I dunno TBH,as I am uncertain whether the issue is more down to the uarch and less about single core performance as I assume there is zero optimisation for Ryzen in those sorts of games. It really would be interesting to see tests with a single CCX and with SMT off,to see if this is the issue.Originally Posted by Sumanji
On the face of it I would be surprised if it is more than 10% to 15% extra performance overall.
Keep in mind, these Ryzen performance issues practically vanish if you bump up to 1440p. I know it's been a bad year for enthusiasts, but keep the savings from not going Intel and putting it towards a 1440p capable GPU upgrade when Volta/Navi comes along is not a bad idea.
Not in the one or two games I am talking about sadly - even at qHD there are performance drops under 30fps.
Just a note that the best technical write up of how Meltdown works that I have seen so far is in the copy of MagPi, the Raspberry Pi magazine, that dropped through the letterbox this week.
When you consider that the Pi is immune to these attacks (no speculative execution) that seems rather odd.
mikerr (31-01-2018)
Thankfully those are super edge cases. Nearly across the board the performance difference virtually vanishes with the resolution bump shifting the bottleneck from the CPU to the GPU. Sadly Kerbal Space Program is one of those edge cases because of the woefully unoptimised engine. Guess what my favourite game is.
Well in my case,the games I keep going back to seem to be FO4,Planetside 2,Overwatch and Diablo 3. The last two run perfectly fine for me,but the first two performance can really tank. Fallout 4 is very dire,and most review sites don't test the game properly. I can tell which review sites know the game,since its settlements where performance is not great and only HardOCP seemed to test it that way. Planetside 2 performance in large battles also drops minimums but its hard to say how much of that is on the server side though.Thankfully those are super edge cases. Nearly across the board the performance difference virtually vanishes with the resolution bump shifting the bottleneck from the CPU to the GPU. Sadly Kerbal Space Program is one of those edge cases because of the woefully unoptimised engine. Guess what my favourite game is.
Someone on one of the slack groups I'm subbed to (can't remember if it was the work one or the community one) linked to a post on the Pi blog which gave a beautifully simplified explanation of the issues as part of their explanation of why the processors used in Pis weren't vulnerable. I'd be surprised if the article in MagPi wasn't based on that blog post, and it makes sense that you need to provide an understandable explanation of how the exploit works if you're going to explain why you're not vulnerable to it.
Some quick searching and attempt to understand it.
CPU attempts to predict in advance the outcome of a conditional statement before it is calculated, and will then in advance perform the next step saving CPU cycles should that prediction be correct.
The exploit is that mispredictions have side effects which can be observed, and through the use of Timing Attacks based on the duration of the calculation, the original input can be determined.
Which in short means that data from another application can be exposed.
My thought as to why the apparent performance hit is that Timing Attacks can be mitigated by returning the result in a fixed interval even when it is finished quicker, so there is nothing given away about the duration of the calculation.
That is a software fix though, and instead protecting the mispredicted execuations from that analysis is what I think the hardware fix will be.
As I mentioned above, the best explanation I've seen is on a raspberry pi blog article that explains why PIs aren't affected (basically they don't do the speculative execution that's key to the attacks): https://www.raspberrypi.org/blog/why...e-or-meltdown/
I'm not sure exactly which bit of the exploit the software fixes target, but since the attacks target a hardware feature specifically designed to improve single-threaded performance, I honestly don't see how any fix can avoid a single-threaded performance penalty. Fixing it in hardware will presumably reduce that penalty, but I can't help thinking that Intel's "fixed" silicon is going to have slightly lower IPC than the "unfixed" silicon. It'll be interesting to see how it all shakes down.
As to johnroe's point about offering discounts; the processors are functioning as designed. Neither Intel and AMD guarantee any particular level of performance for any of their CPUs; if you look carefully you'll see their slide decks are riddled with cya footnotes that detail exactly what the test setups were like and explicitly disclaim that performance is indicative only.
I know it's sucky that software updates can kill your PCs performance, and that this one is down to a security issue that's baked into the silicon, but it's not like Intel and AMD have been designing insecure chips since they found out they were exploitable - every piece of silicon that's currently on a shelf will have been designed a couple of years ago. If Intel can actually get new silicon out, which addresses these vulnerabilities, within 18 months of notification, that would represent a pretty impressive turn around on silicon changes, remasking, sampling, testing and shipping. I suspect it'll take AMD longer to be able to make the relevant changes (although they are affected by fewer of the issues, so perhaps they've got a smaller job to do).
The Linux meltdown patch works by re-mapping the user process so that only a tiny stub of the kernel appears in it. That way, circumventing the kernel/user permissions doesn't expose anything because nothing is there to expose.
Edit: Note that AMD64 splits the address space so that there is a kernel half at the top and a user half at the bottom. That gives a big hint to tread careful when the top address bit changes from a 0 to a 1 which hardware can use.
Last edited by DanceswithUnix; 30-01-2018 at 02:30 PM.
The cynic in me says that:
“We’re working to incorporate silicon-based changes to future products that will directly address the Spectre and meltdown threats in hardware, and those products will begin appearing later this year.”
Really means:
"We're die shrinking those old Atoms and rebadging them as Pentiums while we come up with a better plan.
I would like to know exactly what intel mean by 'next generation' in their 'hardware solutions to the Meltdown and Spectre vulnerabilities, which will be delivered in the next generation of Intel chips' seeing as everyone has been saying for weeks that they wont have time to change the designs for the desktop Cannon lake chips due this year. Only another year or so until they release something that plugs the security holes, or have they known about this long enough to design something in to cannon lake? My bet is on the former
2700X,X470 Taichi,Silverstone Fortress 2,16GB RAM, SSDx3, HDDx4,GTX970 G1 Gaming,24"x2(1xIPS,1xTFT),W10x64Pro
HTPC: AthlonX2 5050e,M4A78-EM,AntecFusion,8GB RAM,ATi3200,32"Sony TV,W7x64Pre
no one here is aware ALL CPUS's including ALL ARM architecture is affected? Why do you think APPLE push out patches for their phones already... you guys are lost
Life is seldom that black and white, and this is one of those times.
I think most people would like to know that also, unfortunately no body knows as they've given out very little detail on how they'll achieve that or even if the hardware solutions to Meltdown and Spectre will be widely available.
It seems unlikely, but not impossible, that they'll be able to include these hardware changes into Cannon Lake as (iirc) that's already been taped-out however as they've also stated they'll have hardware solutions in the same year, this year, that Cannon Lake is due to be released so one would assume they're either doing a hastily put together second revision or that processors with the hardware solutions will be limited in numbers or exclusive to a particular market segment.
There are currently 1 users browsing this thread. (0 members and 1 guests)
Posting Permissions
LinkBack URL
About LinkBacks
Reply With Quote