Is there a way to turn off SMT on a per application basis?
Is there a way to turn off SMT on a per application basis?
The only way that I could see was to switch it in the BIOS.
Even then, the BIOS had to be cleared before the switch took effect.
Well reviewers switched SMT off in some reviews - its hilarious if AMD missed that. I wonder if some of the worst reviews for gaming are done to SMT causing thread stalling.
I mean doesn't AMD write reviewers guides - surely they should have said something about this??
This reminds me of the R9 290X again - a potentially great product which was rushed out half-finished,and yet in the end look how well it aged?? The R9 285 was launched half finished,the Fury X was launched with bugs,the RX480 had issues.
I mean why can't they ever launch anything without some problem??
Intel and Nvidia tend to launch products which on average are more polished.
AMD doesn't and it masks all the very hard work they have done.
Yes,Ryzen should bump AMD CPU sales up quite nicely even for gaming rigs,but many average people will look at these initial reviews and go Intel. AMD is loosing potential sales there.
They really need to get on top of this major issue,otherwise its going to affect the 4C and 6C models which most people are waiting for.
At some point I do need to upgrade this old IB based rig I have,but if AMD really does not get on top of this,I will just get another Core i7 and not bother upgrading for a few years.
I am quite happy to wait a while,but in the end Coffee Lake is will be out at the end of the year or early 2017 so its a limited time-frame indeed.
The problem with all these accurate leaks before hand is that the reviews don't actually have anything new to tell us! it's a huge improvement over Piledriver for single thread but still trails Intel by a bit, and it shines when you load up all the cores.
On the plus side AMD are actually playing in the same division as Intel on single-threaded performance now, even if they're at the other end of it. OTOH that still leaves them trading on "better at heavily threaded workloads" and "cheaper than the alternative" as their two main selling points - much as they have been for the last 6 years....
Is this a potential issue with the task scheduler again like with CMT?
AMD really suffered under the old chips because windows didn't know which cores shared resources so instead of putting light tasks on the second core that shared resources with the first it was simply allocating them out in core number order so the first 4 cores in the system were actually 2 pairs that shared resources causing performance issues whilst the other 2 pairs were mostly idle. Microsoft was going to release a patch for Windows 8 to make this better but it never made it to mainstream.
The above might not be an issue in an SMT architecture though.
Tarinder, did you notice the CPU core utilisation under the Warhammer test? Were some of the cores idle? With SMT off were the cores able to boost to a higher frequency on their own?
Thanks Tarinder nice review. Very decent CPU considering the price and its multi-core performance. Its very nice to see some real competition for Intel. Bummer i should have waited for Zen instead of going for another intel.
People could take exception to being called average you know ;-)
Surely it's not just following the herd to buy Intel it's the logical thing to do.
- Established product
- Less driver issues
- Established hardware platforms
- Know what you are going to get
- And for the overwhelming majority of users (inc. gamers) they are faster.
Where's the logic on taking a punt on a brand new product that's probably going to have many teething problems unless it's much faster or massively cheaper ?
Look at the Hardware.fr tests - upto a 13% performance regression in games and as usual AMD PR was asleep about all of it.
What did AMD think would happen when reviews happened??
I am honestly getting fed-up that at every AMD CPU and GPU launch there is some bloody problem,and "give it a few months we can fix it" and the competition will use that to get more sales at their expense.
What type of internal testing are they doing?? If review sites in less one bloody week can expose such a big problem,then WTF was AMD doing all these months then??
Average Joe or Jane is not considered an insult,its people who are not tech enthusiasts - but the main issue is AMD should have found those SMT issues first. Even in the Hexus and Hardware.fr reviews some of the performance difference due to the SMT issues is making it much worse against the Core i7 6900K than it should have been,and in the Hardware.fr test,without SMT their suite of games would have placed the R7 1800X at around Core i7 4790K level,ie,within 10% of a £1000 Core i7 6900K.
It was the same with the R9 290X - it shipped with one of the worst stock coolers in years,and caused the card to get the "hot and throttling" moniker which Nvidia PR actually managed to push since they actually sent review sites R9 290X cards and said they discovered how under certain settings the cooler had problems.
If AMD had actually made a better cooler,it would have not jinxed the whole R9 290 series line from the start.
Look at how well the R9 390 series rejig did in comparison - they launched with decentish cooling.
This may have factored into their decisions about which processors to launch, of course. The higher end chips are more likely to go to workstation users who'll want to load up the threads and won't have these issues.
Cinebench 15 shows that AMD SMT scaling in some tasks is better than Intel's HT scaling, so it must be a particular feature of game engines that causes AMD to lose performance with SMT on.
Ultimately, AMD had to launch some Zen CPUs eventually, and the longer they wait they longer they risk bumping into another Intel spoiler launch. At least Zen is a solid base going forward - let's see how much they can refine it over the next couple of iterations. Intel seem to have optimised Core about as far as it'll go now, so AMD may have a little breathing room to make those tweaks...
EDIT for crosspost:
See, that's simply not true. For the vast majority of users the difference would be completely unnoticable.
The vast majority of people use their computers to watch youtube videos, send emails, and update social media. The vast majority of gamers are on cards equivalent to an RX 480 or lower (dig out GPU sales by sector - < 20% is "enthusiast" level cards). They'd be absolutely fine with a Bristol Ridge APU or a mobile Core i U (which are basically low-clocked i3s). Give them a Zen + RX 480 rig and an i7 + RX 480 rig and I'd put money on them not being able to tell the difference.
Pleiades (02-03-2017)
Yes but it should have been documented and even passed onto reviewers - what did they expect would happen??
Its not a one off either - every single AMD CPU and GPU launch in the last few years has had some issue like this,which either AMD internal testing did not find or quietly ignored.
Intel and Nvidia are just exploiting all of this.
People will just read all the launch reviews and that is what they will get from Ryzen - crap for gaming just get Intel.
They need to get on top of this before the 4C and 6C models are released - be honest about the SMT gaming issues FFS.
10% performance drops are huge.
BTW,on a side-note did Hexus test the CPU under the high-performance power plan in Windows?? Apparently it adds a bit more performance in games according to some comments I read over on OcUK forums.
There are currently 1 users browsing this thread. (0 members and 1 guests)