 00:03:41,280
Apple's GPU can address more memory than any PCGU, even ones that cost as much as

51
00:03:38,480 --> 00:03:46,720
this entire machine. But we are talking about some pretty specialized workloads

52
00:03:43,760 --> 00:03:51,760
here that what would benefit from a ton of GPU memory that comes at the direct

53
00:03:49,599 --> 00:03:57,920
cost of having basically no system memory. So, we're finding ourselves

54
00:03:54,640 --> 00:04:00,640
asking, if the system configuration is

55
00:03:57,920 --> 00:04:06,400
identical to the Studio and in some ways worse than the old Intel model, what is

56
00:04:03,200 --> 00:04:09,920
the point of the new Mac Pro? What is

57
00:04:06,400 --> 00:04:12,480
all this cooling for? We can answer that

58
00:04:09,920 --> 00:04:17,680
with benchmarks. A quick note before we begin. Apple's product page pits the new

59
00:04:14,959 --> 00:04:23,840
Mac Pro against the previous gen, both baseline and fully kitted out, which is

60
00:04:20,880 --> 00:04:29,600
a 4-year-old machine that started for $1,000 less than this one. But credit to

61
00:04:27,280 --> 00:04:33,840
Apple, it's still a far better frame of reference than their studio comparisons

62
00:04:31,360 --> 00:04:38,400
were. Now, we don't have either a base model or a fully kitted out Intel Mac

63
00:04:36,080 --> 00:04:44,000
Pro, so our testing is going to focus on current gen hardware, including a PC

64
00:04:40,720 --> 00:04:46,320
bench with a Core i9 13900 K and an RTX

65
00:04:44,000 --> 00:04:53,199
4090. Starting with Octane, the Mac Studio is exactly as fast as the Mac Pro

66
00:04:49,520 --> 00:04:55,440
and costs $3,000 less. Not a great

67
00:04:53,199 --> 00:05:00,400
start. And Red Shift doesn't give us much to celebrate unless you consider 1

68
00:04:57,759 --> 00:05:04,560
second a massive win over a machine that is already substantially more expensive

69
00:05:02,320 --> 00:05:10,160
than the PC bench that is obliterating it. Blender 2 shows the Mac Pro's GPU

70
00:05:07,919 --> 00:05:15,680
pushing mostly the same performance on lighter renders, but Gooseberry sees an

71
00:05:13,280 --> 00:05:21,360
8-second gap between the two machines, suggesting that extended GPU loads could

72
00:05:19,199 --> 00:05:24,720
benefit from the extra cooling at its disposal. I mean, [music] they don't

73
00:05:23,039 --> 00:05:28,800
benefit enough to justify Apple's refusal to allow their users to install

74
00:05:26,720 --> 00:05:34,960
more powerful accelerators like the RTX 4090, cuz our PC bench at a lower price

75
00:05:32,639 --> 00:05:40,400
manages to beat both of them. In our mixed 4K 8K HDR timeline in Da Vinci

76
00:05:38,160 --> 00:05:44,320
Resolve, we ran into that same bug that cropped up in our Mac Studio review that

77
00:05:42,720 --> 00:05:48,960
prevents us from comparing these two units. But what we can at least do here

78
00:05:46,400 --> 00:05:53,440
is see the performance that the M2 Ultra is supposed to have, regardless of which

79
00:05:51,280 --> 00:05:57,440
machine it's installed in. Internally, we've investigated this bug a fair bit,

80
00:05:55,280 --> 00:06:01,680
and our best guess right now is that whatever heristics Apple is using to

81
00:05:59,440 --> 00:06:06,000
determine power states and fan speeds is preventing our studio from achieving max

82
00:06:03,840 --> 00:06:09,840
performance. The SSDs are nearly identical between the two machines, with

83
00:06:08,000 --> 00:06:13,199
both of them dramatically outperforming the encoding speed, meaning that they

84
00:06:11,440 --> 00:06:17,680
will never be a bottleneck. [music] We did discover though the Mac Pro's SSD

85
00:06:15,360 --> 00:06:23,680
seems to have a minor advantage in 4K reads. So, point Mac Pro. And another

86
00:06:21,680 --> 00:06:28,160
point for the Mac Pro is in our Chromium compile where it shaved just over two

87
00:06:26,080 --> 00:06:32,639
minutes off of the Mac Studios time, which sounds really impressive until you

88
00:06:30,639 --> 00:06:38,560
realize that that works out to about a 4% performance improvement. But hey, it

89
00:06:36,160 --> 00:06:42,639
does at least show again that the additional cooling may be making a

90
00:06:40,720 --> 00:06:46,720
difference. When we move to synthetics, however, everything kind of falls apart

91
00:06:44,560 --> 00:06:51,039
again. Performance test has the Pro and the Studio at about the same performance

92
00:06:48,720 --> 00:06:56,800
level, and Geekbench gives the Pro only a 1 to 2% lead in the GPU and CPU tests,

93
00:06:54,800 --> 00:07:01,280
respectively. Cinebench surprisingly shows the Pro gain about a 7% lead over

94
00:06:59,360 --> 00:07:05,840
the Studio, which is especially odd considering the renderer that R23 uses

95
00:07:03,599 --> 00:07:10,800
is still Intel optimized. But hey, it's something. And something else is

96
00:07:08,319 --> 00:07:15,680
Blender's CPU renderer, which might not be super relevant to most people, but

97
00:07:13,199 --> 00:07:20,560
here is one more synthetic test that sees real gains over the Mac Studio with

98
00:07:18,160 --> 00:07:25,759
performance more closely matching the PC bench in some scenes. Perhaps the pros

99
00:07:23,360 --> 00:07:30,000
CPU cores are able to more aggressively turbo. Or maybe the massive cooler is

100
00:07:28,479 --> 00:07:35,440
preventing the system from throttling when the studio otherwise would. To test

101
00:07:32,720 --> 00:07:42,000
that theory, we set up our combined CPU and GPU stress test and ran it at room

102
00:07:38,080 --> 00:07:44,400
temperature where oh, lookie here. While

103
00:07:42,000 --> 00:07:49,360
the GPU clocks are roughly the same throughout the run between our two Macs,

104
00:07:46,880 --> 00:07:56,240
it seems that the Mac Studio allows its CPU clocks to drop down before the fans

105
00:07:53,759 --> 00:08:00,479
even get a chance to fully ramp up. That throttle point is visible right here

106
00:07:58,319 --> 00:08:05,199
when we look at the temperatures. For comparison, here's the more gradual ramp

107
00:08:02,720 --> 00:08:10,080
on the Mac Pro without the throttling hump. Curiously though, the fans didn't

108
00:08:07,759 --> 00:08:15,360
kick in for either system until about the 4minute mark of this load. So, the

109
00:08:13,120 --> 00:08:20,960
Mac Pro's ample cooling system seems to be almost purely for show. And given

110
00:08:19,120 --> 00:08:25,759
that at room temperature, the Mac Studios fans never approach their

111
00:08:22,879 --> 00:08:30,400
maximum 3500 RPM, it's pretty clear that with an aggressive fan curve, a Mac

112
00:08:28,080 --> 00:08:35,279
Studio owner should be able to reach the same level of performance as a Mac Pro

113
00:08:32,880 --> 00:08:41,120
owner. [music] There's no additional special sauce in the silicon. And all

114
00:08:37,680 --> 00:08:42,800
while saving enough to buy 150 of the

115
00:08:41,120 --> 00:08:47,040
new mystery water bottles at ltstore.com.

116
00:08:44,959 --> 00:08:51,839
As expected, then lowering our ambient temperature to 10° yielded no change in

117
00:08:49,839 --> 00:08:59,120
performance, and the Mac Pro's fans barely spun. And at 35°, same story,

118
00:08:55,680 --> 00:09:01,440
except the fans finally broke 1,000 RPM,

119
00:08:59,120 --> 00:09:08,720
making the Mac Pro the perfect machine for running heavy loads super quietly in

120
00:09:05,760 --> 00:09:13,519
a really hot environment. Sort of. Actually, I'm not quite sure

121
00:09:11,200 --> 00:09:18,480
how this works given their nearly identical performance, but the smaller

122
00:09:15,920 --> 00:09:22,240
studio draws just a third as much power at idle, meaning that the Pro will

123
00:09:20,480 --> 00:09:29,120
actually kick more heat out into your room. And the Pro also reaches a higher

124
00:09:25,440 --> 00:09:31,440
peak power consumption at 356 W. How

125
00:09:29,120 --> 00:09:36,080
much of this extra draw is from the SOC versus the, you know, extra boards and

126
00:09:33,760 --> 00:09:40,560
fans? We can't say that for sure. But what we can say for sure is that both

127
00:09:38,480 --> 00:09:46,399
machines are substantially less power hungry than our PC. Both at idle and at

128
00:09:43,600 --> 00:09:52,800
full bore. The Core i9 13900K and the RTX490 alone can draw upwards of 600 to

129
00:09:50,480 --> 00:09:57,839
700 watts when both of them are running full tilt. So the Studio offers the same

130
00:09:55,600 --> 00:10:02,560
performance with a smaller footprint and even less power draw. Surely there must

131
00:10:00,880 --> 00:10:11,680
be something the Mac Pro does that the Mac Studio can't, right? Ah, cracking

132
00:10:07,200 --> 00:10:14,800
open the chassis reveals a lot. A lot of

133
00:10:11,680 --> 00:10:17,760
empty space. That is the Intel Mac Pro

134
00:10:14,800 --> 00:10:22,000
had eight PCI Express Gen 3 slots, usually with some expansion cards in

135
00:10:19,680 --> 00:10:27,680
them, while the M2 model features a total of seven PCI Express Gen 4 slots,

136
00:10:25,600 --> 00:10:33,600
which sounds like not much of a downgrade, but it's more than you would

137
00:10:30,720 --> 00:10:39,279
think. While six of the slots are full length, only the bottom two are wired up

138
00:10:36,480 --> 00:10:43,839
for the full 16 lanes with the next four limited to half the total bandwidth at

139
00:10:41,519 --> 00:10:48,800
eight lanes and the slot that's normally occupied by the Apple IO card haved

140
00:10:46,160 --> 00:10:51,920
again to a mere four lanes. Meaning that should you choose to use one of your

141
00:10:50,320 --> 00:10:56,079
Thunderbolt ports as a DisplayPort output and rip that card out and replace

142
00:10:54,320 --> 00:11:01,040
it with something else, your options would be a little limited. Well, more

143
00:10:59,360 --> 00:11:05,200
limited. Um, we're going to get to that later. First, on the subject of

144
00:11:03,120 --> 00:11:08,880
Thunderbolt ports, you might think that you could remove that card, too, and

145
00:11:07,040 --> 00:11:13,680
maybe get another internal expansion slot, but you can't. The design is

146
00:11:11,040 --> 00:11:17,760
proprietary. There's no key like a PCI Express slot would have, and the power

147
00:11:15,839 --> 00:11:22,160
is supplied on the Edge card connector in a shrunk down interpretation of

148
00:11:19,839 --> 00:11:25,680
Apple's one-off proprietary MPX connector for graphics cards. This

149
00:11:24,079 --> 00:11:29,360
enables sufficient power delivery without the use of internal cables. It

150
00:11:27,519 --> 00:11:33,839
really does look pretty on the inside. Now, because MPX is officially gone now,

151
00:11:32,000 --> 00:11:38,800
these three miniature cable connections then are the only way to get a total of

152
00:11:36,399 --> 00:11:42,640
300 W of additional power if you need them for your expansion cards. Not all

153
00:11:40,640 --> 00:11:47,200
expansion cards need that, of course, especially if you're not installing any

154
00:11:44,240 --> 00:11:52,399
GPUs, but this arrangement means that expansion cards like the 4 bay MPX hard

155
00:11:50,160 --> 00:11:56,640
drive module that Promise used to sell are no longer compatible with the new

156
00:11:54,720 --> 00:12:00,480
version of the Mac Pro. That leaves hard drive expansion to just the upper

157
00:11:58,480 --> 00:12:04,399
chamber where a sold separately two drive bay can be fitted that makes use

158
00:12:02,480 --> 00:12:08,959
of the connectors just above the IO cards. Good gravy is there ever a lot of

159
00:12:07,200 --> 00:12:14,000
empty space in here. Turning the computer around reveals that the

160
00:12:10,880 --> 00:12:16,079
shrouded RAM slots are actually still

161
00:12:14,000 --> 00:12:22,880
here from the previous model, complete with easy open locks, except

162
00:12:19,519 --> 00:12:26,639
instead of RAM we get some heat sinks.

163
00:12:22,880 --> 00:12:31,920
Bummer. But wait. Ooh, that's where the

164
00:12:26,639 --> 00:12:34,399
SSD slots are now. All two of them. But

165
00:12:31,920 --> 00:12:39,760
unlike the SSDs used for the Mac Studio, which are bare flash modules, Apple does

166
00:12:37,440 --> 00:12:45,519
in fact still allow users to upgrade these. So that's one major advantage

167
00:12:43,279 --> 00:12:48,480
over the Studio. That is assuming that you don't just kit out your studio from

168
00:12:47,120 --> 00:12:52,639
the factory with all the money you saved by not being a chump and buying a Mac

169
00:12:50,560 --> 00:12:57,839
Pro. The other major advantage is back with those PCI Express slots. So, let's

170
00:12:55,600 --> 00:13:02,399
talk a little bit about what we got working in here. To get the obvious out

171
00:12:59,680 --> 00:13:07,839
of the way first, the Radeon 5700 XT that did work fine in an Intel Mac Pro

172
00:13:05,279 --> 00:13:12,240
was a no-go here. It wouldn't output anything. And while it is detected in

173
00:13:10,399 --> 00:13:17,040
macOS, actually, which is farther than I expected to get, it shows that it

174
00:13:14,720 --> 00:13:21,680
doesn't have a driver. Of course, that is what Apple said would happen. They no

175
00:13:19,200 --> 00:13:26,959
longer offer support for GPUs, even for AMD ones. So, here's all the other cards

176
00:13:24,560 --> 00:13:31,760
we tested, and we did find some surprises. What wasn't surprising is the

177
00:13:29,680 --> 00:13:37,839
biggest issue by far is driver [music] support. Companies like Blackmagic have

178
00:13:34,800 --> 00:13:40,240
large professional Mac user bases, and

179
00:13:37,839 --> 00:13:45,360
their products worked without much ado, but others like Elgato and Aver Media

180
00:13:43,200 --> 00:13:49,894
perhaps rightly feel that there's a larger Mac market for external capture

181
00:13:47,920 --> 00:13:55,200
dongles than there is for internal [music] cards. It was surprising to see

182
00:13:52,399 --> 00:13:59,600
the Melanox 100 gig Ethernet card that we grabbed just pick up and run just

183
00:13:57,760 --> 00:14:03,279
like that. I mean, I would have half expected Apple to remove the driver

184
00:14:01,279 --> 00:14:08,320
despite NVIDIA who acquired Melanox in 2019. [music] Unsurprisingly then, our

185
00:14:05,440 --> 00:14:13,199
Apex Storage X21 SSD carrier card didn't work in fairly spectacular fashion. But

186
00:14:11,279 --> 00:14:16,800
then along came the Honeybadger. Honeybadger don't give a crap.

187
00:14:14,880 --> 00:14:22,639
Honeybadger does what it wants and it wants to work. Not only did Liquids M.2

188
00:14:20,399 --> 00:14:26,800
2 card managed to make the SSDs show up in system information. But we were even

189
00:14:24,959 --> 00:14:31,600
able to use disk utility to create a RAID array on it. Though for some reason

190
00:14:29,680 --> 00:14:38,365
known only to Apple, they are registered as external drives. I mean, hey, at

191
00:14:34,800 --> 00:14:39,040
least we can install SSDs in the slots.

192
00:14:38,365 --> 00:14:45,120
[laughter] Yay. And so between this and the 100 gig

193
00:14:42,320 --> 00:14:48,800
nick, you know, if the dual onboard 10 gig nicks are not enough for some

194
00:14:46,800 --> 00:14:54,399
reason, uh we might have found some of the only expansion cards that really

195
00:14:51,600 --> 00:14:57,839
make more sense inside the Mac Pro than in one of the six Thunderbolt enclosures

196
00:14:56,480 --> 00:15:02,320
that you could buy for the difference in price and install your expansion cards

197
00:14:59,760 --> 00:15:07,519
in with the Mac Studio. Oh, there is another card that pros might want to

198
00:15:03,839 --> 00:15:09,519
use. Avid's HDX. It's $500 and we have

199
00:15:07,519 --> 00:15:14,000
no use for it, so we didn't buy one. But apparently a driver does exist for these

200
00:15:11,519 --> 00:15:17,680
cards and it does work on Apple Silicon. It just requires you to relax the

201
00:15:15,839 --> 00:15:23,680
startup security settings on your Mac, which should be fine, but isn't ideal.

202
00:15:21,920 --> 00:15:28,480
The more I look at this thing, then the more I'm coming to appreciate the Mac

203
00:15:25,920 --> 00:15:32,240
Studio, I guess. As much as we ragged on that machine for not living up to

204
00:15:30,079 --> 00:15:37,279
Apple's hype and for not liking the way that they marketed it, it's safe to say

205
00:15:34,320 --> 00:15:42,399
that it's more than enough computer for 90% of the people who were looking for a

206
00:15:39,279 --> 00:15:46,240
Mac Pro. And hey, at least it doesn't

207
00:15:42,399 --> 00:15:48,880
cost another $3,000 only to do

208
00:15:46,240 --> 00:15:52,320
almost nothing to address the other 10% of users.

209
00:15:50,639 --> 00:15:57,199
I don't know what to say other than that this thing is an abject failure and an

210
00:15:55,040 --> 00:16:03,839
absolute slap in the face to Apple's professional users yet again. Um, does

211
00:16:02,079 --> 00:16:07,920
anyone want to buy a lightly used M2 Ultra Mac Pro that uh got scratched

212
00:16:06,639 --> 00:16:12,639
while we were sliding it out of the thermal chamber? No. If you guys enjoyed

213
00:16:10,720 --> 00:16:19,920
this video, go check out our review of the Mac Studio for our critique of the

214
00:16:15,440 --> 00:16:19,920
somehow better of the two machines.