Discussion AMD Ryzen MegaThread! FAQ and Resources

[juanrga]

We need to keep a couple of things in mind in regards to gaming and reviewing the currently released Ryzen processors. Number one first and foremost Ryzen just released, and they have bios issues, driver issues and optimization issues to work out in regards to gaming. Its not hard to look at gaming in the last 10 years and come to the logical conclusion that games were optimizing their software to Intel hardware as that was really the only player in the game. This is evident when a couple games which were actually optimized for the old FX Piledriver performed better on those processors even though they were not strong enthusiast gamers.

We also have to take into account that the Ryzen R7 line that released IS NOT RYZEN'S GAMING LINE OF PROCESSORS. As experts none of us would recommend a i7 6900K or 6950X over the i7 7700K if the person is interested in gaming first and foremost. First of all both of those processors are way overpriced, and second the i7 7700K smokes both of them in pure gaming benchmarks due to ever so slightly better IPC and much higher clock speeds. Yet we are all comparing the R7 processors directly to the i7 7700K and calling it a bad gamer because it loses to it, by that logic we then have to declare the i7 6700K and 6750X to be bad gamers because they get smoked by the i7 7700K in most benchmarks. While It may be in the same price range as the i7 7700K, the R7 processors were intended to compete against the i7 6900K, not the i7 7700K. In that regard, and especially when both processors are utilized as workstations (which with 8 cores and 16 threads is their main function- if you want a gaming processor the 7700K is much better), the R7 series competes very well against Intel. In workstation related applications the R7 series takes it directly to the i7 6900K which is more than twice its cost.

As you pointed out, the R7 series even at this point isn't a bad gamer and will only get better as drivers, bios and game optimizations are made. AMD's true gaming line will release with the R5 series processors. Those are what should be compared to the i7 7700K as they will feature less cores and should be able to obtain higher overclocks than the R7 series sporting in some cases twice the number of cores. By the time they release hopefully the driver and bios issues are resolved and we have some games optimized for the new architecture so we can truly see the gaming potential of the Ryzen line.

Final though- as I have stated before gaming isn't everything. The R7 series was never intended to be the gaming processors of Ryzen, they are the workstation processors of Ryzen. In that role they truly shine and are a huge success. While we are all muddled down with the gaming performance of a WORKSTATION PROCESSOR we are also largely ignoring the implications the R7 series sets for Naples server processors. The server market is where the money is, where AMD has to make up the most ground, and is where AMD is currently poised to take a huge bite out of Intel.

It is AMD which advertised Ryzen R7 as a gaming chip. It is AMD which provided gaming demos of R7 vs 7700k. It is AMD fault, not everyone else.

The R5 are not better gaming chips. The quad-core Ryzen chips use worse silicon that the flagship R7 models and have lower clocks. The top R5 1500X has 3.5GHz base and 3.7GHz turbo. It will play games worse than the 1800X because has lower clocks and half the cores/threads.

 

[jdwii]

Juan i also agree same old arguments that Amd used with bulldozer i just wish they would stop this and admit the truth. Phenom was amazing it really was i owned a 1100T it wasn't that far behind Intel in gaming at the time. That is what Ryzen is stop for the love of god stop claiming this or that will fix everything i mean even if Amd fans can think logical for a second how many times has Amd repeated this story? Bulldozer-Vulkan-12-now Ryzen. Them telling reviewers to basically make a GPU bottleneck with a CPU test and so forth this is not right.

I say this for the 10th time Ryzen is not bad at all and i expect their 4 core to be a better offering then the I3 for gaming but Amd themselves are acting unprofessional and Amd fans are being crazy like they have been in the bulldzoer days. I said before this processor released that i will not let that fly and that i hoped it wouldn't happen YET again.

I have respect for 8350rocks, Yuka and more i really do but we all have to agree on something if Intel pulled the same stuff you would be freaking out i know i would as i did with Intel on their overclocking claims with haswell and i'm doubting Nvidia on their 35% improvement with their 1080Ti over the 1080 when even a Titan XP doesn't see that return.

I'm actually Pro Amd i started with them and i want to go back with them but i'm a objective person i look at things that way.

 

[juanrga]

Does anybody know why exactly Ryzen overclocks way worse than Intel? I know that getting 8 cores to overclock at a high frequency is much harder, but the headroom is so small that it's almost hard to believe.

I'm also wondering about this because AMD is using the same 14nm FinFET right?

I have been saying during years in the old threads that 14LPP means "14nm Low Power Plus", that this is a node optimized for mobile applications, with an optimal range sub-3GHz, that AMD was going to have a problem with stock clocks, and that Ryzen was going to overclock really bad.

Linear scaling is broken around 3.3GHz and there is a voltage wall around 4GHz. It is not Globalfoundries fault. AMD is using 14LPP for it wasn't designed. The process node optimized for higher clocks is Globalfoundries 14HP, this is "14 nm High Performance". 14HP is the process used by IBM for its Power9 CPUs, which are expected to ship with frequencies in the 4.0--5.0GHz range.

 

[juanrga]

http://Forbes

Verbal confirmation with Forbes that r5 6 core flagship to bump speeds up to 3.6/4.0 a la 1800x?

Question being did AMD plan this earlier - leaking lower core speeds to keep Intel off guard, or is this in direct response to reviewers?

The specs of the flagship six-core have been confirmed by AMD. You can find they here

http://www.tomshardware.co.uk/forum/id-3327589/amd-ryzen-megathread-faq-resources/page-14.html#19376036

The 1600X has same clocks than the 1800X but 2C/4T less. It will not game better than the 1800x. AMD shipped the 1800X to the press because it was the best chip and wanted to generate a halo effect.

 

[Obakasama]

Tentative steps onto this board...

With the Windows scheduler issue coming to light I wonder if anyone could clarify something for me. From what I've heard/read it doesn't handle SMT well which may account for some of the benchmarks for games (performing better in some titles with SMT off).

But wouldn't this particular issue have an effect on (almost?) all tasks in a Windows environment? We see excellent productivity benchmarks which rely upon multiple threads.

What am I missing?

 

[Redneck5439]

juanrga, jdwii,

I just wanted to make something clear- as a professional I can't endorse one brand over another, it has to be based on price to performance. In that aspect in gaming it depends mostly upon budget, but if the budget is there Intel has always been the recommendation and will continue to be so for now. In my personal builds I almost always build with AMD as I don't like Intel's "business model".

I also wanted to make clear that I do support all the postings I've made here, but I'm also a realist. I don't expect that any fixes to Windows, DX12, bios or drivers is going to magically have Ryzen going toe to toe with Intel in enthusiast gaming. Ryzen isn't going to beat the Kaby Lake i7 7700K in gaming period, it isn't going to have the overclocking headroom and doesn't have quite as good IPC, its just not going to happen. What I do expect to see is much better gaming results when Windows, DX12, bios and driver updates occur. I fully expect that Ryzen will be gaming near Skylake levels so instead of benchmarks between Ivy Bridge and Haswell (closer to Haswel) we will be seeing benchmark results between Hawell and Skylake (closer to Skylake). With what Ryzen has already shown us in workstation based applications one can directly correlate that Ryzen should be gaming between Haswell and Skylake levels, with benchmarks much closer to Skylake than Haswell.

Now what that means is the R5 isn't going to be the new go to gaming processor for people with a good budget, but it will be a very good mid range level gaming option and there is a lot of growth and sales in that niche.

I'm also expecting that AMD has learned a lesson or two and this time around actually puts some emphasis on its 6 core and 4 core flagship components. I'm expecting the flagship 6 and 4 core processors for Ryzen to have very good silicon and hopefully AMD even uses the best silicon and best binning for these flagship components. If they do, which would be smart, the flagship 6 and 4 core processors will have good overclock headroom and be able to give them credibility in the gaming market. Using the R7 line as a template and seeing that they can overclock to 4Ghz with 8 cores and 16 threads (while their Intel counterpart can typically hit 4.3Ghz) it is within reason to expect the 6 core components to be able to overclock to at least 4.3Ghz and the 4 core components to hit at least 4.5Ghz. Any better overclocking headroom than that would just be a bonus, but I think those overclocks are well within reason. As stated, no the R5 isn't going to come out and go toe to toe with high end Kaby Lake (especially when overclocked), but they can close the gap and look very good price to performance wise.

 

[Crumpet 1]

To be honest, if the R5 can produce the SAME gaming results as the R7 series, but for $260 vs $350 for the i7 7700k, with vastly cheaper motherboards, for the same price you could stick a significantly better GPU in the AMD build...

 

[jaymc]

@juan
I was just wondering how difficult it would be for them to change over to the 14hp node... an is it even the same size ?

By the way...
I hear Ryzen is running like a swiss watch on linux... an the patch's are out already to fix scheduling... Anyone have any experience with this...

I would love to see some stats on gaming on Linux.... !!!!!

Also might wanna check these guys out:
https://www.youtube.com/watch?v=bIbwuLdHbMg

 

[cdrkf]

The way some are talking here you'd think Ryzen was a terrible gaming chip- it really isn't though is it? In the very *worst case* results it's still 40 - 50% faster than the best FX parts (which is a massive jump). No it isn't as fast as Intel's very best gaming chip- and no it won't overclock far enough to change that (@gamerk just to point out the people 'warning' about clock speeds on the glofo process declared it wouldn't pass 3ghz, they've done much better than I initially thought).

@jdwii, you saying that 'AMD aren't competeing in high end gaming'- I'd argue that the R7 series is fast enough to be considdered. I agree actually that the R7 parts are imo very much like AMD were with Phenom II (although arguably a touch better as Phenom II X6 didn't really outrun the top i7's of the day even in workstation stuff, Ryzen is the outright fastest part in a few tests core for core which is increddible).

I'm just not getting the 'backlash'. As for lower chips- I still think a 6 core, 12 thread part at same speeds will be the best option for gaming on Ryzen as I don't see it giving up anything meaningful in performance and it'll cost less. I don't expect it to overclock much more though (at least until AMD give us per core overclocking). I also agree with @jdwii that the 4 core 4 thread part for i3 money will look very tempting to a lot of people- especially when there are some decent parts higher up the stack if you need more threads (again, probably not so much for gaming right now but there are lots of people who do more than *just* game on a high end PC). I do think that is something that will come forward as a stand out feature of Ryzen moving forward actually- the fact AM4 is this monolithic, unified platform makes it a very interesting proposition as a 'foundation' build. Intel don't really offer anything directly comparable as they switch sockets in the middle of the stack meaning an upgrade necessitates a new motherboard, I'll wait to see benchmarks but I have a feeling as a future proof budget build R3 will be unbeatable. Pair it with a mid range B350 motherboard and you've got something that can be scaled up nicely in the future if required- that was also a strength of Phenom II on AM3, I started with an old X3 710 then later jumped to the 1055t- it's nice to have that flexibility as an option.

 

[gamerk316]

Tentative steps onto this board...

With the Windows scheduler issue coming to light I wonder if anyone could clarify something for me. From what I've heard/read it doesn't handle SMT well which may account for some of the benchmarks for games (performing better in some titles with SMT off).

But wouldn't this particular issue have an effect on (almost?) all tasks in a Windows environment? We see excellent productivity benchmarks which rely upon multiple threads.

What am I missing?

Benchmarks tend to load all cores up to 100%. This hides any scheduler effects. Games don't, and are more sensitive to scheduler problems. So benchmarks such as cinebench wouldn't be affected, but games would.

I note I don't have any hard proof of scheduler problems, but it certainly looks like it to me, and I've speculated a few ways it could be manifesting.

 

[James Mason]

Hmm... if these are lower power chips, maybe AMD wants to bust back into the laptop market.

I mean, a 1800x would be really impressive in a laptop with it's performance, and "workstation" grade CPUs in a mobile form would be very popular with the people who use them.

 

[jwcrellin]

^^^^Next gen console chips.

 

[cdrkf]

http://Forbes

Verbal confirmation with Forbes that r5 6 core flagship to bump speeds up to 3.6/4.0 a la 1800x?

Question being did AMD plan this earlier - leaking lower core speeds to keep Intel off guard, or is this in direct response to reviewers?

The specs of the flagship six-core have been confirmed by AMD. You can find they here

http://www.tomshardware.co.uk/forum/id-3327589/amd-ryzen-megathread-faq-resources/page-14.html#19376036

The 1600X has same clocks than the 1800X but 2C/4T less. It will not game better than the 1800x. AMD shipped the 1800X to the press because it was the best chip and wanted to generate a halo effect.

I agree the 1600X won't game 'better' than the 1800X- I think it will be the 'gaming pick' on the basis that I think it will offer pretty much identical performance to an 1800X at a significantly better price. The price is the key though... at $299 it's probably too close to the R7 1700 (which over clocks to match the 1800X anyway), however I'm guessing it will be priced in the $250 - $270 range at a guess, which makes it look quite interesting to what Intel offers in that range- even in gaming.

My issue with selling the 1800X on it's gaming capability is it doesn't make sense given the $500 price tag (which is appropriate for the thing's it's good at). That is where AMD got the gaming message very wrong imo. I will be interested to see the performance disparity between the 1800X, 1600X and 1500X to see if / where the core count really comes into play. I maintain from what I've seen the majority of games top out at 8 threads (which neatly matches the consoles and the higher end of typical gaming cpu's) so in theory at the same clocks all 3 are likely to be equal (note I acknowledge the 1500X is specified to be a touch slower on clocks, although I'm sure someone will overclock it to match).

 

[cdrkf]

Hmm... if these are lower power chips, maybe AMD wants to bust back into the laptop market.

I mean, a 1800x would be really impressive in a laptop with it's performance, and "workstation" grade CPUs in a mobile form would be very popular with the people who use them.

Well, the R7 1700 is only 65 watts... I could see them bringing that down quite a bit with slightly lower clock speeds. I'm not sure about a full 8 core 16 thread laptop part but 6 core 12 thread down at 40w territory sounds plausible. It would probably have a 2.x ghz base with a very wide turbo range up into the upper 3.x range. It could be a compelling option for a mobile workstation.

 

[ScrewySqrl]

Hmm... if these are lower power chips, maybe AMD wants to bust back into the laptop market.

I mean, a 1800x would be really impressive in a laptop with it's performance, and "workstation" grade CPUs in a mobile form would be very popular with the people who use them.

Well, the R7 1700 is only 65 watts... I could see them bringing that down quite a bit with slightly lower clock speeds. I'm not sure about a full 8 core 16 thread laptop part but 6 core 12 thread down at 40w territory sounds plausible. It would probably have a 2.x ghz base with a very wide turbo range up into the upper 3.x range. It could be a compelling option for a mobile workstation.

a Laptop APU part with a 4C/4T and notional RX 450 graphics in the 35-40W range would be a very solid lappy. a 2/4T with even weaker graphics (notional 430?) in the 1.9GHZ turbo 2.5GHz could be a ultrabook/2-in-1/Tablet chip in the 10-15w range

 

[cdrkf]

Hmm... if these are lower power chips, maybe AMD wants to bust back into the laptop market.

I mean, a 1800x would be really impressive in a laptop with it's performance, and "workstation" grade CPUs in a mobile form would be very popular with the people who use them.

Well, the R7 1700 is only 65 watts... I could see them bringing that down quite a bit with slightly lower clock speeds. I'm not sure about a full 8 core 16 thread laptop part but 6 core 12 thread down at 40w territory sounds plausible. It would probably have a 2.x ghz base with a very wide turbo range up into the upper 3.x range. It could be a compelling option for a mobile workstation.

a Laptop APU part with a 4C/4T and notional RX 450 graphics in the 35-40W range would be a very solid lappy. a 2/4T with even weaker graphics (notional 430?) in the 1.9GHZ turbo 2.5GHz could be a ultrabook/2-in-1/Tablet chip in the 10-15w range

Very true- although when I say 'mobile workstation'- well the last work machine I purchased was a laptop with an i7 6700HQ (8 thread) with 16gb of ram and a Quadro gpu for 3D CAD work. Whilst the i7 has integrated graphics I'd actually rather swap that for a couple more cores / threads and rely on the discrete gpu. It would harm battery life however I don't think I ever use the machine without being plugged in, save opening it up briefly to show something in a meeting.

I view a Ryzen APU as a more mainstream part, I think there's a gap in the market for a proper 'high end' laptop oriented chip though (that doesn't require fitting into a chassis so large it may as well be a desktop anyway).

 

[juanrga]

http://www.techspot.com/review/1348-amd-ryzen-gaming-performance/

In addition to including more games, we're also adding results for the 1800X and 1700X with SMT disabled as Anandtech forum-goers have discovered a problem with the Windows 10 scheduler that can cause Ryzen to perform worse in lightly-threaded applications with SMT enabled.

But:

From the 16 games tested we see that disabling SMT on the 1800X resulted in 3% more performance for the average frame rate and just 1% for the minimum.

Disabling SMT didn't change anything because the problem with RyZen is other.

 

[jaymc]

Bacause Core's are not all using shared cache the contents of cache has to be copied from ccx to ccx as threads are moved from on core to another.
As a tread is allocated to an idle core suddenly the required cache is missing and has to copied from another ccx again, I can see how this could cause a lot of problems.

It's not just as simple as fixing scheduling...

 

[eathdemon1]

Bacause Core's are not all using shared cache the contents of cache has to be copied from ccx to ccx as threads are moved from on core to another.
As a tread is allocated to an idle core suddenly the required cache is missing and has to copied from another ccx again, I can see how this could cause a lot of problems.

It's not just as simple as fixing scheduling...

your right. ms has to treat it as 2 cpus, not one. the question is do they put that in win 10 home. multi cpu support is in win10 pro if ms treats it as 2 cpus this issue goes away.

 

[juanrga]

Confirmation the problem of Ryzen is on unusually high latency

https://www.techpowerup.com/231268/amds-ryzen-cache-analyzed-improvements-improveable-ccx-compromises

as some of us said before launch

https://twitter.com/juanrga/status/836503836260990976?p=v

What I heard is that this problem is expected to be solved for Zen+.

 

[thegentlewoman]

Did anybody read about this ?

AMD WAS: Overclocked at 5.8 and 6.0 Ghz ... 2 days after it came out.
Do you think it is fake? I guess it's true.

<link removed by moderator>
<as this is the English version of Tom's Hardware please post only English language links>

It's an italian Post, advertorialism, (like most of bs today around internet for easy traffic money) just saying ...

Still people saying it's just a ... failure? Please go to bed. Why is everybody still wining here? Can't we continue construct a positive atmosphere and share good opinions... besides the fact that there is not a lot to day but just to wait 1 month...

 

[cdrkf]

http://www.techspot.com/review/1348-amd-ryzen-gaming-performance/

In addition to including more games, we're also adding results for the 1800X and 1700X with SMT disabled as Anandtech forum-goers have discovered a problem with the Windows 10 scheduler that can cause Ryzen to perform worse in lightly-threaded applications with SMT enabled.

But:

From the 16 games tested we see that disabling SMT on the 1800X resulted in 3% more performance for the average frame rate and just 1% for the minimum.

Disabling SMT didn't change anything because the problem with RyZen is other.

Hmm I read that article and you missed out an important part of the quote- the fact is that disabling SMT improved performance in quite a few titles *however* it reduced performance in titles that weren't effected badly in the first place- so the *average* negated the effect.

It's worth keeping in mind however that a patch to the OS / or games themselves should *fix the specific games that have an issue, without adversely effecting those that don't*- which would in fact have a net benefit overall of probably around 10%. No it's not going to change the overall picture that much but taking that one line out of context ignores the point that SMT on Ryzen *is a problem* for a few specific titles.

Edit: The full quote (which I don't think lines up quite how Juan is potraying it):

"The State of Ryzen for PC Gaming
As noted in our full Ryzen review, at this point we're waiting for software optimizations and AMD promises they're coming. Of course, the company also said that about Bulldozer, but Ryzen appears to be in a much better situation.
Fixing SMT support should be a relatively easy first step and it will help improve many of the results. Ryzen's overall average performance didn't look great but that's because the results in some games were actually hurt by turning off SMT. Those titles included Arma 3, Battlefield 1, Mafia III and Watch Dogs 2.
Conversely, disabling SMT boosted performance to some extent in Deus Ex: Mankind Divided, F1 2016, Far Cry Primal, Gears of War 4, Grand Theft Auto V, Overwatch and Total War: Warhammer, while we saw virtually no performance change in Civilization IV, For Honor, Hitman, Mirror's Edge Catalyst and The Division."

*Note that I'm not saying that other factors, including memory latency, aren't a factor. I'm just pointing out that an SMT patch that tidies up the thread priorities for the badly effected titles could go a long way to evening out the overall results.

 

[jaymc]

I take it Naples will not be handicapped with this affliction...

When is Zen+ due to be releaseed ?

Edit: Why don't they just start using faster memory, no need for big announcement.. just change it as soon as possible please !!

 

[thegentlewoman]

And what about this ?
1700 vs i7 7700
https://youtu.be/BXVIPo_qbc4

 

[TMTOWTSAC]

And what about this ?
1700 vs i7 7700
https://youtu.be/BXVIPo_qbc4

Looks the same as his last one, crank up the settings until the GPU bottlenecks. Not sure what he thinks he's saying about V-Sync and 144hz monitors. Once the GPU bottlenecks it can't going to go any faster no matter what CPU or monitor you're running it with. Which is the definition of a bottleneck.