AnandTech benchmarks the two nearly identical Surface Laptop 3s from Microsoft – one with AMD’s latest mobile processor and GPU, and the other one with Intel’s. They conclude:
There aren’t too many ways to sugar coat the results of this showdown though. AMD’s Picasso platform, featuring its Zen+ cores and coupled with a Vega iGPU, has been a tremendous improvement for AMD. But Intel’s Ice Lake platform runs circles around it. Sunny Cove cores coupled with the larger Gen 11 graphics have proven to be too much to handle.
That being said, much like the first desktop Ryzen processors being a huge leap forward for AMD without closing the gap with Intel at the time, the Picasso platform seems to repeat the feat in laptops.
It was fantastic to see AMD get a design win in a premium laptop this year, and the Surface Laptop 3 is going to turn a lot of heads over the next year. AMD has long needed a top-tier partner to really help its mobile efforts shine, and they now have that strong partner in Microsoft, with the two of them in a great place to make things even better for future designs. Overall AMD has made tremendous gains in their laptop chips with the Ryzen launch, but the company has been focusing more on the desktop and server space, especially with the Zen 2 launch earlier this year. For AMD, the move to Zen 2 in the laptop space can’t come soon enough, and will hopefully bring much closer power parity to Intel’s offerings as well.
I can’t wait to see what AMD can offer consumers in the laptop space over the coming years. If it’s going to be a repeat of the desktop space, we’re going to be in for some seriously good times.
Yeah it’s kind of weird that Picasso exists like that… when Renoir is right around the corner (7nm Zen 2 + Vega 7nm)
Thom Holwerda,
You’ve repeated this sentiment a couple of times, but the benchmarks really don’t put AMD ahead for consumer desktop CPUs. Even in the last anandtech.com review you posted praising AMD had year old intel CPUs still besting the new Ryzen chips on singlethreaded benchmarks. AMD does have a clear lead with higher core count processors that are sufficiently loaded. But on singlethreaded applications, which is a bigger bottleneck for most desktops than core count, AMD is still lagging and honestly I was disappointed that the Ryzen CPUs didn’t do better than intel. I’m a bit perplexed at why AMD’s 7nm process isn’t able to easily beat out intel’s 14nm process on sheer speed, but obviously it’s a complex multifaceted topic.
I suspect that both intel and AMD CPUs hit a setback with the mitigations for the spectre and meltdown attacks, although it may have affected them differently.
Single threaded benchmarks are only a very very small part of the story… *overall* AMD has bested Intel solidly, and IPC is higher on AMD than Intel clock for clock… and that frequency lead may evaporate it really is the last shread of lead that Intel has. AMD has higher IPC, more cores, more effective multithreading and SMT, as well as more and faster IO, they can also sustain higher boost clocks across all cores than Intel can because Intel has to resort to exceeding TDP and the chips just can’t go any further and AMD can deliver it at lower prices than Intel….
In short it isn’t a complex topic at all… Intel took shortcuts to make thier CPUs faster and AMD did not. Intel clocks faster but even with the same number of cores cannot match AMD in multithreaded tasks, the only time Intel is faster is when single core boost clocks come into play and frankly that’s pathetic.
AMD was only ever vulnerable to a few spectre attacks and all of those required physical access or were otherwise impossible to take advantage of. AMD has further hardened their CPUs while Intel has barely done anything at all…
cb88,
I agree with you that by conventional wisdom AMD should have the upper hand in theory. But until AMD is able to translate that to better performance in ryzen CPUs, then it would be dishonest to pretend AMD has totally overtaken intel. AMD’s made strong inroads and Intel is being hammered very hard for not making little progress, I agree with this completely. Also I agree that AMD has done a great job at adding lots of cores for highly concurrent workloads. Having 16 or 32 cores at relatively affordable prices is awesome! I couldn’t be happier that AMD is giving intel a run for it’s money.
However I think we need to resist the temptation to let the narrative get ahead of the facts. Single threaded game and application logic do remain significant bottlenecks for most games and desktop applications. This is more the fault of software developers than hardware engineers, but nevertheless if you look at the system load for most computer tasks you’ll see clearly a couple CPUs max out at 100% while the rest float around 0% utilization if you have many cores. In my experience this is a very common bottleneck and intel still scores better than AMD for this bottleneck.
As far as my opinion about AMD’s push for high cores, I welcome that on the one hand, but realistically it’s not enough just to add lots of cores; for high numbers of cores to be relevant to the performance picture, game & software developers will need to do a much better job optimize software for massive core counts. It’s not clear one way or another this will happen though because for their part, CPU intensive software developers have taken a liking to GPGPU architectures that blow CPU performance out of the water. And with GPGPU based software, a faster CPU often outperforms one with more cores. It’s impossible to say what the future will bring, but my prediction, for what it’s worth, is that highly parallel GPUs will continue to become more important while displacing the need for massively parallel CPUs.
You say that you have looked at the benchmarks but you fail to provide them to prove your point.
I’ve looked at the Ryzen 7 3950X review on anandtech and it’s clear that the Ryzen performs on par with Intel (near as makes no difference) while blowing it out of the water in multithreaded loads, i.e. reducing the total waiting time by a significant amount.
Conclusion in very few words: single-threaded performance isn’t a clear win (or not even a clear stalemate) for AMD but that doesn’t really matter because the difference is negligable, while on the other hand the wins in multithreaded performance are so vast that it’s really a silly idea to pay the same amount of money for an Intel CPU that performs not significantly better in single-threaded loads and does far worse in multi-threaded loads, or pay a lot more money for equal multithreaded performance where the singlethreaded performance on those kind of SKUs isn’t necessarily improved over the cheaper parts.
I don’t really know A/ what you are expecting AMD to do; totally annihilate Intel even in singlethreaded performance?; B/ why you are putting so much vague emphasis on singlethreaded performance without providing actual numbers; and C/ why you claim to read into this that AMD has surpassed Intel in every possible metric, of course it isn’t that simple but you seem to blame people being enthousiastic about the current state of AMD of saying exactly that, which isn’t really the case imo.
The first paragraph is quoted from Alfman’s reaction; I really have no idea how to properly format quotes to make them look like quotes. Am I missing something?
It’s not that simple. Let’s split “multi-threaded” into 2 very different cases:
a) embarrassingly parallel. For this case, Intel wins because of AVX512.
b) “not embarrassingly parallel”. If you understand Amdahl’s law you’ll know that as the number of CPUs increases the performance of the “not parallelizable” parts dominates the overall performance. In other words, for this case (with lots of CPUs) good single-threaded performance is necessary to get good multi-threaded performance.
Of course it’s not 3 distinct cases; it’s a continuous line that starts at single-threaded and ends at embarrassingly parallel, with everything falling somewhere in the “not embarrassingly parallel” area between the extremes.
Have a look at most of the games and single threaded benchmarks on the anandtech review that was posted to osnews.
https://www.anandtech.com/show/15044/the-amd-ryzen-threadripper-3960x-and-3970x-review-24-and-32-cores-on-7nm/9
Of course, on multithreaded benchmarks, AMD rocks intel, but when we talk about single threaded, the data clearly shows that AMD is still behind intel. I don’t know why people are in denial about this, but intel comes out ahead just about everywhere there’s a single threaded bottleneck. There’s not much room for interpretation, look at the benchmark scores.
Here are all of the single threaded “system tests”. Note that I’m cherry picking the best AMD scores for each test even though it’s not always the same CPU. This gives AMD a somewhat unfair advantage, but lets go along with it anyways and see how AMD’s best compares with intel’s best for single threaded benchmarks.
The benchmarks reveal a similar picture with low threaded games where intel is the clear winner. Again, AMD does better with more threads, but most games are built to use the GPU for parallelism rather than using high numbers of CPU threads. Just one example:
Saying that ryzen is on par is disingenuous. If the situation were reversed and AMD had an equally large lead, you wouldn’t say that intel was on par with AMD, be honest now
You can make a case that ryzen is good enough for the majority of users despite lower single threaded scores, and you can make a case that ryzen is the better deal price wise, I will not argue against either of those points because I agree with them. However facts are still facts and for now intel is still king of the low thread count performance domain. I don’t know why people are so resistant to the facts.
A) Honestly I did expect more performance, I would have thought that given the theoretical advantages of 7nm process that AMD would have been able to beat intel on raw instructions per second. For whatever reason AMD was not able to make it happen. There could be physical challenges at 7nm and beyond that make it difficult to reliably squeeze more performance out of a CPU with such small transistors.
B) If you read my original post, I was referring to the anandtech benchmarks that Thom posted on osnews, but I included them above anyways.
C) It’s just been hard to get people to acknowledge that the data doesn’t put AMD ahead except for highly multithreaded workloads. It’s absolutely great that AMD does well in high core counts, but realistically most desktop workloads are NOT bottle-necked by core counts. So unless we see a major shift in how software is developed, adding more cores is not going to bring about much, if any, improvement over CPUs with faster single threaded performance. I’m honestly not trying to be anti-AMD, I’m just trying to be realistic.
Alfman,
I was comparing the 3900X/3950X to the i9 9900KS because both are consumer models, the threadrippers have other uses, because seriously who buys a threadripper for singlethreaded performance?
AMD is still behind Intel on a lot of singlethreaded benchmarks, but most of the time that result isn’t really significant or important, meaning the difference it too close to 0% to be perceivable or the difference is too close to 0ms to be perceivable. Tell me, which singlethreaded load is so heavy that it is noticeable by a user?
Speedometer, Octane and Kraken have better scores on the 3950X than on the 3960X but those are still below Intel. Does it matter though? Not really, because the difference is not big enough to notice in real life (or is it? I don’t think 11.22% faster performance in Google Octane 2.0 points to a noticeable difference, OR it must be that webpages load for at least 5 seconds where a 11.22% difference would mean (generalised) 0.56s slower loading on AMD. You see where I’m going, this almost never happens in a browser.
X264 pass 2 is significantly more significant since it is by far taking the most time.
Geekbench is not a good benchmark to interchangebly compare results for different architectures, it’s theoretical and has been called biased in favour of the biggest companies for which it posts results (first Apple, then Intel). I need to look up the source for this.
There is no noticeable performance difference between 395fps and 343fps for any gamer. If you dial up the graphic quality to medium and high, the difference evaporates because of course a game should be so taxing for the GPU that the CPU should make as little difference as possible, with the only requirement of keeping up with the demand of running the game. In any significant high-enough graphic setting, the difference between Intel and AMD evaporates. This even used to turn around for games requiring more than four physical cores or eight SMTd threads, but Intel has mitigated this by selling 8C/16T SKUs now.
This trend of AMD Zen 2 and Intel i9 9900K(S) being on par in gaming is new, because it wasn’t like that with Zen and Zen+ where Intel still had a significant lead in a lot of games. This has dried up though.
So yes, I will agree there are benchmarks to be found where current Intel scores better in singlethreaded loads than AMD Zen 2. I will not agree that those results have any meaning in real-life, except for some corner cases. Most of those results are either from theorical benchmarks or for benchmarks where the load tested is so small or short that even the current difference can’t really be felt.
While you are implying that AMD Zen 2 isn’t good enough because of its poor singlethreaded performance, I’m implying that it really doesn’t matter as much as you say.
A/ Your expectations about seeing AMD trumping Intel just because of a node shrink are a bit unrealistic, because as you have probable seen now Intel have been optimising their 14nm node for ages and by throwing lots of money at it, so much even that their current node shrink isn’t even able to complete with their 14nm products. If Intel can’t do it, why should AMD with less than 10% of Intel’s R&D budget?
B/ Fair enough.
C/ Why must AMD be ahead of Intel? Why is equally fast not good enough? You must realise that having a product that’s equally fast in some cases while being much better in other cases for the same amount of money or even lower is actually a good thing. You imply that AMD fails to catch up to Intel in most desktop workloads. Which workloads are you talking about? In what way will they be hampered by using AMD Zen 2? I already pointed out that most singlethreaded benchmarks where Intel wins don’t extrapolate to desktop workloads where a difference can be felt.
I feel you’re either misunderstanding the enthousiasm for finally having a real competitor in this rather boring landscape that is the x86 CPU market, or you aren’t being fair in critising AMD and their supposed weak points, while not saying anything how Intel has bled the market dry for ten years without any real improvement and how they are the cause for this stagnation and the hardship we consumers had to endure.
Gargyle,
I thought I was clear, but I specifically chose the AMD CPU with the best performance on each test because I didn’t want to be accused of not using the best numbers for AMD. However feel free to compile the scores any way you like.
Don’t say 0%…that’s exactly the sort of hand waving I’m asking people to avoid.
In terms of your claim though, if you have a process that only takes 5ms on one CPU, and half the time on a second CPU that’s 100% faster, that difference will be imperceptible. Heck a third CPU could be 1000% faster and that difference would be imperceptible too. Obviously if the CPU duty cycle is so brief as to not matter, then that’s fine to say you don’t need the additional single threaded performance, but in those cases you probably won’t benefit from more cores either. I’ve already stated repeatedly that AMD’s single threaded performance is good enough for most people, we both can agree on that, but the fact remains that it is not on par with intel. That’s all I’m saying.
Obviously nobody cares about running world of tanks at 300+FPS and therefor you could throw out all benchmarks once they reach ~100+FPS given that such high frame rates aren’t useful for humans, but with all due respect this kind of misses the point of the benchmarks as a tool for measuring relative performance. It’s not the absolute frame rate that matters so much as the system bottlenecks they measure. If you want to disregard the gaming scores anyways, go ahead, but the data is consistent in that it shows the single threaded CPU bottlenecks kicking in before core count.
I agree the difference isn’t likely to matter much to ordinary users, I’ve already said ryzen CPUs are good enough despite lower single threaded scores. I can accept that part of your argument, no problem. However this notion that AMD is “on par” with intel regarding everything that is bottlenecked by single threaded performance is false. Can you admit that 1) AMD still has work to do in order to catch up and and be on par with intel’s single threaded performance and 2) adding many more cores will not do anything to improve performance in games & applications that are bottle-necked by single threaded performance? We *should* agree, because this is all true.
There are plenty of applications that hit single threaded bottlenecks. I’ll mention two that I encouter regularly: the gimp filters on a huge image and searching emails in thunderbird. I didn’t realize this was a problem for people, but my wife was experiencing performance issues with a huge excel workbook with complex formulas. Guess what, it exhibited a single threaded bottleneck.
The truth is many desktop application do not take advantage of many cores and will max out one single CPU for a prolonged period of time. Of course, this isn’t all applications, but are you really going to say these are irrelevant corner cases that have no meaning in real life? Come on now, we all know better than that.
That’s an interesting perspective. Are you suggesting (albeit indirectly) that 14nm may be better than 7nm for maximizing single threaded performance?
AMD doesn’t have to be ahead of intel for single threaded performance and being equally fast would be good enough. The main point that I’m trying to make is that the benchmarks show that AMD isn’t there yet!
Please stop suggesting that AMD is”equally fast” at single threaded loads when we know it’s not true. AMD could reach intel’s single threaded performance with a speed bump, but the data is clear that they aren’t there yet.
You got that wrong, I’m not critisizing AMD, I’m criticizing the portrayal of AMD as the uncontested winner when the truth is that unbiased results are more mixed. AMD blows away intel on highly multi-threaded tasks, but they’re still lagging behind intel for single threaded.
Yes, I don’t disagree with these opinions, intel has become stale and AMD’s progress is solid. Strong competition is better for innovation.
I do think your argument is of interest for those who need / care about single-threaded performance.
But, I’m not sure it’s a practical concern for *most* people. If you’re editing video, using a supercomputer, or doing things where multi-threaded performance counts, then you’ll more likely be willing to deal with less single thread performance for the overall bandwidth.
I know this is anecdotal, but I’m sure many can relate:
In my case, I want Android Studio to compile as fast as possible. My work laptop needs a keyboard repair, so I got a loaner. The former has a quad-core i7 + 32GB RAM. The latter has a dual-core i5 + 16GB RAM. The former can compile in <2 min for a full compile. It takes seconds on an incremental build. The latter took 10+ minutes on a single build and the incremental build was quite long — 100% CPU utilization the entire time in both cases. So, a little better single threaded performance would have done nothing appreciable for me.
Even when talking about a multi-tasking environment, single threaded performance seems more moot. How many people have Chrome/Firefox (with however many tabs) open, maybe a video playing, using Android Studio / Xcode / Office products (MS, Google, whoever), maybe an AV scanning in the background, etc. With more cores, there's fewer apps to preempt and more time with each going at full speed. That, to me, seems like it would create better *real world* single-thread performance *per app* than what a single-core pumps out on a benchmark.
I don't mean to sound argumentative. Rather, I'm just wondering if the average person cares as much about single-thread performance as they do the overall CPU bandwidth for many threads / processes. And if the average person wants better CPU bandwidth, what will device makers naturally choose to increase sales?
cacheline,
I completely agree with one caveat, many of those problems that benefit from parallel algorithms are actually good canddiate for GPGPU which can perform better than CPU cores. As a software engineer embracing GPGPU architectures may bring more gains than highly multicore CPUs.
I don’t disagree with that, however we do need to consider diminishing returns. Going from one core to two enabled a huge reduction in CPU contention. Two to four, less so. Four to eight, outside of specialized tasks, the instances when all of those cores will be in use simultaneously in a desktop is very rare. If you have a build server then having more cores is great. I have my own anecdotal example too: for one project we used a 24core xeon CPU to analyze raw RF data streams in real time. That was some years ago but today these ryzens could have done the job for much cheaper. A GPGPU solution would probably have been better, but requirements were that it fit in a 1U rackmount server. Anyways, I’m not saying they’re never useful to anyone, they certainly can be. But most desktop users will hit their computer’s single threaded bottlenecks much more frequently than the core count bottlenecks.
Alfman,
I feel that you are completely dismissing the potential and advantages of having a high core/thread count on x86 CPUs.
You say that contention relief suffers from diminishing results. I have read since the launch of Zen 1 in March 2017 that people upgrading from 4C/8T Intel to 8C/16T AMD Zen 1 feel liberated from the constraint of having such a low number of cores in today’s (well, 2017 I mean) world of computing. A lot of games were already saturating a 4C/8T setup so much that they were suffering from nasty framedrops that aren’t seen in system with a higher core/thread number, especially while that system is processing a significant alternative load. High average framerate with nasty framedrops is worse for the perceived fluency of a game than having a lower average but more stable framerate. That is also the reason why a lot of reviewers are now including 99th percentile frametimes into their benchmarks, because average framerates really don’t paint the whole picture. That is also one of the strong points of AMD Zen, Zen+ and Zen 2: equal or better frametimes than Intel, which accidentally is also the metric that matters most for gaming.
The problem with GPGPU support is that the load must fit the constraints of that platform and that the software must be adapted to use that platform. Since there is no popular unified GPGPU platform yet, we can’t expect this to happen anytime soon, so high core count x86 SKUs are here to stay for another couple of years at the least.
Gargyle,
That’s really not fair, cacheline gave an example where high core counts helps and I agreed with him, and then I also gave my own example, and of course there are many others too. My point has never been that high cores can never help. My point was that 100% single core bottlenecks are reletively more frequent than multicore bottlenecks.
Correction:
Ugh why is the wordpress 5 minute edit feature gone? Possible to get it back?
Ah so it’s not just me thinking I’ve gone mad while searching for edit functionality that isn’t there.
For similar price ranges the reviews generally have the performance to be to close to care in desktops. It is annoying how AMD fans always overhype their progress, then AMD under-delivers. Still no driver update with ray-tracing for their existing GPU lines as AMD has promised. There’s a rumor of a $699 6 core budget gaming laptop in coming in early 2020 from a site with 0 credibility, etc. I most certainly don’t see a huge leap forward from these benchmarks, quite the opposite as the iris GPU is besting the integrated AMD GPU in a lot of these tests. Other than that we see the exact same results we’ve seen for years. Better single-threading from intel, higher power draw from AMD, etc.
Although I’m still holding out hope for a 6 core AMD laptop processor early next year as that should drop prices considerably and leave only the nvidia rtx GPUs overpriced.
dark2,
I agree with you about the price factor. AMD has long had the better deals.
I really hope AMD makes progress on the GPU side because consumers are in desperate need for more competition in the GPU space. Nvidia has inflated it’s prices to ludicrous levels.
About the overhyping, it can be annoying, but that’s human nature and I think it’s a core tenant for marketing departments everywhere, haha.
Did you read the article carefully?
The current situation is complicated for both, but AMD is in a way better shape than it has ever been, and it keeps making sizable gains with every new generation, Intel not so much.
For now, while Intel is still in trouble trying to fix all that is still not near perfect with their node size shrink, AMD is blowing them away on servers (where it really hurts) and on desktops. Hopefully, next generation of AMD chips will have AVX512 too and AMD mobiles chips will finally catch up their desktop offerings.
Of course, I hope Intel succeed on fixing their production process as more competition is better to everyone (well, except for them).
acobar,
I agree and I’ve stated that AMD is making more progress than intel.
Yes, I’ve stated before that these high core ryzen chips are great for servers that naturally tend to service dozens or even hundreds of independent processes. On the desktop, it’s more of a mixed bag for reasons already explained elsewhere.
https://cpu.userbenchmark.com/Compare/AMD-Ryzen-9-3950X-vs-Intel-Core-i9-9900KS/4057vsm929964
I will concede your point that AMD could be better prepared for this upcoming year. Both intel and AMD are in a race for fastest per core performance under 14nm and you could be right about AMD gaining more ground than intel in upcoming CPUs, I have no evidence to the contrary. I only ask one thing from everybody, and that is to respect the benchmark outcomes whatever they are.
By now, I hope that I’ve gotten my point across, haha. I’m not anti AMD, I’m just trying to balance out what seems to be prominent favoritism.
“But Intel’s Ice Lake platform runs circles around it.”
There’s some bias showing in the article. Actually looking at the benchmarks, the vast majority are neck and neck, with AMD winning more of those than Intel. There’s a handful where the Intel is decently faster, say 10% to 25% or so. And there were only TWO tests where the Intel “ran rings around” the AMD, being twice as fast on one single synthetic CPU benchmark, and a bit more than that on one gpu + physics benchmark. Overall, the average person in their daily usage is probably not going to see any difference at all between the two.