News:

Willkommen im Notebookcheck.com Forum! Hier können sie über alle unsere Artikel und allgemein über Notebook relevante Dinge disuktieren. Viel Spass!

Main Menu

Post reply

The message has the following error or errors that must be corrected before continuing:
Warning: this topic has not been posted in for at least 120 days.
Unless you're sure you want to reply, please consider starting a new topic.
Other options
Verification:
Please leave this box empty:

Shortcuts: ALT+S post or ALT+P preview

Topic summary

Posted by Valantar
 - April 04, 2020, 15:48:54
Quote from: william blake on April 04, 2020, 13:18:09
Quote from: Valantar on April 04, 2020, 10:36:57
Ah, the classic cherry-picking of benchmarks that characterizes someone who desperately wants their factually wrong opinions to be right
very serious question, are you high?
what i said was: speed/productivity per frequency is equal between amd and intel for an average user.
exactly you, not me, desperately wants chery-picking to prove me wrong, and i have no idea why.
show me some average pc program package which measures ipc, i dont mind.
It's quite impressive that you don't seem to remember what you wrote six posts further up this thread. Let me remind you:
Quote from: william blake on April 03, 2020, 18:56:32
we should not care about spec since we have browsing, games and photoshop to measure our cpu speed. please and thank you.
This is in no way equal to saying "speed/productivity per frequency is equal between amd and intel for an average user." You are saying "we don't need a proper benchmark, we should use these three tests that I have chosen, that is enough." It is also worth noting that you switch between saying "IPC" and "CPU speed" - these two are not the same! CPU speed is (roughly) IPC multiplied by actual CPU frequency, at least for single-threaded loads.

Before the above quote, you said
Quote from: william blake on April 03, 2020, 13:12:56
pretty safe and fair to say ipc is equal. some amd advantage if latencies does not matter, some intel advantage if they matters.
which I then responded to by providing you with test data demonstrating that this is untrue. IPC between Zen 2 and Coffee Lake (and other Skylake-based chips) is not equal, but on average about 7% in favor of AMD according to an industry-standard benchmark. Yes, of course there are scenarios where Intel has better IPC (such as 462.libquantum in AT's testing), as different workloads stress different parts of the core, which means different architectures will perform differently. A different benchmark selection would obviously also present different results. This is why a properly developed benchmark application runs a wide range of different workloads, to ensure that as many scenarios and types of load as possible are represented - and that these loads are based on real-world examples, with the choice of benchmark not being done by any person or small group of persons but by a wide selection of people from across the computing industry.

Now, you could of course argue that the workloads in which Intel is better are more important (though I haven't seen you actually present any arguments for this), but that doesn't change the fact that overall Intel needs much higher frequencies to beat AMD's chip - which, again, tells us that AMD's IPC is better.

As for cherry-picking: you do understand what that term means, right? I am referring to an thoroughly documented industry-standard benchmark which includes 18 different tests. You, on the other hand, said
Quote from: william blake on April 03, 2020, 18:56:32
we should not care about spec since we have browsing, games and photoshop to measure our cpu speed. please and thank you.
In other words: My preferred test has 18 workloads, yours has three. Mine is an established and well documented industry standard developed by an industry-wide organization, yours is seemingly pulled out of thin air based on being workloads where Intel historically does well. Now who is cherry-picking?

"Browsing, games and Photoshop" does not tell us anything about IPC unless you both broaden the test selection and heavily refine the testing criteria (for example, "Photoshop" is not a single workload, but hundreds of different ones with variable threading, cache/RAM utilization, etc. Which one(s) do you pick? Why?). There are both too few tests and too many variables in your choice for that to be possible (especially in games, which are generally more reliant on GPU performance). Browsing is also highly bursty, which means that to use such a workload to determine IPC you would need to disable all boost modes of the CPU, which would again negate Intel's advantage of higher boost clocks, which is largely why Intel tends to win browsing benchmarks. Remember, IPC means Instructions Per Clock, i.e. it compares how many operations a CPU core is capable of processing for each clock cycle. Clocking higher with lower IPC can still be faster overall - which Intel demonstrates in some workloads. In fact, the 9900K in Anandtech's SPEC testing wins 14 out of 18 tests - but it does so by running at much higher clocks than the 3900X it's compared to. The scores divided by their respective clock speeds then tell us that the AMD chip is, in general, processing more work per clock than the Intel chip. That doesn't mean that the AMD chip is faster in this case - in fact it's only faster in four out of 18 tests! - but it does have higher IPC. That is beyond question.

How this extrapolates into real-world performance is of course both dependent on the actual clocks of the chips people are using and how many threads the applications in question actually use. Both of which has extreme variability. But in general, we know that Zen 2 consumes less power than Coffee Lake (and now Comet Lake) while performing at the same level, partly due to its slightly higher IPC and partly due to its process advantage. AMD also has better SMT than Intel, and generally gives you more cores/threads for the same money. Intel on the other hand can boost quite a lot higher (a 9900K with good cooling can run at a steady 5GHz, while a 3900X or 3700X will generally not exceed 4.3GHz in heavily threaded loads. Whether these properties gives the ultimate performance advantage to Intel or AMD depends on the application in question. But none of this brings into question whether AMD has better IPC - it does. Period.

As for "some average PC program package which measures IPC" - sorry, there's no such thing. It doesn't exist. SPEC is the closest you'll get, as measuring IPC (and not just absolute performance) requires a controlled test environment which most users won't be able to provide. Of course you can roughly measure IPC on your own by having two computers you want to compare, manually running them at the same clocks (how fast or slow is irrelevant, as you're not measuring absolute performance and IPC-measuring workloads should scale linearly with frequency, so something reasonable like 3GHz all-core is perfectly fine), and then running your desired benchmarks on each. Of course you also need to stick to single-core workloads unless you want multi-core scaling, the Windows Scheduler, SMT, and other uncontrollable variables to foul up your results. Ideally you would also need to compile your own applications to negate any compiler-based advantages (such as benchmarking an AMD chip with an application complied with an Intel-optimized compiler, or the other way around). Another possible method is letting frequencies vary, but logging them continuously across the run and then normalizing the score for the average frequency over each benchmark run - though this would require advanced logging software and some time-consuming excel work to produce anything that can be called comparable IPC numbers.
Posted by william blake
 - April 04, 2020, 13:18:09
Quote from: Valantar on April 04, 2020, 10:36:57
Ah, the classic cherry-picking of benchmarks that characterizes someone who desperately wants their factually wrong opinions to be right
very serious question, are you high?
what i said was: speed/productivity per frequency is equal between amd and intel for an average user.
exactly you, not me, desperately wants chery-picking to prove me wrong, and i have no idea why.
show me some average pc program package which measures ipc, i dont mind.
Posted by _MT_
 - April 04, 2020, 12:36:24
Quote from: Valantar on April 03, 2020, 15:12:26
In other words, 5.3GHz for two cores is around 135W, which will be entirely unsustainable for anything else than the biggest laptop designs out there.
...
Zen 2 has a >7% IPC advantage over Coffee Lake on desktop according to AnandTech's testing in SPEC2017, which covers a wide range of different workloads that stress both FP and INT. It is thus safe to say that IPC is not equal, even if 7% isn't massive.
...
Neither Intel nor AMD define TDP as a direct function of chip power draw (if they did, turbo modes would be impossible). The difference is that AMD currently has a ~85% delta between base TDP watts and boost power draw in their HS chips, while Intel has a 300% delta between the same numbers for the "45W" 10980HK and its PL2 Turbo power. Both are disingenuous; one slightly, the other massively. Only one of these will deliver close to the advertised performance at the advertised power.
At these numbers, cooling isn't the only limiting factor. Anything drawing 135 W can't really be used on battery. Partly because of the unfortunate TSA (or whoever is responsible for it) limit on built in batteries. And even without it, it's so large that an appropriate battery might require wheels. :-)

One problem with measuring IPC is that caches get in the way. I think there is consensus that Zen2 has superior caching. Intel still might be able to achieve higher IPC, they just get more cache misses and can't feed the processor as well as AMD. End user doesn't have to be concerned with such details, but it actually matters if you want to dig into it.

Shouldn't that be +200 % (45 vs. 135 W) if the AMD is +86 % (35 vs. 65)? Also, if you want to use percentage difference (to which you allude by "delta"), then it's 100 % (90/90) for Intel and 60 % (30/50) for AMD (that's the difference divided by average).

Intel really lost on two counts. They didn't push for higher core counts. Which would force them to solve some problems (and I imagine that's the reason they avoided it) AMD had to solve and solved. And they hit a rock with their 10 nm node. They've been struggling with it for what seems like ages now. AMD got very, very lucky there with the timing. In the end, we're the winners. AMD really needed to catch a break.
Posted by _MT_
 - April 04, 2020, 11:39:24
5.3 GHz at 45W, seriously? How can you write such nonsense? How can anyone take you seriously anymore? I thought it's well known that TDP is for base clocks. So why spread this nonsense in a title and fuel false expectations?
Posted by Valantar
 - April 04, 2020, 10:36:57
Quote from: Adam Kullai on April 03, 2020, 22:05:02
Before declaring victory, you should look at what intel says closely: 45 w tdp, but turns out intel measures tdp as the heat generated by the sustained clock, on the other hand and is measuring tdp as the highest possible heat generated, so intel is basically cheating
No they aren't - AMD's definition of TDP is a rather complex calculation, and while boosting their CPUs/APUs pull significantly more power than the TDP number (the 35W 4900HS hits 65W). TDP is roughly equal to steady-state power draw, which will run at base clock or higher if possible under thermal/power constraints.

Quote from: william blake on April 03, 2020, 18:56:32
Quote from: Valantar on April 03, 2020, 15:12:26
Zen 2 has a >7% IPC advantage over Coffee Lake on desktop according to AnandTech's testing in SPEC2017,
we should not care about spec since we have browsing, games and photoshop to measure our cpu speed. please and thank you.
Ah, the classic cherry-picking of benchmarks that characterizes someone who desperately wants their factually wrong opinions to be right. Yes, Intel has a minor advantage in web browsing, but... let's say they have a 50% advantage (which they don't) -that means a 0.5s page load turns into 0.75s, or a 2s load into 3s. The latter is somewhat noticeable, the former not at all - and unless your browsing consists of constantly and rapidly navigating to new pages (which it doesn't), the actual difference matters little as long as it's quick enough. In either case, the perceived difference is much smaller than the percentage would indicate. And of course the real number is nowhere near 50%. In games, Intel needs dramatically higher clock speeds (at much higher power) to eke our a few percent advantage over AMD in high FPS games. And frankly, what is the difference between 200fps and 210? And we know Adobe is working through their application suite to increase multi-core performance as CPU core counts increase - look at how Premiere performance has changed over the past 2-3 years. Photoshop likely won't see as large gains, but can nonetheless improve.

Now, this of course doesn't say that Intel isn't still better in these and some other workloads - no matter how marginal the differences might be - but SPEC also shows that. In memory and cache latency sensitive subtests (such as libquantum) Intel has clear victories. Look at Anandtech's test data. Even with that taken into account the average across different workloads is ~7% in AMD's favor.

Arguing to dismiss outright an industry-standard and well-researched benchmark set because you don't like the results is quite silly. SPEC covers similar workloads to all the cases you mention, and shows where Intel is stronger, but that still doesn't mean they are faster or better overall.
Posted by Adam Kullai
 - April 03, 2020, 22:05:02
Before declaring victory, you should look at what intel says closely: 45 w tdp, but turns out intel measures tdp as the heat generated by the sustained clock, on the other hand and is measuring tdp as the highest possible heat generated, so intel is basically cheating
Posted by Adam Kullai
 - April 03, 2020, 21:58:12
In the picture intel is showing the performance increase in gaming over a 3 year old CPU. But the 3 year old CPU is paired with a gtx 1080 and the new CPU is paired with a freaking rtx 2080 ti, so most of the performance increase is because of the graphics cards, intel is outraging.
Posted by william blake
 - April 03, 2020, 18:56:32
Quote from: Valantar on April 03, 2020, 15:12:26
Zen 2 has a >7% IPC advantage over Coffee Lake on desktop according to AnandTech's testing in SPEC2017,
we should not care about spec since we have browsing, games and photoshop to measure our cpu speed. please and thank you.
Posted by Valantar
 - April 03, 2020, 15:12:26
5.3GHz at 45W is outright false. To quote AnandTech:
QuoteThe base frequency of this chip is 2.4 GHz, and it has a regular 45 W TDP (sustained power), which can be run in cTDP up mode for 65 W. [...] Intel's recommended PL2 (turbo power) for the Core i9 is 135 W, and Intel says the recommended 'Tau' is set to 56 seconds for the i9, and 28 seconds for all the other CPUs.
In other words, 5.3GHz for two cores is around 135W, which will be entirely unsustainable for anything else than the biggest laptop designs out there. Given the temperature targets noted in this news post, 5.3GHz will be gone in the blink of an eye, and 5.1 will last just long enough for the chip to hit 99 degrees and throttle - which will also be very, very quick in most cases. 45W will be at or slightly above base clock - which is how Intel specifies TDP, after all. I'd be impressed if any of the 8-core chips hit even close to 3GHz at 45W. The 65W cTDP-up mode for the 10980HK is at 3.1GHz base clock.

On the other hand the 4900HS in the Asus G14 (yes, this is a lower power SKU of a top bin, so not generally representative - most other SKUs will need slightly more power for slightly lower clocks) boosts to 65W peak for a short while (at top boost), steps down to 55W for about three minutes, before settling in at ~3.5GHz at 35W. Note that this is about .5GHz above base clock. At 35W power draw.


Quote from: william blake on April 03, 2020, 13:12:56
pretty safe and fair to say ipc is equal. some amd advantage if latencies does not matter, some intel advantage if they matters.
Zen 2 has a >7% IPC advantage over Coffee Lake on desktop according to AnandTech's testing in SPEC2017, which covers a wide range of different workloads that stress both FP and INT. It is thus safe to say that IPC is not equal, even if 7% isn't massive. It still means that a 4900HS at ~3.5GHz sustained will match a theoretical Intel chip at 3.75GHz sustained - which none of them come even close to without consuming significantly more power.

Quote from: william blake on April 03, 2020, 13:12:56
Quote from: ariliquin on April 03, 2020, 09:59:49
Where are the flagship gaming laptops at that use AMD? Manufacturers seriously under estimated AMD here and over estimated Intel demand. 
ask lisa su. i heard she said she is very proud.
Your oddly intense dislike for Dr. Su is well established at this point, and really has to make one wonder what exactly it is you have against her considering the turnaround of AMD under her direction. I might be wrong, but this smells of sexism to me. Su is killing it, AMD is killing it, and you seem not to want to accept those facts.

Beyond that, the reason for there being far more Intel 10th gen laptops than Ryzen 4000 ones is more or less twofold: one is that Intel is a behemoth with a de facto monopoly in this segment up until now, and thus OEMs have a lot of familiarity with their platforms and only need minor updates to make a new laptop with a new CPU. Intel also provides huge amounts of marketing support money to their partners. Making a Ryzen 4000 laptop means designing it entirely from scratch, which takes time. The other is that R&D on these laptops mostly started 1-1 1/2 years ago, and back then AMD had nothing to compete in this performance class. While ES silicon and early samples have no doubt been sent to manufacturers long ago to start development of new models, this is still too late to have them ready for launch outside of the few models we have now - all of which have had direct AMD engineering support. Give it six months, and this will no doubt change.

Quote from: william blake on April 03, 2020, 13:12:56
Do you think AMD is any different? According to them thermal watts don't equal power consumption and the law of conservation of energy must only be a guideline.
Neither Intel nor AMD define TDP as a direct function of chip power draw (if they did, turbo modes would be impossible). The difference is that AMD currently has a ~85% delta between base TDP watts and boost power draw in their HS chips, while Intel has a 300% delta between the same numbers for the "45W" 10980HK and its PL2 Turbo power. Both are disingenuous; one slightly, the other massively. Only one of these will deliver close to the advertised performance at the advertised power.
Quote from: william blake on April 03, 2020, 13:12:56as much as i dont like lisa su and robert "4,75ghz" hallock, this type of marketing-enjoy your 5ghz printed on the box while gaming on 3ghz-is not very common.
For once we actually agree on something - this kind of marketing is idiotic. Intel introduced TVB as a response to AMD's XFR, which AMD abandoned with Zen 2, while Intel is introducing it into mobile now as they don't have anywhere to go for performance improvements but increasing boost speeds. AMD never advertised XFR speeds as the main "up to" speed of the chips though, but Zen 2 chips were obviously launched in a state where their on paper boost speeds were nonetheless significantly higher than what users saw. Even if this was mostly fixed by later BIOS updates it is definitely a shady practice.
Posted by william blake
 - April 03, 2020, 13:12:56
Quote from: Mohammed Bhram on April 02, 2020, 14:20:23
Shame on you intel compare between core i7 7th generation and core i9 10th generation
shame? sad and pathetic you mean? a monopoly declares i'm an outsider now after all these years
Quote from: Gianni on April 02, 2020, 19:51:11
But with AMDs higher IPC, their 4.3GHz is on par with like 5.06GHz on an Intel processor ... So does breaking this "5 GHz barrier," mean much?
.
pretty safe and fair to say ipc is equal. some amd advantage if latencies does not matter, some intel advantage if they matters.
Quote from: JMac on April 03, 2020, 01:18:59
Do you think AMD is any different? According to them thermal watts don't equal power consumption and the law of conservation of energy must only be a guideline.
as much as i dont like lisa su and robert "4,75ghz" hallock, this type of marketing-enjoy your 5ghz printed on the box while gaming on 3ghz-is not very common.
btw its funny you said about tdp and consumption difference and mentioned amd but not intel. cringeworthy even.
Quote from: ariliquin on April 03, 2020, 09:59:49
Where are the flagship gaming laptops at that use AMD? Manufacturers seriously under estimated AMD here and over estimated Intel demand. 
ask lisa su. i heard she said she is very proud.
Posted by Kirill
 - April 03, 2020, 12:51:46
How many seconds they can maintain clocks above 5Hz? Two or three?
Posted by gerger
 - April 03, 2020, 12:25:48
No, not these! Gimme those Tiger Lakes goddammit!
Posted by ariliquin
 - April 03, 2020, 09:59:49
YAWN... Theoretical Intel single core performance which is significantly lower than advertised under sustained load due to thermals in laptops does not compete IMHO with true power/performance benefit of 7nm AMD. Where are the flagship gaming laptops at that use AMD? Manufacturers seriously under estimated AMD here and over estimated Intel demand. 
Posted by Vaidyanathan
 - April 03, 2020, 05:41:53
Quote from: Rohan on April 02, 2020, 20:15:51
Now that the laptop processors are unveiled, is there a possible timeframe we can expect the desktop comet lake processors to launch in
Not too far away, I suppose. Earlier rumors said both H and S series launch could be on a similar timeframe. The later they defer the launch, the more ground they are likely to cede to AMD, which is making huge inroads in desktop right now.
Posted by A
 - April 03, 2020, 01:58:02
Quote from: JMac on April 03, 2020, 01:18:59
Quote from: GURVINDER S PARMAR on April 02, 2020, 13:59:50
Intel does NOT have any CPU that can run at these speeds using 45 watts...the marketing they get away with is beyond ridiculous at this point.

Do you think AMD is any different? According to them thermal watts don't equal power consumption and the law of conservation of energy must only be a guideline.

Of course it is different.

AMD's 4900HS is 3ghz and boost clocks to 4.3ghz, a 1.3ghz boost.

The 10980HK is 2.4ghz and boost clocks to 5.3ghz, a 2.9ghz boost.

AMD's boost clocks are far more sustainable than Intel's.

Intel is trying to cheat on benchmarks by showing their single core performance to be higher than AMD (though not by much), but that can only be done on 1 core. And even then not sustainable.

Not to mention AMD is making those speeds at 35W. So my guess is they aren't even showing the full potential of their 45W yet.