![[H]ard|Forum](/styles/hardforum/xenforo/logo_dark.png){kind=logo}
Interesting perspective on Ryzen for gaming.
- Thread starter SpeedyVV
- Start date
Hi, still quite new lurker. It was definitely interesting as you say, and I've heard some of these ideas this guy has expressed before and thought I might go for Ryzen. The only thing holding me back right now from Ryzen are mb features I want (was looking at formula code for 7700k). I might postpone next build a little longer now though and hope for better board features, or buy BT and NIC separately, and use lower speed memory and get this thing going.
ZeroBarrier
[H]ard|Gawd
- Joined
- Mar 19, 2011
- Messages
- 1,495
If you're waiting like I am, I'd recommend waiting for Intel's next release to compare it to Ryzen. Coffee Lake might be a good competitor to Ryzen and I'm sure the boards will have more features than what Ryzen can offer.
I'm going to weigh Ryzen against Coffee Lake to see which one better suits my wants and needs. It will likely include a 6C/12T chip in the line up which should be promising.
Gideon
Supreme [H]ardness
- Joined
- Apr 13, 2006
- Messages
- 4,291
Coffee Lake will be going against Zen +.
ZeroBarrier
[H]ard|Gawd
- Joined
- Mar 19, 2011
- Messages
- 1,495
Highly doubtful, the only buzz going around is Raven Ridge (mobile APU) for the second half of this year. And I don't care about mobile APU's; I'm not looking to buy an OEM laptop or a barebones laptop, I'm looking at building a desktop.
JustReason
razor1 is my Lover
- Joined
- Oct 31, 2015
- Messages
- 2,483
If you look at the time between Bulldozer and Piledriver it was one year and a month. So it is still possible Zen+ will surface by early next year.
Liquid_Static
Gawd
- Joined
- Feb 9, 2013
- Messages
- 628
This is an incredibly interesting video! Thank you for sharing.
Coffelake is going to be Kabylake with slightly better power efficiency, some slightly higher clocks, 2 more cores on the high to mid range parts, and a somewhat better iGPU.
Mobo wise not much will change at all for the desktop. For HEDT Purley platform should be there around that time and that may bring some interesting features for power users but won't mean much of anything for gamers and overclockers.
Zen+ is supposedly Q1-Q2 2018 which is pretty close the Coffeelake's desktop launch (Feb. 2018) time line. The problem there is its all rumors at this point. Particularly for Zen+ the rumors are very squishy since supposedly they might be going for GF's "7nm" process which won't be starting production until mid 2018 which means a late 2018 launch. Hard to say at this point.
If GF's "7nm" (using scare quotes because its complete BS marketing at this point, its more like a much improved 16/14nm process) does manage to pull of its rumored specs (30% more clocks OR 60% better power efficiency) and Zen+ really does end up giving 10-15% more IPC then they could actually have a product that beats Intel's best for the first time in years. That is a very big "could" at this point of course but still damn interesting to think about.
ZeroBarrier
[H]ard|Gawd
- Joined
- Mar 19, 2011
- Messages
- 1,495
Better clocks and better power efficiency on a very roubust and feature rich platform doesn't sound at all bad to me. It will provide almost no issues and you don't have to wait for the promise of potential fixes that may never come to fruition.
Going to 7nm; I have my doubts they will skip 10nm entirely, especially on desktop chips. It just doesn't make sense.
Neither Zen nor Zen+ have peaked my interest any further than hoping that AMD at least remains a competitor. All the hub bub about Zen on these forums when the vast majority probably only benchmark handbrake and CB ad nauesum to feel better about their purchase because most of them aren't content creators in the slightest. Which is funny though, since tests have show Ryzen has a very, very long way to go.
http://www.tomshardware.com/reviews/amd-ryzen-7-1800x-cpu,4951-9.html
As it says in Tom's hardware review; the bottom line, it's a general purpose CPU. Content creators will still likely go with Intel system because they still provide the best performance (which in some cases like cad and Maya is miles ahead when compared to Zen). This also means developers will likely not be putting a lot of time and money into trying to optimize for a small fraction of users that will use Ryzen.
This is just my opinion, theory and feeling; but I think AMD has probably created a very competitive and great server chip in Naples. And they decided to release Ryzen and drum up hype plus open preordering (something never before done for CPUs, ever!) in an attempt to get as much funds as possible off the mainstream market. Why else would they open up preorders when they strictly enforced review NDAs until preorders where done and over?
The differences will be very minor and you won't see them both at the same time either. That is the higher clocking parts will draw even more power since its all still on the same process as Kabylake.
The biggest change that will matter to gamers is 2 more cores for a (possibly) non-stupid price from Intel.
Its not really 7nm. That is why they're able to make such a "jump". Public process names are pure marketing at this point.
That doesn't make any sense for you to say though. Zen+ really could beat Intel's Coffeelake in terms of performance at stock clocks. OC is a different story but the difference may not to turn out to be all that big and the market at large doesn't care about OC stuff. That is going to make some real market changes happen one way or another.
He's correct in that ultra low res tests are probably not as useful as we like to think. Games evolve and distribute workloads in a different fashion than they did 3 years before, so using low res benches as a predictor for future performance is unreliable at best.
I think he misses the mark on two aspects though.
First, he suggests that games need to be optimized for Ryzen. Outside of a couple games where there seems to be some real SMT problems, this seems unlikely. Yes, it will help when games are optimized for *more threads*, but that's not a Ryzen specific optimization. This idea that the so-so gaming performance is because game developers didn't optimize for AMD seems far fetched to me. If game developers were supposedly not optimizing for AMD because of performance/market share, then it would stand to reason that no developers of any kind software were optimizing for AMD. And this excuse would hold true for all types of performance. If game devs didn't feel the need to optimize for AMD, I doubt devs of rendering software felt the need to either. And yet Ryzen performs brilliantly at that. Ryzen will continually benefit from increasingly threaded workloads for sure, but this is not a Ryzen specific trait.
Secondly, I somewhat agree with him poo-pooing ultra low res tests. I disagree with him (and others) who poo-poo the 1080p tests. There are a lot of us 1080p high refresh rate gamers out there. 1080p is certainly not an invalid or contrived test, the same way 640x480 is.
I think he misses the mark on two aspects though.
First, he suggests that games need to be optimized for Ryzen. Outside of a couple games where there seems to be some real SMT problems, this seems unlikely. Yes, it will help when games are optimized for *more threads*, but that's not a Ryzen specific optimization. This idea that the so-so gaming performance is because game developers didn't optimize for AMD seems far fetched to me. If game developers were supposedly not optimizing for AMD because of performance/market share, then it would stand to reason that no developers of any kind software were optimizing for AMD. And this excuse would hold true for all types of performance. If game devs didn't feel the need to optimize for AMD, I doubt devs of rendering software felt the need to either. And yet Ryzen performs brilliantly at that. Ryzen will continually benefit from increasingly threaded workloads for sure, but this is not a Ryzen specific trait.
Secondly, I somewhat agree with him poo-pooing ultra low res tests. I disagree with him (and others) who poo-poo the 1080p tests. There are a lot of us 1080p high refresh rate gamers out there. 1080p is certainly not an invalid or contrived test, the same way 640x480 is.
I do get the feeling this video is from an AMD fanboy.
But still interesting to hear his points, and use my own knowledge and experience.
As far as Intel goes, i have the feeling they have been sandbagging for a long time, so if they want to, they would be able to leave AMD in the dust in future releases.
Mind you, I'm also pretty confident they were surprised by how much AMD caught up.
Last but not least, if there is continued use of AMD on consoles, the gaming CPU landscape will be ever more interesting in the years ahead.
But still interesting to hear his points, and use my own knowledge and experience.
As far as Intel goes, i have the feeling they have been sandbagging for a long time, so if they want to, they would be able to leave AMD in the dust in future releases.
Mind you, I'm also pretty confident they were surprised by how much AMD caught up.
Last but not least, if there is continued use of AMD on consoles, the gaming CPU landscape will be ever more interesting in the years ahead.
I don't think they've been sand bagging, so much as "necessity is the mother of invention."
Radical changes in architecture is risky in the CPU world. With no real threat to their lead, the consistent and safe, "tweak things, get slightly better IPC and clock speed" was a reliable strategy with little downside for them.
Radical changes in architecture is risky in the CPU world. With no real threat to their lead, the consistent and safe, "tweak things, get slightly better IPC and clock speed" was a reliable strategy with little downside for them.
No argument there.
I would have done exactly the same.
Sandbagging, as in, if they had to, they could have.
- Joined
- Apr 29, 2005
- Messages
- 19,197
I have a theory. Ryzen is doing exactly what its supposed to be doing, crunching, and its very good at it. I think AMD is delivering its 1-2 punch, first Ryzen for non gaming processing, then Vega for gaming.
When you combine the two?
When you combine the two?
They're not sandbagging. They've been focusing on getting CPU power consumption down both under load and at idle while also improving iGPU performance. If they put the same engineering effort into getting more CPU performance and increased their TDP's you'd see plenty more performance from Intel chips right now out of the box.
Since process improvements are slowing down greatly, Intel and everyone else, is going to have to start getting real creative with their designs and engineering to provide more performance. Its really hard to say how all that will pan out at this point. Its quite possible that AMD and Intel will eventually only be able to add more CPU performance under certain situations rather than in general.
You raise some good points. It's quite possible and I believe more than likely that we'll see the the old x86 instructions go by the wayside and be software emulated as they focus more transistors into specialty purposes for things like server oriented instructions etc.. I mean they can only shrink so far, silicon costs are not going to decrease, so the obvious answer is to do more specific specialization where competitors like ARM can't easily catch up and gain market share.
Peppercorn
Limp Gawd
- Joined
- Dec 8, 2016
- Messages
- 259
I went for the Tomahawk board. I was hoping to get some G.Skill FlareX that is optimized for Ryzen but i couldnt wait.
Its a blast.Peppercorn
Limp Gawd
- Joined
- Dec 8, 2016
- Messages
- 259
Maybe AMD will stay on 14nm for a couple years yet. I think they have some breathing room there because intel is somewhat handcuffed now by not being able to get to 10nm in production. They hit a roadblock that gives AMD valuable time to optimize and advance the platform. I think AMD is in a better position than anyone expected.
The size of a carbon atom is around 0.23nm. At 10nm we are about pushing the limits for transistor size. If you could stand on a 10nm transistor you would probably be able to see an ocean of atoms looking like little balls of energy forming crystalline lattice around you.
We are going to need to move to a new platform. I think we have hit the limit of silicon as far as approaching to atomic size limit of transistors.
The next big realm to enter is Quantum computing. Qubits are 3 and 4 dimension depending on the opinion behind defining them. Where it stands now a transistor that is approaching atomic size can only gate on and off and nothing else. However a quantum "transistor" for lack of a better term, can open close, change shape, reorganize it's self. Do all kinds of crazy things that can be harness for data processing.
They've been doing this for years.
AMD started doing that with the K5 (which emulated all of the x86 ISA, the "front end" would translate x86 ISA instructions into RISC instructions for the 29K CPU "back end" at its core) and Intel started doing it with the Pentium Pro back in the mid 90's. Some of the more arcane and seldom used x86 instructions can take quite a while to execute and I believe AMD once said publicly they could take up to 1000 cycles on the K7. Who knows how long on their more modern arch's. Its very slow to do software ISA emulation even in the CPU but that is OK since it's virtually never used outside of very old software that was designed with much slower CPU's in mind so they still run fast as hell on a modern chip anyways.
Its kind've a given that more x86 ISA instructions will be dumped into software emulation but how exactly AMD or Intel go about this is anybody's guess at this point.
Simplifying and cutting the cruft out of the ISA or adding ever wider and more complex vector FPU's will help for some things but 3D die stacking has far more potential to boost general software performance without any re-writes or re-compiling which is a pretty big deal. Even if its just to throw a giant cache on package it'll help. 1GB of fast low latency eDRAM, SRAM (lol not happening), or HBM could really help here. Especially with core and thread counts now starting to get high.
I think it'll all come down to how well GF executes on their process development. If everything goes well and its ready by early 2018 I don't see why they wouldn't use GF's "7nm" process. 30% more clock OR 60% reduced power usage is a pretty big deal. Smaller CPU dies will be nice too once the process matures and costs come down.
If it really does turn out that GF won't have their "7nm" process ready until mid 2018 then I think AMD might launch Zen+ on 14nm. You probably won't see much if any improvement on clockspeed, OC'ing, or power usage but 10-15% more IPC is still a nice jump and worthwhile.
http://www.realworldtech.com/intel-10nm-qwfet/
Quantum computing is still in its "learning to crawl" stage and will remain, at best, only sort've practical for people in the deep pocket HPC arena. Its entirely possible they never get affordable enough for avg. consumers. Also for some things quantum computers (ie. DWave's work) are inherently no faster than existing CPU's. They are inherently blazing fast at stuff like decryption so that is part of the reason why so much work is being done on them. They will however not be a performance silver bullet. Its quite possible they'll just end up as a specialized hardware on a classically architected silicon CPU.
Last edited:
Peppercorn
Limp Gawd
- Joined
- Dec 8, 2016
- Messages
- 259
I agree. I think they are going to want to take 14nm as far as they can, at least until 7nm has been in production for more than a few months. I think it would be suicide to risk 7nm so early in production as it would be very difficult to match 14nm volume and price at such an early stage. They have a great product now to refine and improve still on 14nm.
5nm is the theoretical limit, then there are other technologies that can push it further than 7nm, like nano tubes, graphene etc. But those are still very early in development.
Listening to you guys, the more I think the future will have to rely on more parallel software.
Its a combination of both, but by the time 5nm is there will be technologies that will push further.
Intel already tried that with EPIC and it didn't really pan out so well with general purpose work loads. Basically you need ~~magic~~ compilers to extract huge amounts of ILP in order to make parallel software work for general purpose stuff. The stuff that is relatively easy to parallelize has probably already been shifted onto either the vector units (ie. SSE, AVX256, AVX512) or GPGPU's already, both of which excel at parallel tasks.
As I said some time ago 2018 is Zen refresh on a improved '14nm' node (I guess we can expect 200--400MHz higher clocks than RyZen) and Zen+ (Zen2) on '7nm' for late 2019 or early 2020. Glofo has just confirmed they expect '7nm' production to start in 2019
http://www.reuters.com/article/us-globalfoundries-japan-idUSKBN18R1J9
I use quotes for the nodes because
Glofo '14nm' ~ 20nm
Glofo '7nm' ~ 14nm
Gideon
Supreme [H]ardness
- Joined
- Apr 13, 2006
- Messages
- 4,291
He is doing damage control for Intel so he posts things to try and make AMD look bad. Your better off to just ignore him, most everyone else has and hes down to mostly talking to Shintai.
sirmonkey1985
[H]ard|DCer of the Month - July 2010
- Joined
- Sep 13, 2008
- Messages
- 22,412
lol i was trying to figure out who mesyn was talking to then i saw your post and realized exactly who it was even though i can't see his posts.
This. People can't seem to grasp that some tasks by their very nature do not benefit from parallel processing. No magic compiler or software trick we have today will change that. You still need a processor to have either an excessive single core clock speed or solid IPC performance to improve performance in those tasks that do not lend themselves well to parallel processing.
Wshmaster0
2[H]4U
- Joined
- Jul 11, 2007
- Messages
- 3,713
What about a magic dongle? :-D
Pieter3dnow
Supreme [H]ardness
- Joined
- Jul 29, 2009
- Messages
- 6,784
Hence the IPC power of Jaguar cores those must be beastly
in current generation consoles.I get your point that we would stop quoting what Globalfoundries have to say about its own roadmap and leave silly rumors about 7nm next year take the forums. The funny part is that the people is hyping AMD products with impossible/nonsensical claims is the people really hurting the company.
Hence the IPC power of Jaguar cores those must be beastly
He mentioned both IPC and clocks. Funny enough several devs complained precisely about the anemic performance of those Jaguar cores. And latest consoles are trying to correct that by a huge increase on clocks (e.g from 1.6GHz to 2.1GHz) instead adding moar cores up to a total of 16 threads.