Intel Bug - Did this actually tip the balance for any AMD CPUs?

[juanrga]

You're mixing denominators to make Intel look better. Come on man. If you're going to use the R7 as denominator for power, you have to use it for speed as well and vice versa.

So r7 is 32.7% faster while i5 uses 30% less power. Or, i5 is 49% slower while r7 uses 41% more power.

You are the one mixing the baselines. I am using the same baseline for power and performance; I am using the 8400 as baseline:

Your own graphs show that the 1700 is only 32.7% faster than the 8400 whereas consuming 41.3% more power (79.95W vs 56.58W).

If you want a more explicit and formal discussion, let Eff_A denote the efficiency of the AMD chip

Eff_A = Perf_A / Power_A

since the AMD chip is 32.7% faster

Eff_A = ( 1.327 Perf_I ) / Power_A

and with the AMD chip consuming 41.3% more power, we obtain

Eff_A = ( 1.327 Perf_I ) / ( 1.413 Power_I )

So

Eff_A = 0.939 ( Perf_I / Power_I )

or, what is the same,

Eff_A = 0.939 Eff_I

The AMD chip is less efficient. Inverting the relation, you can check that the Intel chip is 6% more efficient.

 

[juanrga]

You... you know he's not going to stop obfuscating, right? It's just the pattern, but I'll say he's very good at talking circles around almost anything.

Yes, AMD is more than competitive now. When I do my new build in April / May, it'll be an AMD system, personally. First time in more than a decade.

If the 1700 consumes 41.3% more power than the 8400 to be only 32.7% faster, then it is obvious that the 1700 is less efficient than the 8400. Blender is not the only workload where the i5 is more efficient. Next you have efficieny graph measured for x264

The i5 8400 is the more efficient chip. It is 11% more efficient than the R7 1700 on all-core load and 99% more efficient than the R7 1700 in single-core load. Your ad hominems will not change anything of this.

No one here is negating that AMD is more competitive with RyZen than with Piledriver! What is being contested here in the forums are some incorrect claims about performance, pricing, efficiency, power consumption, desktop marketshare,... about RyZen products.

Back on topic: the Meltdown/Spectre flaws could make RyZen chips a bit more competitive, but don't expect miracles. Meltdown+Spectre patches are reducing the performance of Kabylake chips by 1--2% on Blender and Handbrake (x264). Even if Spectre patches don't reduce the performance of RyZen chips (which is difficult to believe), the i5 8400 will continue being much more efficient than the R7 1700, And the i7 8700k will continue being faster than the R7 1800X. That is all.

 

[OrangeKhrush]

Your own graphs show that the 1700 is only 32.7% faster than the 8400 whereas consuming 41.3% more power (79.95W vs 56.58W). So your own graphs show that the i5 is more efficient on a throughput workload despite having less cores. In other workloads that aren't Blender the i5 is even much more efficient than the 1700.

Not even bother to reply the rest of your post. It is a waste of time.

it has less threads to power, lower package power needed, but the offset is considerable amount of performance, the 8600K overclocked struggles to match the stock 1600X while after 4.8Ghz Intel's power just falls apart. That is essentially the idea here. In parallelism the 1600 on perf/watt remains better than the 8400 and 8600K by a significant margin, on equal clocks the 8700K is essentially a 1600 but at a 2x the cost premium that is excluding the rest of the system.

Don't bother to reply because your replies are often circular rhetoric that is false but passed as fact.

 

[OrangeKhrush]

You're mixing denominators to make Intel look better. Come on man. If you're going to use the R7 as denominator for power, you have to use it for speed as well and vice versa.

So r7 is 32.7% faster while i5 uses 30% less power. Or, i5 is 49% slower while r7 uses 41% more power. Either way, the r7 is more efficient.

The real way is to measure watt*sec. So 80W for 28.8 seconds vs 56W for 42.8 seconds. so 2304 Ws for r7, 2396 Ws for i5 8400 making the r7 4% more efficient than the i5 8400.

Not only is it 32.7% faster, it's also 4% more efficient while doing so.

You are supposed to pretend to let him think he is correct. Never go full correct.

 

[OrangeKhrush]

No one here is negating that AMD is more competitive with RyZen than with Piledriver! What is being contested here in the forums are some incorrect claims about performance, pricing, efficiency, power consumption, desktop marketshare,... about RyZen products.

yet it is fine when you pass off incorrect information?

As an example mindfactory.de post their quarterly sales per unit reports, AMD outsold Intel units with over 50% of the total sales for about 5 month, if you take the total annual aggregate fo sales for mindfactory alone AMD had around 45% of total CPU's mind factory sold. This reflects desktop pc only maket.

The alcorn article focuses on total addressable markets and AMD did not feature in a lot of those, IE AMD had no presence in the extremely lucrative mobility market, they also do not have any involvement in 30% market space presented by the dGPU market. howver that is set to change with AMD APU's and mobile being released this term and AMD features on Intel mobile parts now.

DT marketshare for AMD if working on small sample is more closer to 50-50, over a broad spectrum of total units sold AMD probably has about 40% of the desktop space in total addressable markets about 12-13% without real epyc data which is probably only due in 2018.

 

[juanrga]

it has less threads to power, lower package power needed, but the offset is considerable amount of performance, the 8600K overclocked struggles to match the stock 1600X while after 4.8Ghz Intel's power just falls apart. That is essentially the idea here. In parallelism the 1600 on perf/watt remains better than the 8400 and 8600K by a significant margin, on equal clocks the 8700K is essentially a 1600 but at a 2x the cost premium that is excluding the rest of the system.

You started discussing the 1800X, after being proven wrong you changed to the 1700, after being proven wrong now change to the 1600.

The R7 1600 has an efficiency of 19.91. The i7 8700k has an efficiency of 19.72. The difference is an insignificant 1% that falls witin the margin of error. But the i5 8400 continues being the more efficient chip with a score of 27.33. The i5 8400 is 37.3% more efficient than the R5 1600.

 

[OrangeKhrush]

You started discussing the 1800X, after being proven wrong you changed to the 1700, after being proven wrong now change to the 1600.

The R7 1600 has an efficiency of 19.91. The i7 8700k has an efficiency of 19.72. The difference is an insignificant 1% that falls witin the margin of error. But the i5 8400 continues being the more efficient chip with a score of 27.33. The i5 8400 is 37.3% more efficient than the R5 1600.

if you use the incorrect computation then yes.

 

[juanrga]

yet it is fine when you pass off incorrect information?

As an example mindfactory.de post their quarterly sales per unit reports, AMD outsold Intel units with over 50% of the total sales for about 5 month, if you take the total annual aggregate fo sales for mindfactory alone AMD had around 45% of total CPU's mind factory sold. This reflects desktop pc only maket.

The alcorn article focuses on total addressable markets and AMD did not feature in a lot of those, IE AMD had no presence in the extremely lucrative mobility market, they also do not have any involvement in 30% market space presented by the dGPU market. howver that is set to change with AMD APU's and mobile being released this term and AMD features on Intel mobile parts now.

DT marketshare for AMD if working on small sample is more closer to 50-50, over a broad spectrum of total units sold AMD probably has about 40% of the desktop space in total addressable markets about 12-13% without real epyc data which is probably only due in 2018.

Minfactory.de only reports sales of one store in Germany. Its numbers don't represent wordlwide marketshare.

Wordwide marketshare numbers can be found in the passmark graph and in the Alcorn article. As stated in the Alcorn article. AMD had a desktop marketshare of 9.9% in 4Q16, just before Zen launch, and increased its presence in desktop to about 12.9% (projection) in 4Q17. But those numbers are before CoffeeLake started stooling desktop marketshare from AMD (check passmark graph). Now AMD marketshare is back to pre-Zen time.

 

[juanrga]

if you use the incorrect computation then yes.

I am using data from the efficiency graph in #43.

 

[OrangeKhrush]

Minfactory.de only reports sales of one store in Germany. Its numbers don't represent wordlwide marketshare.

Wordwide marketshare numbers can be found in the passmark graph and in the Alcorn article. As stated in the Alcorn article. AMD had a desktop marketshare of 9.9% in 4Q16, just before Zen launch, and increased its presence in desktop to about 12.9% (projection) in 4Q17. But those numbers are before CoffeeLake started stooling desktop marketshare from AMD (check passmark graph). Now AMD marketshare is back to pre-Zen time.

I already stated what us baseball analytics term "small sample size", I already stated it is one etailer but on the data of actual sales of desktop PC only AMD controls nearly 50% of the years total units sold. This trend is likely common throughout global resellers. I cannot disclose the information of a friends reselling business but lets say its very similar.

the 9% is total addressable markets and as I explained AMD has very little presence in mobile and server space for now, but as the article covered and market analysts stated the Desktop marketshare is significantly higher. The only data we have in number sold were from mindfactory so they become a bonafide source of actual market data.

Systemmark or whatever it is called is a survey run by benchmarking, most I know don't even use it so and they own Skylake X parts, it is a random census on all clicks to their bench tool and not a product of sales and it is clearly put in their disclaimer. lets just go through this slowly.

• This graph counts the baselines submitted to us during these time period and therefore is representative of CPUs in use rather than CPUs purchased.
• The Quarters are by the calendar year rather than financial. (i.e. Q1 starts January 1st)
• Baselines can be submitted from anywhere therefore these are global statistics.
• We do receive a small number of submissions of CPU types other than AMD and Intel however the percentage is so small as to make it not worth graphing. This combined with rounding off the percentages to 2 decimal places will account for each quarter not always adding up to exactly 100%.
• This chart only includes x86 processors and does not include other chip architectures these manufacturers may sell.
• This chart only includes CPUs installed into PCs and does not include game consoles.
• As the PerformanceTest software only runs on Windows OS and counts on user submitting their benchmarks. This chart may be non reflective of non Windows user base.
• The figures don't represent just new CPU sales. They also reflect to some degree the installed base from the last 5 to 10 years. So large monthly swings in CPU sales will take time to convert the installed base.
• As the chart is updated daily, but data points are for the quarter, the first few days of a new quarter will only have a few samples. i.e. The first few days of a new quarter are less accurate compared to the end of a quarter.

1) representative on pc's in use not purchased, what they are saying is people who run their bench tool but has zero corrolation to sales, can't make a marketshare dependent on who uses this software.

2) self explainatory

3) can be run anywhere and as many times as you want, ie one person can log multiple results to affect the data, brilliant lets spam this benchmark.

4) no relevance

5) boting

6) self explainatory, they have no data on one market AMD feature heavily in but herp derp irrelevant for DT marketshare.

7) they need time to catch up their database because they are so slow, data could be representative of older data more than newer but sit back they will deal with it eventually.

8) their baseline is only likely to be more or less accurate at the end of the month when they have more data to work with, or saying that while they may do things daily, they normally do it monthly.

For the purpose of the whole argument is that point #1 basically means this is not accurate, it just means that many people don't actually use this tool and most that do are Intel users.

 

[OrangeKhrush]

Wait are we now taking motherboard ATX12V over actual EPS12V power numbers.

 

[bobzdar]

You are the one mixing the baselines. I am using the same baseline for power and performance; I am using the 8400 as baseline:



If you want a more explicit and formal discussion, let Eff_A denote the efficiency of the AMD chip

Eff_A = Perf_A / Power_A

since the AMD chip is 32.7% faster

Eff_A = ( 1.327 Perf_I ) / Power_A

and with the AMD chip consuming 41.3% more power, we obtain

Eff_A = ( 1.327 Perf_I ) / ( 1.413 Power_I )

So

Eff_A = 0.939 ( Perf_I / Power_I )

or, what is the same,

Eff_A = 0.939 Eff_I

The AMD chip is less efficient. Inverting the relation, you can check that the Intel chip is 6% more efficient.

Your math is wrong. Just multiply the wattage by the time for total power used - that's what we pay for (you know, kilowatt hours, that's what the electric company charges us for) . The 1700 uses 4% less total power to finish the job (and does it 32.7% faster).

If you're going to enter these debates you better at least learn basic math because right now you look like an idiot.

 

[Lunas]

So far what google has shown me is that gaming performance is minimally impacted, which would seem to indicate that the huge lead Intel has maintained, will continue to be maintained. I'm not running a database.

Just to be clear, i'm not against AMD, or inherently pro Intel. I would love to see actual competition bring down prices on all CPUs. I'm wondering if that is right around the corner yet, or not.

Huge lead? Not so much...

 

[Zinn]

FWIW I returned my 7820X / X299 to Microcenter and bought a Threadripper 1920X instead. The Spectre/Meltdown fixes were actually wrecking my compile performance. Bye, Intel!

 

[juanrga]

Systemmark or whatever it is called is a survey run by benchmarking, most I know don't even use it so and they own Skylake X parts, it is a random census on all clicks to their bench tool and not a product of sales and it is clearly put in their disclaimer. lets just go through this slowly.


1) representative on pc's in use not purchased, what they are saying is people who run their bench tool but has zero corrolation to sales, can't make a marketshare dependent on who uses this software.

2) self explainatory

3) can be run anywhere and as many times as you want, ie one person can log multiple results to affect the data, brilliant lets spam this benchmark.

4) no relevance

5) boting

6) self explainatory, they have no data on one market AMD feature heavily in but herp derp irrelevant for DT marketshare.

7) they need time to catch up their database because they are so slow, data could be representative of older data more than newer but sit back they will deal with it eventually.

8) their baseline is only likely to be more or less accurate at the end of the month when they have more data to work with, or saying that while they may do things daily, they normally do it monthly.

For the purpose of the whole argument is that point #1 basically means this is not accurate, it just means that many people don't actually use this tool and most that do are Intel users.

Passmark estimates marketshare from the benches. If passmark was used alone then it would be taken with a huge grain of salt, but it is being used here only as a complement to the other data. And passmark agrees with the other data.

Passmark shows that Zen only produced a 3.2% marketshare for AMD in the period from 4Q16 to 4Q17. This agrees with the 3% that Mercury research and Rolland found using sales data for the same period 4Q16 --> 4Q17.

Passmark shows that AMD gained marketshare before CoffeeLake. After CoffeeLake launch, Intel has stolen marketshare from AMD. This agrees with sales data from Mindfactory and Amazon, which show Intel is back to #1 in sales.

 

[juanrga]

Your math is wrong. Just multiply the wattage by the time for total power used - that's what we pay for (you know, kilowatt hours, that's what the electric company charges us for) . The 1700 uses 4% less total power to finish the job (and does it 32.7% faster).

If you're going to enter these debates you better at least learn basic math because right now you look like an idiot.

Guess what? You are right on that the R7 is 4% more efficient than the i5 on this workload.

 

[juanrga]

FWIW I returned my 7820X / X299 to Microcenter and bought a Threadripper 1920X instead. The Spectre/Meltdown fixes were actually wrecking my compile performance. Bye, Intel!

That is weird. Compile times aren't affected by the patches

 

[ManofGod]

That is weird. Compile times aren't affected by the patches

That chart means nothing to the person you quoted though. His experience is exactly as he explained and why he went with Threadripper instead. Cannot say I blame him.

 

[Zinn]

That is weird. Compile times aren't affected by the patches

It was not a gcc compile, it was file concatenation and compression compilation based on filesystem watch events in a VM container. My use case involves reloading hundreds of files through a VM network every time I change a file, basically tons of kernel IO. The Meltdown patch made things noticeably more sluggish, the Spectre patch was threatening to make it worse. Security is critical for my work, so I didn't want to fuck with disabling the "fixes."

I was still in the return period and Microcenter was very kind to let me jump ship with no restocking fee - I had been on the fence between Intel and AMD when I bought this a month ago and would never have bought it if I had known this shit was going down. For the money, X399 is a way better platform now.

 

[OrangeKhrush]

Passmark estimates marketshare from the benches. If passmark was used alone then it would be taken with a huge grain of salt, but it is being used here only as a complement to the other data. And passmark agrees with the other data.

Passmark shows that Zen only produced a 3.2% marketshare for AMD in the period from 4Q16 to 4Q17. This agrees with the 3% that Mercury research and Rolland found using sales data for the same period 4Q16 --> 4Q17.

Passmark shows that AMD gained marketshare before CoffeeLake. After CoffeeLake launch, Intel has stolen marketshare from AMD. This agrees with sales data from Mindfactory and Amazon, which show Intel is back to #1 in sales.

Passmark is naively used because not many companies will put their returns up for public access as there is no obligation to do that, in the absence of such incorrect data is pushed as fact

 

[juanrga]

Passmark is naively used because not many companies will put their returns up for public access as there is no obligation to do that, in the absence of such incorrect data is pushed as fact

This sounds as babbling, and it has zero relevance to my post.

 

[OrangeKhrush]

This sounds as babbling, and it has zero relevance to my post.

Very relevant, you want to pass Passmark as a legitimate figure when it has zero corrolation to parts sold, something their very own disclaimer makes clear. Without multiple data sources from other resellers, mindfactory.de remain the best source and it shows near 50-50% sales of desktop CPU's through 2017.

 

[juanrga]

It was not a gcc compile, it was file concatenation and compression compilation based on filesystem watch events in a VM container. My use case involves reloading hundreds of files through a VM network every time I change a file, basically tons of kernel IO. The Meltdown patch made things noticeably more sluggish, the Spectre patch was threatening to make it worse. Security is critical for my work, so I didn't want to fuck with disabling the "fixes."

Ok. And do you know which is the impact of Spectre on ThreadRipper?

I was still in the return period and Microcenter was very kind to let me jump ship with no restocking fee - I had been on the fence between Intel and AMD when I bought this a month ago and would never have bought it if I had known this shit was going down. For the money, X399 is a way better platform now.

My post was only about the technical part of yours. I was interested in what do you mean by "wrecking my compile performance".

It is your money and your job and you have all the right to purchase/return/change anything you want. I only wish you took the correct decision.

 

[juanrga]

Very relevant, you want to pass Passmark as a legitimate figure when it has zero corrolation to parts sold, something their very own disclaimer makes clear. Without multiple data sources from other resellers, mindfactory.de remain the best source and it shows near 50-50% sales of desktop CPU's through 2017.

I already explained you how Passmark marketshare is not being used alone but only as complement to rest of data available. I also explained you how Passmark marketshare confirms that other data.

 

[ManofGod]

Ok. And do you know which is the impact of Spectre on ThreadRipper?



My post was only about the technical part of yours. I was interested in what do you mean by "wrecking my compile performance".

It is your money and your job and you have all the right to purchase/return/change anything you want. I only wish you took the correct decision.

He did take the correct decision and that will not change. Personally, I would love me some threadripper but, cannot really afford to spend the money regardless. Good thing is, the person you quoted will move along and enjoy what he is using well you are secretly raging against his good choices.

 

[Zinn]

It is your money and your job and you have all the right to purchase/return/change anything you want. I only wish you took the correct decision.

Damn right it's my money and Intel can suck my dick. Between raising prices this gen to get full PCI-e lanes on HEDT, milking the same microarchitecture for a decade with minimal IPC gains, chips riddled with Management Engine and speculative execution backdoors, and collecting monopoly money the whole time, my tolerance for them fucking me as a consumer has reached its limit.

It was so funny to watch Intel panic and flailing around this gen when Threadripper caught them with their pants down. They were totally actually planning on releasing an 18-core desktop anyway, sure. Such a shithole company.

 

[juanrga]

you are secretly raging against his good choices.

Please don't make me laugh so hard.

 

[juanrga]

Damn right it's my money and Intel can suck my dick. Between raising prices this gen to get full PCI-e lanes on HEDT, milking the same microarchitecture for a decade with minimal IPC gains, chips riddled with Management Engine and speculative execution backdoors, and collecting monopoly money the whole time, my tolerance for them fucking me as a consumer has reached its limit.

It was so funny to watch Intel panic and flailing around this gen when Threadripper caught them with their pants down. They were totally actually planning on releasing an 18-core desktop anyway, sure. Such a shithole company.

As I already said I couldn't care less about what you do with your money. I am only interested in the technical stuff.

Intel is providing minimal IPC gains per gen, because Intel did hit a IPC wall (also named ILP wall) with x86. The existence of this wall has been known for a while and it is the reason why Intel (and HP) tried to abandon x86 by a scalable ISA.

Threadipper caught Intel with pants down, but Threaripper was AMD response to initial Skylake-X models, which caught AMD with pants down.

I see you didn't answer my question about the impact of Spectre patch on ThreadRipper performance.

 

[Zinn]

As I already said I couldn't care less about what you do with your money. I am only interested in the technical stuff.

Okay, technically Intel can suck my dick,

 

[juanrga]

I see you still didn't answer my question. So I conclude that you are changing your Intel setup by a ThreadRipper setup because the patches affected badly the performance of the Intel chip. But you are changing to ThreadRipper without knowing if the patches affects the performance of ThreadRipper and how much.

Guess what? I already saw some preliminary bench with ThreadRipper loosing 30% of performance on IO after Spectre patch.

 

[Zinn]

Cool story bro

 

[kirbyrj]

Guess what? I already saw some preliminary bench with ThreadRipper loosing 30% of performance on IO after Spectre patch.

After your claims in the other threads about 30% performance boosts or losses without any supporting info (and actually false info when you read into it), you'll forgive us for being skeptical of your claims. I'd have thought you'd have some foreign language, obscure graph by now to show us already. Claims of 30% "preliminary" benches without any supporting documentation, especially coming from a known AMD hater, is suspect.

 

[juanrga]

After your claims in the other threads about 30% performance boosts or losses without any supporting info (and actually false info when you read into it), you'll forgive us for being skeptical of your claims. I'd have thought you'd have some foreign language, obscure graph by now to show us already. Claims of 30% "preliminary" benches without any supporting documentation, especially coming from a known AMD hater, is suspect.

Don't attack the messenger only because you don't like the message. But forget for a minute those performance loses. Zinn is moving from Intel to ThreadRipper, but he has not seen/know any benchmark of ThreadRipper with the Spectre patch. I asked him three times and he refuted to answer. So he is moving from a system with a known performance impact to a system whose performance impact is unknown. And you and others are ignoring this fundamental fact.

 

[Zinn]

You are good at trolling or mental gymnastics or both. AMD chips are unaffected by Meltdown. Intel and AMD are affected by Spectre, but AMD to a lesser degree. Sure, there could conceivably be a catastrophic performance decrease that only affects AMD, but 30% seems in-line with a "preliminary" result that disables speculative execution across the board, which is of course a stupid result because in practice there will be context-sensitive flushing of kernel cache that apps will be able to hook into. It's more likely that the correct approach to Spectre mitigation will affect Intel and AMD similarly, but Intel will also have the gigantic wart of Meltdown also decreasing performance further.

So to answer your question, I don't care about the Ryzen Spectre fix because my understanding is that at worst Intel and AMD are equally fucked, and more likely AMD is less fucked (not even considering Meltdown).

 

[kirbyrj]

Don't attack the messenger only because you don't like the message. But forget for a minute those performance loses. Zinn is moving from Intel to ThreadRipper, but he has not seen/know any benchmark of ThreadRipper with the Spectre patch. I asked him three times and he refuted to answer. So he is moving from a system with a known performance impact to a system whose performance impact is unknown. And you and others are ignoring this fundamental fact.

The point is we haven't seen the message. We only know the messenger is biased .

 

[juanrga]

The point is we haven't seen the message. We only know the messenger is biased .

Keep ignoring reality, but it will not go away

 

[juanrga]

You are good at trolling or mental gymnastics or both. AMD chips are unaffected by Meltdown. Intel and AMD are affected by Spectre, but AMD to a lesser degree. Sure, there could conceivably be a catastrophic performance decrease that only affects AMD, but 30% seems in-line with a "preliminary" result that disables speculative execution across the board, which is of course a stupid result because in practice there will be context-sensitive flushing of kernel cache that apps will be able to hook into. It's more likely that the correct approach to Spectre mitigation will affect Intel and AMD similarly, but Intel will also have the gigantic wart of Meltdown also decreasing performance further.

So to answer your question, I don't care about the Ryzen Spectre fix because my understanding is that at worst Intel and AMD are equally fucked, and more likely AMD is less fucked (not even considering Meltdown).

I better don't say what you are good at.

There are some benches in #24 showing how some Zen systems (with Spectre patch only) take a higher performance hit than an Intel system (with Meltdown+Spectre). The patches don't affect in the same way to different CPUs. The patches even affect differently to CPUs from the same vendor. So the idea that Intel will always take a higher performance hit because it is patched for Spectre and Meltdown is not based in hard data.

The point is that you are moving from a system with a known performance hit to another system for which you don't know which is the performance hit. Then you are assuming/believing that the performance impact will be slower in your new system, but you don't have any data to support your belief.

 

[Charlie_D]

Keep ignoring reality, but it will not go away

I better don't say what you are good at.

There are some benches in #24 showing how some Zen systems (with Spectre patch only) take a higher performance hit than an Intel system (with Meltdown+Spectre). The patches don't affect in the same way to different CPUs. The patches even affect differently to CPUs from the same vendor. So the idea that Intel will always take a higher performance hit because it is patched for Spectre and Meltdown is not based in hard data.

The point is that you are moving from a system with a known performance hit to another system for which you don't know which is the performance hit. Then you are assuming/believing that the performance impact will be slower in your new system, but you don't have any data to support your belief.

This and just... everything. I love this whole thread.

 

[Zinn]

There are some benches in #24 showing how some Zen systems (with Spectre patch only) take a higher performance hit than an Intel system (with Meltdown+Spectre).

There are some benches in the same article showing really bad Intel-specific slowdowns too. Particularly write performance, which is catastrophically bad for Intel. Generally speaking everything I said is true - to the extent AMD is affected, it is affected less.

You seem fond of cherry-picking results to cast AMD in a bad light, but the reality is these fixes are stacking up in AMD's favor overall, and combined with the platform and threading advantages of X399, there's almost no reason to buy Intel on the high end at this point, unless they drop their prices significantly (lol).

 

[TurboGLH]

There are some benches in the same article showing really bad Intel-specific slowdowns too. Particularly write performance, which is catastrophically bad for Intel. Generally speaking everything I said is true - to the extent AMD is affected, it is affected less.

You seem fond of cherry-picking results to cast AMD in a bad light, but the reality is these fixes are stacking up in AMD's favor overall, and combined with the platform and threading advantages of X399, there's almost no reason to buy Intel on the high end at this point, unless they drop their prices significantly (lol).

Some of the AMD performance loss is mitigated with the AMD specific retpoline branch. It's worth a read, although I wonder why in some cases TR doesn't recovery while Epyc and Ryzen do.

https://www.phoronix.com/scan.php?page=news_item&px=AMD-Retpoline-Linux-4.15-FX-Zen