Intel bug incoming? Meltdown and Spectre exploits

[AndreiD]

@Bonjour Still pretty far behind, the single core difference between Bulldozer & derivatives and 2nd to 6th gen Cores was pretty huge. With a 40% jump in IPC AMD still hasn't caught up yet, albeit it's close enough now.

@humbug There are far too many outlets without a working relationship with Intel which will jump at the opportunity to publish performance numbers if there's anything shady happening. My money is on EULA mix up because every way you look at it, it's just dumb of them to do if it's serious.

They didn't enforce no benchmarks in EULAs for L1TF and Meltdown, which were big performance drops.

 

[CAT-THE-FIFTH]

They can't, you're right, i said that earlier, however for Intel that's not the point, all they need is the threat and people will fall into line, its a message, we are serious about wanting you to shut the _____ up!

Humbug its basically what I posted here:

https://forums.overclockers.co.uk/posts/32059458/

 

[NinjaCool]

Wait,isn't that what I posted in the last few pages - TPU picked up on it??

Edit!!
It was in this thread:
https://forums.overclockers.co.uk/threads/foreshadow.18828376/

I completely missed that post but it sure seems like you are right

 

[humbug]

Humbug its basically what I posted here:

https://forums.overclockers.co.uk/posts/32059458/

Yup, i responded to that.

Irregardless if its a EULA mix up or not the fact remains they are attempting to stop people publishing performance impacts, what document its written in is completely irrelevant, its the fact that its written at all.

 

[Aretak]

@Bonjour Still pretty far behind, the single core difference between Bulldozer & derivatives and 2nd to 6th gen Cores was pretty huge. With a 40% jump in IPC AMD still hasn't caught up yet, albeit it's close enough now.

It was more than 40% from Piledriver to Ryzen - more like 70%+ in pure IPC terms. The FX chips were a joke.

It's essentially an even playing field at this point, though Intel still have their clockspeed advantage for now. Of course, there are other aspects that affect real-world workloads than purely the strength of the cores. Skylake-X was the ultimate example of that, given that it uses essentially the exact same cores as Intel's mainstream chips, yet was hobbled by its changes elsewhere for gaming especially compared to them.

 

[Panos]

Intel Gags Customers from Publishing Performance Impact of Microcode Updates!
https://www.techpowerup.com/247028/...shing-performance-impact-of-microcode-updates

Intel has updated their license terms regarding their microcode update distribution to explicitly forbid its users from publishing comparative "before/after" performance numbers of patched processors
Edit: It looks like Intel will be rolling back the offending clauses in the license!

@CAT-THE-FIFTH @humbug

No wonder they do not want the benchmarks

https://www.phoronix.com/scan.php?page=article&item=l1tf-early-look&num=2

And for FULL Protection you must disable Hyperthreading on the Intel CPUs also

Now that i7 8c/8t makes sense

 

[EsaT]

No wonder they do not want the benchmarks

https://www.phoronix.com/scan.php?page=article&item=l1tf-early-look&num=2

And for FULL Protection you must disable Hyperthreading on the Intel CPUs also

Now that i7 8c/8t makes sense

I wonder how big changes to Ice Lake Intel has had to do...

 

[humbug]

@CAT-THE-FIFTH @humbug

No wonder they do not want the benchmarks

https://www.phoronix.com/scan.php?page=article&item=l1tf-early-look&num=2

And for FULL Protection you must disable Hyperthreading on the Intel CPUs also

Now that i7 8c/8t makes sense

Christ, yeah that's not good.

 

[AndreiD]

@Aretak You're using 1 workload to compare IPC between different CPU architectures, that's not how it's done and guru3D should know better, you need many different workloads to actually get a clear picture:
https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/

excl 256b is what's most relevant to consumer workloads.

 

[humbug]

@Aretak You're using 1 workload to compare IPC between different CPU architectures, that's not how it's done and guru3D should know better, you need many different workloads to actually get a clear picture:
https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/

excl 256b is what's most relevant to consumer workloads.

Why?

Spreadsheets are not more important than rendering, not when compiling spreadsheets takes seconds or minutes while rendering can take hours or even days, Excel is much of a muchness irrelevant saving a minute here or there when you're potentially saving hours with 10% better rendering performance.

Even then those results are symptomatic of Excel not spreadsheet compiling, its typical of Microsoft using very old unoptimised compilers because they are cheap and lazy and given to them by Intel who designed them to use slower code paths whenever an AMD CPU is detected, don't deny it there is an existing lawsuit about it.
The comparison results will be very different using better up-to date software like Libra Office which don't use Intel compilers.

 

[Panos]

@Aretak You're using 1 workload to compare IPC between different CPU architectures, that's not how it's done and guru3D should know better, you need many different workloads to actually get a clear picture:
https://forums.anandtech.com/threads/ryzen-strictly-technical.2500572/

excl 256b is what's most relevant to consumer workloads.

Interesting you picked that benchmark only from the link you posted above. Not all the other ones with the CPUs locked at 3.5Ghz comparing them, and are dozens....

Like the relative one




Or the ones showing that the Zen IPC is pretty good on number crunching single thread applications and rendering.

Yes all these benchmarks are on your link you posted. Pretty interesting read tbh.

 

[AndreiD]

@humbug because that's how you do IPC differences between architectures or should we use 256bit wide instructions to make Intel seem light years ahead or should we use AES ones to make AMD seem light years ahead? You do a little bit of everything to see proper differences in IPC, not just 1 workload.
If you took that Cinebench """IPC""" slide for granted you'd think the 2700X would be faster than an i5 8400 in gaming, given the Pinnacle Ridge to Coffee Lake """IPC""" difference, according to that slide, is only 2.7% but the R7 2700X turbos 300Mhz or 7.5% higher than the i5 8400, but it's not: https://www.techpowerup.com/reviews/AMD/Ryzen_7_2700X/13.html

@Panos
I posted the summary with the CPUs at 3.5Ghz, and I'm not sure how that relative one shows anything different?

 

[Panos]

@humbug because that's how you do IPC differences between architectures or should we use 256bit wide instructions to make Intel seem light years ahead or should we use AES ones to make AMD seem light years ahead? You do a little bit of everything to see proper differences in IPC, not just 1 workload.
If you took that Cinebench """IPC""" slide for granted you'd think the 2700X would be faster than an i5 8400 in gaming, given the Pinnacle Ridge to Coffee Lake """IPC""" difference, according to that slide, is only 2.7% but the R7 2700X turbos 300Mhz or 7.5% higher than the i5 8400, but it's not: https://www.techpowerup.com/reviews/AMD/Ryzen_7_2700X/13.html

Could we get these gaming benchmarks with Vega 64 please also.
Because we know that Nvidia still hasn't resolved it's driver issues with the AMD CPUs even as we speak.

And that showed again few weeks ago on the Threadripper benchmarks.

 

[Panos]

@AndreiD and before you try to argue, have a look here.

So either the Vega 64 is that miles faster than the GTX1080 or there are issues with the Nvidia drivers, don't you think?

 

[AndreiD]

@Panos Phoronix has some benches under Linux: https://www.phoronix.com/scan.php?page=article&item=ryzen-2700-gaming&num=2 with Vega64 and 1080 Ti

And Nvidia's driver prefers lower latency non-NUMA CPUs, which a lot of software also does, not sure how that's a new thing.

 

[AndreiD]

Also, if you're referring to this golem.de article, I do want to point out that the 2990WX is getting 38 to 76fps in those games at 720p, the performance is still really bad even with the Vega 64, just less bad than with the 1080 Ti.

 

[Panos]

Also, if you're referring to this golem.de article, I do want to point out that the 2990WX is getting 38 to 76fps in those games at 720p, the performance is still really bad even with the Vega 64, just less bad than with the 1080 Ti.

I pointed that we need a Vega benchmark because there are issues with the Nvidia drivers & Ryzen CPUs.

8700K has the edge bellow because of the 10%+ higher clocks. (2700X works are around 4Ghz when gaming at stock), yet the performance doesn't traslate to more than 2-3fps

So counting "absolute IPC" numbers is completely irrelevant, don't you agree?

Also compared to the 2700X + GTX1080 video above, you will see that the 2700X performs way worse with GTX1080 than with the Vega 64.
Surely is not the CPU given that has no issue with Vega 64 to keep up on the much faster 8700K.

Also in relation to 8400..... Here is an interesting benchmark.
True the 2600 is clocked to 4.2Ghz, which is easy to do, and the 8400 has overclocked GTX1060.
Lets see the results, shall we?

So I believe the whole IPC discussion can go out of window as completely irrelevant these days.

 

[AndreiD]

I gave you benchmarks from a high quality outlet with the R7 2700, R7 2700X, i7 8700K with Vega64 and 1080 Ti.
And the difference in your first video is because the R7 2700X isn't as good for gaming as an i7 8700K because of the inherent Ryzen issues like memory & infinity fabric latency for one.

Also your other two videos are just GPU bottlenecked, using a 1060 to see differences between 2 CPUs in gaming is silly.

 

[humbug]

@humbug because that's how you do IPC differences between architectures or should we use 256bit wide instructions to make Intel seem light years ahead or should we use AES ones to make AMD seem light years ahead? You do a little bit of everything to see proper differences in IPC, not just 1 workload.
If you took that Cinebench """IPC""" slide for granted you'd think the 2700X would be faster than an i5 8400 in gaming, given the Pinnacle Ridge to Coffee Lake """IPC""" difference, according to that slide, is only 2.7% but the R7 2700X turbos 300Mhz or 7.5% higher than the i5 8400, but it's not: https://www.techpowerup.com/reviews/AMD/Ryzen_7_2700X/13.html

@Panos
I posted the summary with the CPUs at 3.5Ghz, and I'm not sure how that relative one shows anything different?

Agreed and that's not what you did, you quoted one slide and then said.

excl 256b is what's most relevant to consumer workloads.

No it isn't, and its just one workload, across most productivity workloads the 2700X beats the 8700K, in games at 1080P with a 1080TI the 8700K is about 10 to 15% better.

On the 8400, TPU's results are all strangely clustered within 10% of each-other at the top, clearly the way they do reviews it more a GPU bottleneck than CPU, the fact that the 4 core CPU's result the same as the 6 core ones i think speaks volumes about the poor quality of the review, they also only used about 10 games. which again is strange as before Ryzen they used to do 25 games, is the selection of games Intel win in running very low?

When done right actually the 2600 is better than the 8400 for gaming.



I would never recommend the 2700X as a pure gaming CPU, the 2600 for £160 results in exactly the same performance, that's 15% slower than the 8700K for £200 less money.
The 2700X is a good all rounder, a better rounded CPU than the 8700K.
Oh and far more secure.

Anyway, why did you bring AMD vs Intel back into this thread?
You're off topic can we get back on topic please?
It was going so well before you started the same tired old arguments again.

 

[AndreiD]

The IPC slide I linked is the summary of 26 tests, it's a lot more relevant than just 1. And again, this is IPC, not multithreaded performance where I agree the 2700X is better than the 8700K.

And I only brought up the fact that some people in this thread don't understand how IPC testing should be done and why it's important to use more than 1 workloads, and you @humbug and Panos got very defensive while disproving nothing I said.

But yeah, this should go back on topic because this isn't the another Ryzen fanclub thread.

And actually related to the thread: https://www.tomshardware.com/news/intel-cpu-microcode-benchmark-mitigation,37684.html
The EULA with the clause prohibiting benchmarking the microcode updates has been changed to remove that line.