Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Clockspeed is not the problem with Ryzen,as it does not scale well.
https://www.youtube.com/watch?v=X6RSEU1d-g8
That channel tested the Ryzen 7 3800X overclocked to 5GHZ with some extreme cooling.
This indicates there are other bottlenecks in the design. The Ryzen 7 4750G testing also indicates having the memory controller on-die,does not have a big effect. It seems to be inter-CCX latency,and games not being optimised for Ryzen which are the main problems in terms of gaming performance with the current generation. It seems IF speed also might be another bottleneck,as if you look at other testing with the Ryzen 3 3300X that channel does with WoW,there are some very impressive gains in performance.
This is probably why Zen3 is rumoured to have an 8 core CCX/CCD with a unified L3 cache,and I suspect IF frequency will also increase.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
CAT-THE-FIFTH
Clockspeed is not the problem with Ryzen,as it does not scale well.
This is probably why Zen3 is rumoured to have an 8 core CCX/CCD with a unified L3 cache,and I suspect IF frequency will also increase.
Yes exactly its the fabric speed between the dies and the caching. Looks like Zen3 is going to fix some of these issues + IPC gain + a little bit of clock speed its all shaping up nicely tbh all being well this will be a new system for me from my quad i7
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
CAT-THE-FIFTH
Clockspeed is not the problem with Ryzen,as it does not scale well...
Hopefully the new consoles will usher in a focus on working with Ryzen.
MS are doing a CPU rundown at HotChips too later this month, so will be interesting to see if they have anything borrowed from Zen 3.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
Terbinator
Hopefully the new consoles will usher in a focus on working with Ryzen.
MS are doing a CPU rundown at HotChips too later this month, so will be interesting to see if they have anything borrowed from Zen 3.
We've had 7 years of AMD CPUs in consoles, another generation won't change too much.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
Xlucine
We've had 7 years of AMD CPUs in consoles, another generation won't change too much.
We've had that many years of a completely different AMD microarchitecture used in consoles; it's hardly comparable.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
watercooled
We've had that many years of a completely different AMD microarchitecture used in consoles; it's hardly comparable.
TBH I think giving games devs more powerful CPUs will make them less bothered with threading as they can get away with less effort and a single core will carry them further.
So basically, if they didn't get the hint with the current gen, then next gen isn't going to improve multi core support in games.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Yeah that's a concern I share. Being bound to a 'small' CPU means they need to put some effort into optimisation and tuning. Having PC-equivalent cores means a fairly lazy port would work, albeit not optimally.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Do devs really care that much these days, I remember when we had lower spec machines and they'd spend huge man hours tweaking and optimising stuff, these days, just chuck more RAM or a faster CPU at the problem and it goes away....
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
Xlucine
We've had 7 years of AMD CPUs in consoles, another generation won't change too much.
There's a double the thread increase and (almost) the same again on clock speed (depending on console version) + uArch changes. In no way comparable to the Jaguar parts from 2012/3 relevant to existing desktop parts of the day.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
Xlucine
We've had 7 years of AMD CPUs in consoles, another generation won't change too much.
Big difference though this time. The previous generation were a totally different design to the desktop AMD CPUs,and I suspect many of the desktop development machines were using Intel CPUs,and then the games were optimised down to the console ones.. This time it is literally the same Zen2/Zen3 core being used. So if a game is developed on the dev kit(which probably will be Ryzen based),it should help with desktop AMD CPUs too.
Quote:
Originally Posted by
BigBANGerZ
Yes exactly its the fabric speed between the dies and the caching. Looks like Zen3 is going to fix some of these issues + IPC gain + a little bit of clock speed its all shaping up nicely tbh all being well this will be a new system for me from my quad i7
IMHO,the biggest improvements might be in games based on older engines.
Quote:
Originally Posted by
Terbinator
Hopefully the new consoles will usher in a focus on working with Ryzen.
MS are doing a CPU rundown at HotChips too later this month, so will be interesting to see if they have anything borrowed from Zen 3.
Considering the consoles and desktop Ryzen CPUs will use the same uarch it should help with optimisations with AMD CPUs.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
watercooled
We've had that many years of a completely different AMD microarchitecture used in consoles; it's hardly comparable.
Kabini wasn't renowned as a brilliant gaming CPU though, it was beaten by atom in dGPU gaming
Quote:
Originally Posted by
Terbinator
There's a double the thread increase and (almost) the same again on clock speed (depending on console version) + uArch changes. In no way comparable to the Jaguar parts from 2012/3 relevant to existing desktop parts of the day.
It's a step up in performance, but if developers didn't optimise for AMD then why would they now?
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
Xlucine
I'm not sure what you're getting at? 4C Kabini (a niche product anyway) vs 8C server-oriented Atom? And it still traded places depending on the game. They didn't even test the comparable Celeron part in the charts you linked because of platform limitations. However, the Kabini SoC convincingly beat the Celeron and even some 'big' processors in the IGP tests...
My point, similar to what others have said, is that the situation is not comparable as the past consoles were based on Jaguar cores, unlike AMD's desktop CPUs of the day, let alone Ryzen. Jaguar had little in common with either, so tuning for those cores would make little difference to AMD's desktop CPU performance outside of generic (not microarchitecture-specific) improvements from things like threading, which would have helped just as much on Intel platforms. Expecting Jaguar-specific optimisations to specifically help Ryzen performance is pretty much like expecting Skylake-specific optimisations to do the same i.e. they're totally different cores.
Another difference is that Ryzen is actually commonplace in gaming systems, unlike Kabini, so there's more incentive to actually carry across any optimisations/tuning.
Quote:
Originally Posted by
Xlucine
It's a step up in performance, but if developers didn't optimise for AMD then why would they now?
What makes you think they didn't optimise for the AMD cores actually used in the console? There's no such thing as AMD optimisations; such thing are specific to cores, not brands. This is particularly the case with AMD as they have had some drastically different cores over the past decade or so. Bulldozer was pretty much a fresh start from it's predecessor Phenom cores. Likewise Ryzen another fresh start. And the small cores separate again. In contrast, many of Intel's designs of the past decade or so have been more iterative rather than throwing everything out and starting over, so many optimisations will still help.
If you want to compare a time when Intel made a more significant change, check gaming performance on Nehalem when games were written for Core2. Games can be really quite sensitive to microarchitectural changes.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
watercooled
Jaguar had little in common with either, so tuning for those cores would make little difference to AMD's desktop CPU performance outside of generic (not microarchitecture-specific) improvements from things like threading, which would have helped just as much on Intel platforms.
Threading was TBH all I was really hoping for. Games could have had a kernel compiled with -march=bdver2 for me but that seemed unlikely and I don't expect CPU specific tuning beyond simple compiler flags.
There was a gradual push towards better threading in games, and I do wonder how much of that came from these consoles with sea of puny cores approach. It helped my FX8350 stay relevant in some games benchmarks well beyond when it should have done, but it always felt like in most games the threading was just good enough to hit some target. Like work was offloaded from the main thread until the main thread was going fast enough, so you get one core at 100% while the other cores fight over the scraps of work and are largely idle.
Now that might be because game engines being used had some highly tuned version on the consoles, and on the PC we didn't get the same engine. Dunno.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
I wonder how much games make use of CPU dispatchers, to pick the most optimal binary for different CPUs? Even if it is just compiler flags, it could make a substantial difference in some games. I'm assuming they must already to it to some degree to allow for different extension compatibility.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
DanceswithUnix
Like work was offloaded from the main thread until the main thread was going fast enough, so you get one core at 100% while the other cores fight over the scraps of work and are largely idle.
That's what happens with Bethesda RPGs (both TES and Fallout). One core is 100% while the rest are mostly idle. Found Wabbajack (helps download list of mods), and been playing Living Skyrim (800 mods and stable!), and too many AI doing stuff totally kills performance even if other cores aren't doing much.
Re: AMD Ryzen 9 4950X claimed to be a 4.8GHz boost 16C/32T CPU
Quote:
Originally Posted by
kompukare
That's what happens with Bethesda RPGs (both TES and Fallout). One core is 100% while the rest are mostly idle. Found Wabbajack (helps download list of mods), and been playing Living Skyrim (800 mods and stable!), and too many AI doing stuff totally kills performance even if other cores aren't doing much.
Its the same with Fallout 4 - I tested a ton of mods,and have nearly 400 running in my current load order(all manually installed). More NPCs tends to hammer one of the threads massively. However,another problem is draw calls,and that too hammers another thread. For example,in some of the biggest settlements I made,draw calls were in the 10s of 1000s!!
Edit!!
Look at my Ryzen 5 2600 review for example:
https://forums.hexus.net/pc-hardware...ml#post4020716
That is a load order with 100s of mods,many 4K mods,and upto 60 NPCs per settlement. I also had between 10~20 followers too.