1 00:00:00,000 --> 00:00:03,520 AMD's Ryzen 9000 is a true generational step forward. 2 00:00:03,520 --> 00:00:08,440 I'm talking bigger level one cache, TSMC's new and improved 4nm FinFET process, 3 00:00:08,440 --> 00:00:13,120 and a double digit IPC uplift thanks to its Zen 5 architecture. 4 00:00:13,120 --> 00:00:17,480 And the best part? They're claiming they finally got a non-X3D chip 5 00:00:17,480 --> 00:00:22,480 that can, on average, beat their giga-chad 5800X3D 6 00:00:22,640 --> 00:00:28,920 all while using less power. As for Intel, well, they're in the middle 7 00:00:28,960 --> 00:00:32,600 of an absolute dumpster fire to the point where I actually think 8 00:00:32,600 --> 00:00:38,080 the more interesting question is whether Team Red can slay themselves. 9 00:00:38,080 --> 00:00:43,360 Or whether you should just skip this launch until they inevitably come out with an X3D variant 10 00:00:43,360 --> 00:00:49,920 of these chips that is so strong it makes me forget to segue to our sponsor. 11 00:00:49,920 --> 00:00:54,360 Because of AMD's staggered launch, we're only looking at the Ryzen 5 9600X 12 00:00:54,360 --> 00:00:59,880 and Ryzen 7 9700X today. Compared to last-gen, they boost slightly higher 13 00:00:59,880 --> 00:01:04,680 and level one cache has increased, but the other caches are otherwise untouched 14 00:01:04,680 --> 00:01:10,640 as our core counts compared to last-generation. However, thanks to TSMC's manufacturing process 15 00:01:10,640 --> 00:01:17,280 improvements, AMD has managed to pack an extra nearly two billion transistors into these little guys, 16 00:01:17,280 --> 00:01:22,580 about 26.5%. Too bad that wasn't enough transistors. 17 00:01:22,580 --> 00:01:26,400 Starting with F123, we see our previous-generation 18 00:01:26,400 --> 00:01:31,880 X3D chips leading the pack. Now, AMD never claimed that the 9700X 19 00:01:31,880 --> 00:01:36,920 would be the new gaming champ, but what they did say is that it's supposed to beat 20 00:01:36,920 --> 00:01:41,460 5,000 series 3DV cache chips in gaming on average. 21 00:01:41,460 --> 00:01:45,720 We found that the story was a bit more complicated than that, but we'll get to that later. 22 00:01:45,720 --> 00:01:50,360 For now, I'm happy to tell you that our new 9,000 series chips beat everything else 23 00:01:50,360 --> 00:01:55,360 on the field except for Intel's 14900KS, and let's be real. 24 00:01:55,400 --> 00:02:01,600 I can't personally tell the difference between 381 frames per second and 378, 25 00:02:01,600 --> 00:02:06,460 and I certainly wouldn't wanna deal with the extra heat, power consumption, or cost. 26 00:02:06,460 --> 00:02:10,240 Moving to a very CPU-bound eSports title in Rocket League, 27 00:02:10,240 --> 00:02:14,880 wow, did these things ever rip. Do I need over 800 frames per second? 28 00:02:14,880 --> 00:02:17,880 No, but is it cool that our 9700X 29 00:02:17,880 --> 00:02:24,520 is heads and tails above everything else? Absolutely, and the 9600X isn't very far behind. 30 00:02:24,520 --> 00:02:28,680 Now, it's clear that any of the CPUs we tested will do this job just fine, 31 00:02:28,680 --> 00:02:34,200 even our inexpensive 14100F, but if you were looking for a generational gain, 32 00:02:34,200 --> 00:02:39,840 it's clear that they can be found. Like in Returnal, where our 9700X wins again. 33 00:02:39,840 --> 00:02:46,040 Not by much over AMD's go-to-X3D chips, but if we narrow in on the apples-to-apples comparisons 34 00:02:46,040 --> 00:02:52,340 against the non-X3D predecessors, we're looking at gains of over 20%. 35 00:02:52,340 --> 00:02:56,900 Now, considering these processors are barely different when we compare the spec tables, 36 00:02:56,900 --> 00:03:01,100 and that they're using way less power, that is pretty dang impressive 37 00:03:01,100 --> 00:03:05,100 and makes me very excited for the inevitable X3D refresh. 38 00:03:05,100 --> 00:03:10,300 Of course, those aren't here yet, and we haven't finished painting a full picture of these. 39 00:03:10,300 --> 00:03:15,740 You might have noticed that most of what we've tested so far has been at low settings instead of ultra, 40 00:03:15,740 --> 00:03:19,940 so let's change it up a bit and look at maxed out Total War Warhammer 3. 41 00:03:19,940 --> 00:03:24,180 This is still a fairly CPU-dependent game, which is why we chose it, 42 00:03:24,180 --> 00:03:28,380 because we're testing CPUs, but you can see that once you start to turn the details up, 43 00:03:28,380 --> 00:03:32,580 things get pretty close at the top of the charts. To show you why this matters, 44 00:03:32,580 --> 00:03:38,540 we also ran this title at low, where you can see that the AMD CPUs do pull farther ahead, 45 00:03:38,540 --> 00:03:43,300 but realistically, if you're spending this kind of money on your CPU, you're probably not 46 00:03:43,300 --> 00:03:47,580 gonna be running everything at low. So it's worth taking all of these results as what they are, 47 00:03:47,580 --> 00:03:51,060 an illustration of how much faster these CPUs are 48 00:03:51,060 --> 00:03:54,620 in a completely non-GPU bottleneck scenario. 49 00:03:54,620 --> 00:04:01,700 Probably the most interesting thing here, though, is our little 1400F that could just chugging along down there, 50 00:04:01,860 --> 00:04:06,560 but we'll come back to that later. First, I wanna talk about our results from our X3D chips. 51 00:04:06,560 --> 00:04:11,560 The 7800X3D does typically be the prior generation 5800X3D, 52 00:04:12,580 --> 00:04:18,740 but not by as much as you might think, both in averages and in all important 1% lows. 53 00:04:18,740 --> 00:04:23,940 So if you're a gamer, realistically, our recommendation from two and a half years ago 54 00:04:23,940 --> 00:04:28,380 hasn't really changed other than that, you might be able to save yourself a buck 55 00:04:28,380 --> 00:04:31,820 with the newer, cheaper 5700X3D. 56 00:04:31,820 --> 00:04:36,580 We didn't actually run the numbers on this chip, but it's only marginally slower than its big brother, 57 00:04:36,580 --> 00:04:39,860 and when paired with a budget board and some DDR4 memory, 58 00:04:39,860 --> 00:04:45,020 it is an outstanding value. X3D's weakness, however, is in productivity, 59 00:04:45,020 --> 00:04:50,380 and this is where the fun begins. Intel was already behind in PugetBent Photoshop, 60 00:04:50,380 --> 00:04:55,180 but now AMD has increased their lead with a 10 plus percent gain 61 00:04:55,180 --> 00:05:00,560 over their last-gen 7000 series. In handbrake, our H.264 results were disappointing 62 00:05:00,560 --> 00:05:04,380 with the last-generation Ryzen 7 beating the new hotness, 63 00:05:04,380 --> 00:05:07,500 but AV1 encoding is a whole other story 64 00:05:07,500 --> 00:05:11,460 thanks to AM5's switch to a true 512-bit data path 65 00:05:11,460 --> 00:05:16,900 instead of double-pumping 256. We see a similar improvement on the 9600X in Blender, 66 00:05:16,900 --> 00:05:21,020 but curiously, the 9700X isn't gonna give you a reason 67 00:05:21,020 --> 00:05:27,620 to upgrade over last-gen. Perhaps it's not as powerful as we thought it would be? 68 00:05:27,660 --> 00:05:32,180 More on that later. And unfortunately, that trend of barely changing 69 00:05:32,180 --> 00:05:36,420 continues with Godot Compile, where we have some marginal-generational uplift 70 00:05:36,420 --> 00:05:40,100 on the Ryzen 7 and then slightly better results on the Ryzen 5. 71 00:05:40,140 --> 00:05:44,580 7-Zip is a tad better for both in compression, but then worse in decompression, 72 00:05:44,580 --> 00:05:47,700 and while Cinebench does see an improvement, it's minimal. 73 00:05:47,700 --> 00:05:51,540 So what's going on here? As it turns out, a lot of it has to do 74 00:05:51,540 --> 00:05:54,860 with the lower power consumption that we mentioned earlier. 75 00:05:54,860 --> 00:05:59,860 In Cinebench, our 7700X consumes an average of 144 watts, 76 00:06:00,140 --> 00:06:04,420 while the 9700X is steady at 88 watts. 77 00:06:04,420 --> 00:06:09,100 That is almost a 40% decrease in power consumption 78 00:06:09,100 --> 00:06:14,860 while netting slightly more performance. And here in this part of Canada, where power is cheap, 79 00:06:14,860 --> 00:06:17,940 it's hard to get too excited about this kind of efficiency, 80 00:06:17,940 --> 00:06:21,340 but for other parts of the world, you could save so much on power, 81 00:06:21,340 --> 00:06:24,360 you could treat yourself to a PTM7950 thermal pad 82 00:06:24,360 --> 00:06:32,020 from LTTstore.com. And check this out. Sure, our 14700K is streets ahead in terms of raw numbers, 83 00:06:32,020 --> 00:06:36,700 but it's using almost 250 watts out of the box to get there. 84 00:06:36,700 --> 00:06:40,780 That is almost triple the power for just under twice the score. 85 00:06:40,780 --> 00:06:44,780 And if you toss out those e-cores, we can see that in single core testing 86 00:06:44,780 --> 00:06:49,780 on both the 97 and 9600X, Intel is losing by a wide margin. 87 00:06:50,500 --> 00:06:53,540 Now part of this could be because AMD seems to be leaving 88 00:06:53,540 --> 00:06:59,820 some performance on the table. Check this out. In Cinebench, our Ryzen 7 refuses to reach 89 00:06:59,820 --> 00:07:05,540 the advertised 5.5 gigahertz boost clock, capping out instead at around 5.2. 90 00:07:05,540 --> 00:07:09,780 And I know what you're thinking. Well, that 5.5 is just a single core boost. 91 00:07:09,780 --> 00:07:13,380 It's just for gaming, except that on our Ryzen 5, 92 00:07:13,380 --> 00:07:17,140 it does manage to hit its maximum advertised boost 93 00:07:17,140 --> 00:07:22,980 across all cores in multi-threaded workloads. Not all the time, mind you, but sometimes, 94 00:07:22,980 --> 00:07:29,540 our Ryzen 7 doesn't seem to be defective. It hits 5.5 in games and even in other stress tests, 95 00:07:29,540 --> 00:07:33,020 like OCCT-Linpack and outside of games, 96 00:07:33,020 --> 00:07:36,380 it's using the same or a similar power envelope to do so. 97 00:07:36,380 --> 00:07:41,020 So what gives here? And I know what you're thinking. It's gotta be cooling, right? 98 00:07:41,020 --> 00:07:45,460 Well, here's the thing. This might be clear to some of you already, 99 00:07:45,460 --> 00:07:50,700 but less watts means less heat. So where AMD was already crushing Intel, 100 00:07:50,700 --> 00:07:55,060 now we're seeing even better performance. These chips run so cool. 101 00:07:55,060 --> 00:07:58,620 I'm talking a 22 degree drop on average on the Ryzen 7 102 00:07:58,620 --> 00:08:01,900 and 16 degrees on the Ryzen 5. 103 00:08:01,900 --> 00:08:05,020 So for anyone who hates seeing high CPU temps, 104 00:08:05,020 --> 00:08:09,060 AMD just dummied Intel and did it efficiently. 105 00:08:09,060 --> 00:08:12,500 So it's not thermals. What is it? We'll get to it. 106 00:08:12,500 --> 00:08:15,820 First, I wanna talk about where this efficiency uplift comes from. 107 00:08:15,820 --> 00:08:19,860 It's partially thanks to AMD's new two-ahead branch prediction. 108 00:08:19,860 --> 00:08:23,260 This is gonna be really simplified down, but here we go. 109 00:08:23,260 --> 00:08:27,780 Let's say you're running a program and that program is trying to get from point A to point B. 110 00:08:27,780 --> 00:08:33,060 Without branch prediction, at every fork in the road, it has to stop and figure out which way to go, 111 00:08:33,060 --> 00:08:36,220 then go down the right path. With branch prediction, 112 00:08:36,220 --> 00:08:41,540 your CPU can predict the path ahead of time, removing the need to stop and think at each fork. 113 00:08:41,540 --> 00:08:46,100 With two-ahead prediction, the CPU is guessing the next two steps. 114 00:08:46,100 --> 00:08:50,860 And by dual porting the instruction fetching, AMD is able to basically check the guesswork 115 00:08:50,860 --> 00:08:54,980 about twice as quickly, which speeds up the entire process. 116 00:08:54,980 --> 00:08:58,460 Now there is still a penalty if they get a prediction wrong, 117 00:08:58,460 --> 00:09:03,460 but it's not any worse than how they were doing things before. So it's all gains, baby. 118 00:09:03,460 --> 00:09:07,180 So what's not to love here? Well, there are a few things. 119 00:09:07,180 --> 00:09:10,380 We are not getting the XDNA2 AI component 120 00:09:10,380 --> 00:09:14,140 that is present in the StrixPoint mobile counterpart to these chips. 121 00:09:14,140 --> 00:09:18,180 Instead, AMD is claiming that the CPU can act as an AI accelerator, 122 00:09:18,180 --> 00:09:22,740 which is a pretty honest takedown of the whole AI frenzy 123 00:09:22,740 --> 00:09:27,780 that AMD themselves is participating in, but it could make a slight difference down the line, 124 00:09:27,780 --> 00:09:31,300 especially for power consumption. A bigger issue for me though, 125 00:09:31,300 --> 00:09:34,420 is that more Ryzen 9000 chips are coming, 126 00:09:34,420 --> 00:09:39,980 but everything AMD has announced is going to be more expensive than what we're looking at today, 127 00:09:39,980 --> 00:09:45,340 which continues AMD's troubling trend of ignoring their Ryzen 3 customers. 128 00:09:45,340 --> 00:09:49,060 We also think that while efficiency is exciting to the data center folks 129 00:09:49,060 --> 00:09:54,860 who are eagerly awaiting Epic Turin with its unthinkable 192 core counts, 130 00:09:54,860 --> 00:09:57,860 most desktop users might have preferred to see AMD 131 00:09:57,860 --> 00:10:00,940 give these chips a little more juice by default. 132 00:10:00,940 --> 00:10:04,900 And it's finally time to talk about the weirdness that we observed. 133 00:10:04,900 --> 00:10:08,420 We tested the 9700X with PBO enabled 134 00:10:08,420 --> 00:10:14,820 and with its power set to unlimited. And what we found was that it easily drew over 140 watts, 135 00:10:14,820 --> 00:10:19,060 putting it in the same neighborhood as its predecessor in terms of power consumption 136 00:10:19,060 --> 00:10:22,340 and resulting in some measurable performance gains. 137 00:10:22,340 --> 00:10:27,180 Of course though, this is technically overclocking and with the recent reliability issues 138 00:10:27,180 --> 00:10:32,360 that have plagued Intel users, I can understand why some of you might shy away from this. 139 00:10:32,360 --> 00:10:37,460 So I guess what I'm saying is, I wish AMD had just officially under warranty 140 00:10:37,460 --> 00:10:41,620 given these things more juice so that we could reach those advertised boost clocks 141 00:10:41,620 --> 00:10:45,940 across all cores like we did once we enabled PBO. 142 00:10:45,940 --> 00:10:50,860 Also, last thing, the lowest tier AM5 B840 motherboards 143 00:10:50,860 --> 00:10:54,540 that are supposedly supposed to help make this platform more affordable 144 00:10:54,540 --> 00:10:58,100 don't officially support PCIe Gen 4. 145 00:10:58,100 --> 00:11:03,220 So if you do try to save a buck on your motherboard, you could end up seriously limiting your graphics 146 00:11:03,220 --> 00:11:07,060 and your storage bandwidth, which is just, it's one more knife in the side 147 00:11:07,060 --> 00:11:12,300 of AMD's budget conscious customers. With that in mind, we thought it would be interesting 148 00:11:12,300 --> 00:11:17,460 to see how AMD's new offerings stack up against what you can't buy from AMD right now. 149 00:11:17,460 --> 00:11:23,860 A CPU that costs less than half as much money, fits in a cheaper board that does support PCIe Gen 4 150 00:11:23,860 --> 00:11:27,500 and comes with a stock cooler. The results aren't amazing, 151 00:11:27,500 --> 00:11:31,700 but when you don't even have a competitor in the weight class, you kind of forfeit the fight 152 00:11:31,700 --> 00:11:35,260 and Intel's 14100F puts up pretty respectable results 153 00:11:35,260 --> 00:11:38,700 at less than half the price. Now it struggles in productivity 154 00:11:38,700 --> 00:11:44,580 and struggles even more in CPU-bound games, but if you're on a budget, it's a pretty compelling value. 155 00:11:44,580 --> 00:11:48,940 Of course though, it's based on a platform that has no future upgrade path 156 00:11:48,940 --> 00:11:52,580 and you aren't saving that much compared to AM4, 157 00:11:52,580 --> 00:11:56,700 which by the way, AMD just released another refresh for 158 00:11:56,700 --> 00:12:00,140 with the XT lineup. And they also confirmed that AM5 159 00:12:00,140 --> 00:12:05,260 is gonna be supported until at least 2027. In conclusion then, solid chips, 160 00:12:05,260 --> 00:12:10,700 even better if you're adventurous and turn on PBO, but not really for gamers. 161 00:12:10,700 --> 00:12:16,420 So I'm kind of more excited to see the matchup that's coming then between the 9800X3D 162 00:12:16,420 --> 00:12:21,620 and Intel's Aero Lake, which supposedly is dropping hyperthreading altogether. 163 00:12:21,620 --> 00:12:26,100 Huge move and I don't think anybody knows what that's gonna mean for gaming performance 164 00:12:26,100 --> 00:12:29,140 as they go to a purely big little architecture. 165 00:12:29,140 --> 00:12:33,220 Just like nobody knows what on earth would happen if I didn't segue to today's sponsor. 166 00:12:33,220 --> 00:12:37,540 If you guys enjoyed this video, maybe go check out our review of the 7800X3D. 167 00:12:37,540 --> 00:12:41,860 Somehow that chip and even its predecessor are just killing it. 168 00:12:41,860 --> 00:12:44,780 Years later, outstanding gaming CPUs.