With all the ssecurity issues with Intel having an impact on performance is now the right time to move to AMD?

[Dibble93]

Hi,

 

I am currently considering an upgrade from Intel 5820K and am wondering if with all the flaws such as Zombieland and the impact the possible fixes will have on performance is it now time to join team AMD. I am doing this as I have been for a while doing virtualisation and want to give myself a bit more breathing room when it comes to the number of VMs that I can run at once, I also game and some video editing.

I have been considering the i9 9920X/9960X and threadripper 2950X (I do realise there is a difference in cores with the 9920X) but my biggest concern is what kind of perfomance impact the fixes will have, I know Intel tends to be better for gaming but with the hit to performance will it be?

Any advice or information would be grearly appreciated.

Thanks

[porina]

AMD have a Keynote at Computex next Monday, where they are expected to announce Zen 2 (Ryzen 3000) series desktop CPUs. Definitely worth waiting at least until then to see what's available. 

[Morgan MLGman]

I don't think the fixes are fully out yet. Intel's current recommendation to protect against MDS attacks is to disable HyperThreading, meaning that you lose up to 40% of multitheaded performance if you do that. The fixes will probably allow you to keep HT on, but at some (unknown) performance impact which most likely won't be a negligible 2-3%.

About the move to AMD - normally I'd recommend looking at Threadripper CPUs but the new Zen 2-based lineup that's about to be unvelied soon is confirmed to feature up to 16-cores, meaning that the best value Threadripper CPUs may be pointless when it does launch. I'd recommend waiting and maybe overclocking the 5820K for the time being? It has quite a bit of overclocking headroom provided you cool it well-enough.

[porina]

5 minutes ago, Morgan MLGman said:

Intel's current recommendation to protect against MDS attacks is to disable HyperThreading,

 

Quote

Is Intel recommending that I disable HT?

No. Intel is not recommending that users disable Intel® Hyper threading. It’s important to understand that doing so does not alone provide protection against MDS, and may impact workload performance or resource utilization that can vary depending on the workload.

https://www.intel.com/content/www/us/en/architecture-and-technology/mds.html

 

5 minutes ago, Morgan MLGman said:

meaning that you lose up to 40% of multitheaded performance if you do that.

Outside possibly of the hardly known benchmark 3DPM, I'm not aware of anything that could be impacted remotely near that much. Something like Cinebench R15 or R20, which sees about 30% gain from HT/SMT compared to not, would only suffer a 23% drop if disabled. Most software gains far less benefit (and thus less loss) than that.

[Morgan MLGman]

19 minutes ago, porina said:

https://www.intel.com/content/www/us/en/architecture-and-technology/mds.html

I stand corrected, it was Apple, not Intel that recommended disabling HT on Intel CPUs for maximum protection (along with updates obviously).

19 minutes ago, porina said:

Outside possibly of the hardly known benchmark 3DPM, I'm not aware of anything that could be impacted remotely near that much. Something like Cinebench R15 or R20, which sees about 30% gain from HT/SMT compared to not, would only suffer a 23% drop if disabled. Most software gains far less benefit (and thus less loss) than that.

I was talking about it with Cinebench in mind, I think it's between 35-40% which is why I said "up to 40%" I phrased it that way deliberately, although it might not be that extreme (as you mentioned) and is certainly not nearly as extreme in games. Although this one depends on how many cores does your CPU have in the first place. If it's a dual core then HT will help greatly in more intensive games, but when you have a high-end 8-core the performance uplift is usually negligible

I don't understand the part where you mention that CB sees about 30% gain from HT/SMT compared to when it's turned off, and then you say that it would only suffer a 23% drop if disabled? Can you explain what you meant?

Also - I think it's funny that AMD's SMT implementation in Ryzen chips is not only more effective in terms of performance gains but is also more secure

[Stefan Payne]

26 minutes ago, Dibble93 said:

with all the flaws such as Zombieland and the impact the possible fixes will have on performance is it now time to join team AMD.

No, because AMD is reading their 7nm Lineup and will talk about that in some way next monday or so.

Its so close and the step that big that it makes little sense to buy AMD now.

 

The only reason to do it is if your system blows up and you need something now.

But even then, I'd recommend a lower cost CPU such as the 1600 or 1600X, maybe 2600 for 150€ or less...

 

Well, we know more next week at this time.

 

As for the Spectre/Meltdown inspired security attacks: 
They pop up every other month and its far from over. We will see that for a couple of years. And it will kost 2-5% Performance every time, sometimes more. And Intel is more suceptable for that...

[porina]

1 minute ago, Morgan MLGman said:

I don't understand the part where you mention that CB sees about 30% gain from HT/SMT compared to when it's turned off, and then you sai that it would only suffer a 23% drop if disabled? Can you explain what you meant?

Say we start with 100 units of something. Increase it by 30%, you get 130. Let's decrease that by 30%. 30% of 130 is 39, so we're down to 91, not the 100 we started with. To get back to 100, we would reduce by 23% of 130.

 

1 minute ago, Morgan MLGman said:

Also - I think it's funny that AMD's SMT implementation in Ryzen chips is not only more effective in terms of performance gains but is also more secure

AMD's core is generally stated to have more throughput potential than Intel's cores, but it is unable to deliver as much of it running a single thread through it. So it is a bit below for 1 thread per core, a bit above for 2 threads per core. So the gain is maybe overstated a little from starting a little lower in the first place. That's a thought, AMD are improving the front end of Zen 2, so that could in particular increase the single thread performance. I'm looking forward to testing it as soon as I can get one myself.

 

On testing, below is a comparison I attempted focusing more on the CPU architecture than a specific model. Interesting to see areas of relative strengths and weaknesses between them. Again I look forward to adding Zen 2 to the mix, but given all the security updates I should re-run the Intel ones fresh, but each CPU is about half a day of testing...

 

 

[Morgan MLGman]

22 minutes ago, porina said:

Say we start with 100 units of something. Increase it by 30%, you get 130. Let's decrease that by 30%. 30% of 130 is 39, so we're down to 91, not the 100 we started with. To get back to 100, we would reduce by 23% of 130.

Oh okay, I see. I think it's a bit overly complicated way of thinking about it but I suppose you're right

23 minutes ago, porina said:

AMD's core is generally stated to have more throughput potential than Intel's cores, but it is unable to deliver as much of it running a single thread through it. So it is a bit below for 1 thread per core, a bit above for 2 threads per core. So the gain is maybe overstated a little from starting a little lower in the first place. That's a thought, AMD are improving the front end of Zen 2, so that could in particular increase the single thread performance. I'm looking forward to testing it as soon as I can get one myself.

That's quite interesting, I always thought the main focus of improving Ryzen's singlethreaded performance was reducing cache latencies and improving the IMC inside, on top of improving Infinity Fabric in terms of both throughput and latency

That said, I think that AMD has a very good base architecture to improve upon in regards of singlethreaded performance (unlike Skylake LOL, which ++++ are we on nowadays?)

27 minutes ago, porina said:

On testing, below is a comparison I attempted focusing more on the CPU architecture than a specific model. Interesting to see areas of relative strengths and weaknesses between them. Again I look forward to adding Zen 2 to the mix, but given all the security updates I should re-run the Intel ones fresh, but each CPU is about half a day of testing...

Nice work man, interesting results! Yeah, I think you should update your Intel-mobo BIOSes as well before conducting testing as there may be a performance difference when full MDS mitigations are applied, same goes for AMD BIOSes although I don't expect a huge differences there

[porina]

16 minutes ago, Morgan MLGman said:

That's quite interesting, I always thought the main focus of improving Ryzen's singlethreaded performance was reducing cache latencies and improving the IMC inside, on top of improving Infinity Fabric in terms of both throughput and latency

They may well be factors also, but they're not things I can easily test. Part of my aim in that testing was to reduce factors external to the core as far as practical, and I was using 3000 ram so IF should be an ok speed.

 

16 minutes ago, Morgan MLGman said:

(unlike Skylake LOL, which ++++ are we on nowadays?)

I forget the name right now, but the 10nm mobile chips Intel are putting into volume production this year will be the first substantial architecture update since Skylake. I'm debating if I will try to get one to test, but I don't have need for a new laptop which I doubt will be cheap. It will be much more interesting if they did desktop, but that's not on the short term radar it seems. Maybe if they put it in a NUC?... Actually I don't need Zen 2 either, but as a desktop platform I can at least use that in more scenarios.

 

16 minutes ago, Morgan MLGman said:

Nice work man, interesting results! Yeah, I think you should update your Intel-mobo BIOSes as well before conducting testing as there may be a performance difference when full MDS mitigations are applied, same goes for AMD BIOSes although I don't expect a huge differences there

Yup, it will have to be up to date at the time. I don't think I did it last time as I used my competitive benching Win7 installs which may not be up to date.

[Jay Deah]

This zombieload thing is getting blown out of all proportions. Honestly it’s not worth factoring that into any buying decision it’s just background noise.

 

When choosing  your next processor definitely consider AMD but choose based on real reasons of what best meets your requirements and budget. 

[Dibble93]

Thank you all for replying, for those who suggested that I wait until the AMD announcement that is happening soon I will most certainly do so I didn't realise how soon that was happening.

 

Jay, I would respectfully disagree with you, I believe that these flaws must be considered as when they fix them using microcode updates there will most likely be some sort of performance impact on the CPUs which must be considered, however I do agree that it shouldn't be the only consideration but due to its nature I believe it has to be thrown in with the rest.

[Oalei]

Not really it only affects some group or people, well for Linux user it seems they have 15% performance degradation when you compare it to windows.

[Morgan MLGman]

On 5/22/2019 at 5:06 PM, porina said:

They may well be factors also, but they're not things I can easily test. Part of my aim in that testing was to reduce factors external to the core as far as practical, and I was using 3000 ram so IF should be an ok speed.

 

I forget the name right now, but the 10nm mobile chips Intel are putting into volume production this year will be the first substantial architecture update since Skylake. I'm debating if I will try to get one to test, but I don't have need for a new laptop which I doubt will be cheap. It will be much more interesting if they did desktop, but that's not on the short term radar it seems. Maybe if they put it in a NUC?... Actually I don't need Zen 2 either, but as a desktop platform I can at least use that in more scenarios.

 

Yup, it will have to be up to date at the time. I don't think I did it last time as I used my competitive benching Win7 installs which may not be up to date.

Hey, this video just came out - might be interesting to you:

[porina]

25 minutes ago, Morgan MLGman said:

Hey, this video just came out - might be interesting to you:

Stopped watching them ages ago. Did skim that video regardless, nothing unexpected looking at result charts, didn't listen to any FUD.