LinkBack URL
About LinkBacksIsn't that in danger of driving people to hold on to their old intel chips or replacing them with cheaper intel low end ones?Originally Posted by scaryjim
Isn't that in danger of driving people to hold on to their old intel chips or replacing them with cheaper intel low end ones?
Its very worrying unpatched Windows 7 is fine and Windows 10 isn't??
I do wonder if a fresh install of Windows 10 will do the same??
Dude,you need to consider some of us have better CPUs,and are still hitting CPU limitations. For me actually getting a Core i7 6700 now secondhand would actually improve performance in the games I play the most in - despite that I actually tempered my own expectations and the performance jump in what I will play for the foreseeable future,just as a hedge to support AMD and because games will tend to thread better(was looking at the R5 1600X and a mini-ITX motherboard).
The issue is that upgrading to Ryzen in its current state won't do that - so basically whats the point?? Yes games will thread better and yes AMD will get better games optimisations in new titles,but between then and now,I still want to run things on my computer.
If its not going to improve on what I run now,then whats the point of upgrading - I might as well wait another 12 to 24 months when games are more threaded,I got bored of the games on older engines I am playing now,and then look at whats available.
ATM,we are just waiting in hope that fixes to the windows and BIOSes will improve performance.
This would have all been not needed,if AMD actually did wait a month and launched Ryzen with these in place. I have said so many times AMD needed to get on top of its launches,and repeatedly it does not. Its always something is not quite right.
I linked to the JayzTwoCents for a reason - listen to what he has to say about it and places like TH touched on the same things.
Last edited by CAT-THE-FIFTH; 05-03-2017 at 01:32 PM.
taken from anandtechI did some 3D testing and eventhou there is not nearly enough data to confirm it, I'd say the SMT regression is infact a Windows 10 related issue.
In 3D testing I did recently on Windows 10, the title which illustrated the biggest SMT regression was Total War: Warhammer.
All of these were recorded at 3.5GHz, 2133MHz MEMCLK with R9 Nano:
Windows 10 - 1080 Ultra DX11:
8C/16T - 49.39fps (Min), 72.36fps (Avg)
8C/8T - 57.16fps (Min), 72.46fps (Avg)
Windows 7 - 1080 Ultra DX11:
8C/16T - 62.33fps (Min), 78.18fps (Avg)
8C/8T - 62.00fps (Min), 73.22fps (Avg)
Already posted here and in the review thread!!
thanks CAT
did have another thought though about the disabling of SMT.
with it disabled you get say 15% more performance.
that doesn't mean that when the patch is finished and rolled out that you'll only get 15% more with it turned on. you might get 30% for instance. it has been mentioned that the AMD chip is more efficient with memory and stuff . so the SMT bug may be holding it back quite a lot.
(% numbers just examples)
not for the people who don't want windows 10 it isn't
if what I said above comes true though, then they'll have reason to worry.
just need to wait and see..
as for the rushed launch and waiting another month. will the US tax deadline of April 15th have anything to do with it?
maybe AMD wanted to shift some physical stock to keep shareholders happy by attempting to post a gain for once. the chips were ready software/drivers can wait, especially if they use everyone who buys one as surrogate Q&A testers.
it may be an even worse situation if they did hold back another month or 2 for 'driver optimising' AMD then saying 'this is the best we can do, now launch' and them still having problems.
they just need to work on their PR communication a lot, as you say, then maybe these problems wont turn into even bigger problems where no-one even considers AMD anymore because of the sheer perceived incompentence in doing anything, even though quite a lot of it is then patched quickly after launch.
that my thoughts on this whole matter.
this is the review threadAlready posted here and in the review thread!!
If someone made the new AMD chitchat thread this would all be so much simpler...Already posted here and in the review thread!!
Ok,the other one!!this is the review thread
Hint taken!!If someone made the new AMD chitchat thread this would all be so much simpler...
Hmm, there was rumour of 1M chips made for the launch. If true, at an average of $400 to make the maths easy, that would be $0.4B in stock that could either be shifting or sitting there attracting interest payments. I can see an attraction in getting them sold, end of year or not!
(WRT results mentioned by HalloweenJack): Those sort of results obviously imply this should be resolvable with a Windows scheduler patch, going back to what I said in the first place - the suggestion that this would require individual game patches never made any sense, and as far as game optimisations go, this is no different to any other CPU release where developers can deploy architecture-specific optimisations e.g. use a CPU dispatcher to select a binary depending on what CPU family the code is running on.
It's perplexing that AMD never got this sorted before day-1 reviews, which as CAT says is exactly the sort of thing that they stumble on time and time again, and it's something you can be very sure the engineers were well aware of.
Just a thought, maybe it's one reason behind the decision to launch the lower core-count parts later in the year, which at least on some sites will encourage a re-visit of the 8 cores. Having a product out there puts it in the hands of developers without them having to worry about NDA's, and it allows time for things to settle before launching other products, including the APUs.
WRT the inter-CCX communication - this is yet another scheduler responsibility, and in some ways similar to how NUMA is treated - the scheduler has to be aware of latency between cores (in the case of NUMA, because of different sockets with their own memory controllers) in order to make the best decision about where to place threads. Just because the bandwidth is lower or latency is higher for what will be very occasional cross-CCX cache accesses (especially given the L3 is a victim cache), doesn't imply a problem - that's what prefetchers are for. The L3 is not used for prefetching on Zen, the L2 is, and remember the L2 is very large - twice the size of Intel's. Swings and roundabouts. What won't help is if Win10 is not aware of this and scatterguns threads around the CPU completely at random, but this is true to some extent with most CPUs - local access will nearly always be considerably faster than accessing cache from neighbouring cores (like Intel's L3 cache slices, connected with the ring bus).
I do wonder if the Win7 AMD scheduler patch actually helps zen , as results show win7 is a better OS at this time
Time to buy SJ,a big virtual beer??
Tests with 4C on two CCX or one CCX indicate the single CCX has generally better performance:
http://www.pcgameshardware.de/Ryzen-...eview-1222033/
Yep,and it seems the 4C versions which appear to use only one CCX might be fine.
Phew!!
Doesn't DX12 allow developers to do their own optimisation and stuff though, I've not bothered looking into what each part of that involves but could that explain AMD's comment about game developers needing to patch their games, so DX12 games need patching while DX11 need an OS patch.
Thats the thing - if they actually made a distinction it would have helped,but somehow it is not surprising.Doesn't DX12 allow developers to do their own optimisation and stuff though, I've not bothered looking into what each part of that involves but could that explain AMD's comment about game developers needing to patch their games, so DX12 games need patching while DX11 need an OS patch.
Last edited by CAT-THE-FIFTH; 05-03-2017 at 05:34 PM.
How to install Ryzen under Windows 7:
https://forums.anandtech.com/threads...500572/page-11
It depends on what sort of optimisation you're referring to. Things like task scheduling are down to the OS kernel, along with e.g. time slicing, prioritisation and interrupts. Even with DX12, games don't get bare-metal access to the computer hardware with no OS involvement, and that can't happen simply because many processes besides the game need to run alongside it. That's the sort of thing that used to happen on consoles like the 360, where IIRC the dashboard was more akin to a BIOS, and every game included and booted its own kernel, and hardware management was, in that case, the responsibility of the game developers.Doesn't DX12 allow developers to do their own optimisation and stuff though, I've not bothered looking into what each part of that involves but could that explain AMD's comment about game developers needing to patch their games, so DX12 games need patching while DX11 need an OS patch.
I suspect that AMD comment was a bit of a misunderstanding on one side or another. Game developers, regardless of DX11 or 12, are still the ones responsible for compiling their code and, just for example, including vendor-specific codepaths to extract better performance*, choosing compiler flags, what instruction sets they allow the compiler to use, etc.
*A couple of random examples - conventional x86 vs AES-NI for cryptographic processing, different instruction latency for divides vs multiplies where in some cases they can be used interchangeably. And just because an instruction is available, doesn't mean it's the best option e.g. the y-cruncher developer found that using AVX on Bulldozer led to a performance regression for that code vs the SSE3 codepath. That's the sort of thing I think of when code optimisation is brought up.
On the subject of this discussion, the SMT issues are something that seems more to do with the OS scheduler than the games themselves. But given Ryzen is a new uArch, it's to be expected that developers will have the opportunity to learn the idiosyncrasies of the microarchitecture and get more out of it in future games (or patches if it's worth it).
Last edited by watercooled; 05-03-2017 at 06:57 PM.
Corky34 (06-03-2017)
There are currently 1 users browsing this thread. (0 members and 1 guests)
Posting Permissions
Reply With Quote
