• WANTED: Happy members who like to discuss audio and other topics related to our interest. Desire to learn and share knowledge of science required. There are many reviews of audio hardware and expert members to help answer your questions. Click here to have your audio equipment measured for free!

Intel destroyed by AMD, let's talk like audio

Racheski

Major Contributor
Forum Donor
Joined
Apr 20, 2020
Messages
1,116
Likes
1,699
Location
Chicago
Benchmarks are benchmarks.
Without a (fairly) standardised set of testing methods, one simply cannot reliably compare the performance of something.

A car does 0-30 mph in x seconds on a hot sandy beach, and another does 0-62 mph in y seconds, on rainy day on grass. What use would x and y be to anyone, when comparing the two vehicles are for very similar purposes???

Intel have always had an instructions per clock advantage, so MHz for MHz, on a single thread, the Intel chip would complete a task quicker than an AMD chip with the same clock speed. The waters all got muddied with boost clocks, turbo, throttling etc. All well and good have x GHz single core speed, but when it tails off after a few minutes, that sucks.

I was looking to build a new PC to replace my old Intel Core 2 Duo E6750 machine. I started at Intel i7 4770k and kept looking at every new release and many things put me off. I didn't like the motherboard layouts, and I didn't like the shoddy putty being used to conduct heat from the chip die to the heat spreader.
AMD Ryzen and X370 chipset came along, and these seemed to offer everything I wanted in a platform. The heat spreader is soldered to the die, so I'm happy about that. The ASrock Tai-Chi X370 motherboard layout ticked all of the boxes. There are a couple of small teething issues, some BIOS tweaks are a bit odd, temperature reporting using an offset (really, still doing that?), a difficulty in getting exactly what the core voltage is supposed to be and how to read it without physically probing the board. Sure, the equivalent Intel i7 chip would be a little quicker in single threaded workloads, but I doubt I'd notice in real world use.

...and here is the thing - real world use.
I've read a couple of reviews, and one site did a huge comparison of performance on similarly spec'd machines AMD vs. Intel with the then current Adobe suite.
Benchmarks and stop watches showed which was faster in what areas and specific tasks (different filters in Photoshop maybe multi-threaded, despite a generally single threaded environment). But, the differences were not that huge really. A few seconds when a task took a few minutes - I doubt that anyone would really notice that on their home machine if they didn't have a stop watch.

The article mentions laptops as an example where benchmark performance stats. aren't everything. I agree, you are buying a product where usability, form factor, battery life, weight and other factors are important. It's like having the cleanest, most perfect DAC on the planet, but it's a stupid shape, weighs so much that four beefy fellas need to move it around, lights up like Blackpool tower as soon as you turn in on and the remote control randomises it's buttons every time you press a button. It'd be utterly unusable to the vast majority of people, but it measures soo well.
When it's about desktop PC components though, nah, benchmarks are important.
So what AMD chip did you get?
 

Svperstar

Senior Member
Joined
Jul 4, 2018
Messages
342
Likes
222
I ran all AMD from 2000-2015. When I went laptop only Intel had the best or only game in town.

Looking to upgrade my 2016 laptop this year but sadly there were no Ryzens with 2070 or above level graphics. Taps out at 2060. I don't know why AMD and the laptop companies couldn't come up better unless Intel bribed them not too.

Anyway. I ordered a model with a Intel 10875 & 2070 Super. Nice upgrade over my 6700hq & 1070. Should ship this week :)
 

blueone

Major Contributor
Forum Donor
Joined
May 11, 2019
Messages
1,179
Likes
1,494
Location
USA
How can someone allergic to BS support Intel? The company that still doesn't use ECC for their caches or support ECC RAM on their non Xeon lines...

Is there evidence that ECC for on-die or in-package caches is worth the performance and energy consumption costs?
 

blueone

Major Contributor
Forum Donor
Joined
May 11, 2019
Messages
1,179
Likes
1,494
Location
USA
Worth it to whom?

Worth it to any microprocessor. Or is it like using silver interconnects. Measurably better, but perhaps insufficiently so to justify the additional cost.
 

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
Worth it to any microprocessor. Or is it like using silver interconnects. Measurably better, but perhaps insufficiently so to justify the additional cost.
Depends on the application. What are the consequences of a fault?
 

Spocko

Major Contributor
Forum Donor
Joined
Sep 27, 2019
Messages
1,621
Likes
2,999
Location
Southern California
I meant single threaded gaming performance, which takes advantage of the Intel's ring bus and the lowered memory latency - this shortcoming of AMD (way less than 10% in real terms if you know how to tweak your b-die memory) compared to Intel should be fixed completely in Zen 3. I hear rumors of 30%+ improvements in singled threaded performance from improvements in memory, IPC, clock and core packaging...
THIS is the last stand for Intel - gamers who require every last ounce of single threaded performance for that extra 10FPS (whether competitive or just vanity), it nevertheless is the only advantage Intel currently has. But for everybody else who needs the CPU for general productivity applications (multi-core and multi-thread) with decent gaming performance? AMD is sooo far ahead of Intel it's a joke. We do a lot of video editing, and Intel gets zero respect. Go to Puget Systems and they have some great tests comparing the best CPUs/GPUs for every budget. Sadly for Intel, AMD's next release this fall is expected to have significantly better single threaded performance that eliminates Intel's one last advantage. Sadder still, AMD is already preparing its 5 nanometer CPU for release in 18 months while Intel is still trying to get their act together.
 

blueone

Major Contributor
Forum Donor
Joined
May 11, 2019
Messages
1,179
Likes
1,494
Location
USA
Depends on the application. What are the consequences of a fault?

It depends, but since caches, even the large last-level caches, are usually fully utilized, it could be pretty bad. Corrupted data, OS or application crash. The question is, what is the probability of a cache line corruption? On-die or in-package, very low apparently. Otherwise Intel CPUs wouldn't make it in cloud data centers. Interestingly, Itanium CPUs had ECC on the second level caches, but HP designed those CPUs, not Intel, so my guess is HP thought ECC on the cache was required for mission-critical CPUs.

Even in client DRAM the probability must be pretty low, because the number of DIMMs is small, often just two. In servers, where 256GB of DRAM is nothing special these days, so you might have 16 or 32 DIMMs, it's more worth it. Personally, I wouldn't want to be without it. And the ECC logic is on the DIMMs and not in the CPUs, so it's not burning CPU power budget on those hot Intel sockets.
 

mansr

Major Contributor
Joined
Oct 5, 2018
Messages
4,685
Likes
10,700
Location
Hampshire
It depends, but since caches, even the large last-level caches, are usually fully utilized, it could be pretty bad. Corrupted data, OS or application crash. The question is, what is the probability of a cache line corruption? On-die or in-package, very low apparently.
Right, what are the consequences of a system crash and how rare do you need it to be? If the alternative to ECC is (more) redundancy at the system level, then it's cheaper and thus worth it.
 

q3cpma

Major Contributor
Joined
May 22, 2019
Messages
3,060
Likes
4,416
Location
France
It depends, but since caches, even the large last-level caches, are usually fully utilized, it could be pretty bad. Corrupted data, OS or application crash. The question is, what is the probability of a cache line corruption?
Probably very low. But well, I put my seatbelt on even if I never had an accident; wouldn't want a silent file corruption due to this.
On-die or in-package, very low apparently. Otherwise Intel CPUs wouldn't make it in cloud data centers.
Xeons have them, though. And it's really the same as saying that speakers can't be bad otherwise you wouldn't find them in studios; even amongst "professionals", the people who know and can convince the decision makers of taking the risk of a less popular brand are few. Which is exactly why Intel is pulling the "trusted proved brand" stunt.
The HPC segment is getting bad for Intel, though.

Now, this isn't the biggest problem at all, just mentioning a small part of the entreprise/consumer featureset divide that Intel intentionally fosters.
 

blueone

Major Contributor
Forum Donor
Joined
May 11, 2019
Messages
1,179
Likes
1,494
Location
USA
Xeons have them, though.

That's an interesting question. I'm pretty sure Xeon Platinum CPUs have ECC caches, because the old E7 series Xeons did, but I can't find even one reference to ECC on caches for Xeon Silver and Gold CPUs. Can you?
 

q3cpma

Major Contributor
Joined
May 22, 2019
Messages
3,060
Likes
4,416
Location
France
That's an interesting question. I'm pretty sure Xeon Platinum CPUs have ECC caches, because the old E7 series Xeons did, but I can't find even one reference to ECC on caches for Xeon Silver and Gold CPUs. Can you?
Indeed, I may have been hasty. The only things I got are:
Zen+: https://en.wikichip.org/wiki/amd/microarchitectures/zen+#Memory_Hierarchy
L0 µOP cache: Parity protected
L1I Cache: Parity protected
L1D Cache: SEC-DED ECC
L2 Cache: DEC-TED ECC
L3 Cache: DEC-TED ECC

Zen 2: https://en.wikichip.org/wiki/amd/microarchitectures/zen_2#Memory_Hierarchy
L0 Op Cache: Parity protected
L1D Cache: ECC
L2 Cache: ECC
L3 Cache: ECC

Xeon E7: http://www.intel.com/content/dam/ww...te-papers/xeon-e7-family-ras-server-paper.pdf
"ECC is used to protect processor registers, processor caches, and system memory from
transient fault" with an implication that SEC-DED is used.
Not being familiar with Intel's huge roster with obscure naming, I "think" that E7 was their flagship? So no dice about the lower Xeons. It would be surprising, though, as the entire Xeon line is for workstations, servers and compute farms/supercomputers, which are all used for workloads that can suffer greatly from a single fault.

That marketing flyer also points to an interesting study about the rate of errors on RAM: http://www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf
 
Last edited:

blueone

Major Contributor
Forum Donor
Joined
May 11, 2019
Messages
1,179
Likes
1,494
Location
USA
Indeed, I may have been hasty. The only things I got are:
Zen+: https://en.wikichip.org/wiki/amd/microarchitectures/zen+#Memory_Hierarchy
L0 µOP cache: Parity protected
L1I Cache: Parity protected
L1D Cache: SEC-DED ECC
L2 Cache: DEC-TED ECC
L3 Cache: DEC-TED ECC

Zen 2: https://en.wikichip.org/wiki/amd/microarchitectures/zen_2#Memory_Hierarchy
L0 Op Cache: Parity protected
L1D Cache: ECC
L2 Cache: ECC
L3 Cache: ECC

Xeon E7: http://www.intel.com/content/dam/ww...te-papers/xeon-e7-family-ras-server-paper.pdf
"ECC is used to protect processor registers, processor caches, and system memory from
transient fault" with an implication that SEC-DED is used.
Not being familiar with Intel's huge roster with obscure naming, I "think" that E7 was their flagship? So no dice about the lower Xeons. It would be surprising, though, as the entire Xeon line is for workstations, servers and compute farms/supercomputers, which are all used for workloads that can suffer greatly from a single fault.

That marketing flyer also points to an interesting study about the rate of errors on RAM: http://www.cs.toronto.edu/~bianca/papers/sigmetrics09.pdf

That's what I found. The E7 and now Platinum series are the top of the Xeon line, and are intended to displace IBM Z-series systems, Itanium-based systems, and old Unix/RISC platforms, so they have several so-called "enterprise" and "high availability" features. ECC on caches is one of them. And they also support a high degree of scale-up computing, both on-chip and multi-chip. For the old E3/E5 and now Silver and Gold CPUs I'm guessing - and this is just a guess - that the caches use parity bits but not ECC. Parity checking is cheap and efficient to implement, and silent data corruptions are unacceptable, so I suspect they detect but do not correct cache errors, but do support ECC DIMMs. The fact that Intel doesn't talk about ECC on caches for Silver and Gold CPUs tells me they are unlikely to have that capability. Intel markets everything they think is important.

There have been studies published that say single-bit cache errors decline in frequency as fabrication geometries get smaller, and that metal packages, like most CPUs have, reduce single-bit errors in caches. Also, scale-out software, as with most cloud applications, are better designed to handle memory errors than the highly scaled-up older enterprise (especially financial) applications. For the cloud it appears as if detection of very rare errors is good enough.

Of course, my guesses could be incorrect, because as you found, facts are hard to come by.
 
D

Deleted member 16502

Guest
I refuse to buy Intel simply based on their history with bribery, customer deceit, use of inferior thermal engineering, socket deprecation for any infinitesimal processor line upgrade and some more.
 

VintageFlanker

Major Contributor
Forum Donor
Joined
Sep 20, 2018
Messages
4,940
Likes
19,692
Location
Paris
AFAIK the Intel 9700K generates less heat than a Ryzen 3800x. This could be the only reason to pick an Intel CPU over AMD when it comes to an SFF build.
Ther's absolutely no temp/airflow issue with the Lazer3D LZ7 XTD case. It is only feed by a 140X15mm intake fan, and a single 92X25mm exhaust one, wihich is enought considering the overall low volume here:

Temps.jpg


I also tried dual-fan configuration with my Noctua U9S:

IMG_20200520_142127.jpg


...But it only lowered CPU temp by 3-4°C, with much more noise. At last, I prefered the single fan setup, with the second 92mm as exhaust fan, much quieter that way and also pretty cool.

The gaming performance delta is neglible and within variance imho
Not to me. I'm the kind of guy who cares for this few more FPS.;)

When it comes to any other task, AMD is better though.
Sure thing!
 

Rizzle

Member
Joined
Jun 9, 2020
Messages
82
Likes
117
Ther's absolutely no temp/airflow issue with the Lazer3D LZ7 XTD case. It is only feed by a 140X15mm intake fan, and a single 92X25mm exhaust one, wihich is enought considering the overall low volume here:

[image removed]

I also tried dual-fan configuration with my Noctua U9S:

[image removed]

...But it only lowered CPU temp by 3-4°C, with much more noise. At last, I prefered the single fan setup, with the second 92mm as exhaust fan, much quieter that way and also pretty cool.
That's what I said, the 9700K generates less heat than a 3800X, which is better when using an SFF case.
Not to me. I'm the kind of guy who cares for this few more FPS.;)
With variance I meant that I'm not sure whether there is a significant difference, or it might be coincidence, I'm not sure. The performance difference is fully dependant on the games/settings.

I sometimes do wish I had an Intel, just for the Thunderbolt 3 support. But AMD's implementation of TB3 is slowly rising.
 
Top Bottom