Discussion AMD Ryzen MegaThread! FAQ and Resources

[TMTOWTSAC]

Well, personally, I still think the scheduler can be tweaked, but I do agree the improvements won't be magic bullets for performance and won't make Ryzen leapfrog any Intel CPU it hasn't already beaten. Still, a 2-3 percent change in favor is still good and might even be quantifiable in day to day stuff, since the scheduler is the main responsible for having a snappy/responsive OS. I do remember the first days of the P4+HT, when the experience in the desktop was improved a lot after the schedulers were able to handle the extra thread correctly. Well, maybe nowadays it won't be as pronounced, but still.

I'm still waiting on AnandTech's part 2. I wonder if they'll put some new stuff into the light-spot.

Cheers!

I think you'd be surprised what can be achieved through incremental updates. A while back when the FX-8000 series released, all the benchmarks had it about 10-15% behind the Sandy Bridge i5's. Next year, it was only about 5% behind. Fast forward two years, it's 5-10% ahead in the same game against the same SB i5. One of the major YT reviewers dug up that info.

I'd like to see the review pt 2 as well.

Are you sure it was in the same game? I saw something like that on Adoredtv, but it took 4 years, and he was using aggregate benchmarks from Computerbase that changed games (and OS's and GPU's etc) every time.

 

[Gon Freecss]

Well, personally, I still think the scheduler can be tweaked, but I do agree the improvements won't be magic bullets for performance and won't make Ryzen leapfrog any Intel CPU it hasn't already beaten. Still, a 2-3 percent change in favor is still good and might even be quantifiable in day to day stuff, since the scheduler is the main responsible for having a snappy/responsive OS. I do remember the first days of the P4+HT, when the experience in the desktop was improved a lot after the schedulers were able to handle the extra thread correctly. Well, maybe nowadays it won't be as pronounced, but still.

I'm still waiting on AnandTech's part 2. I wonder if they'll put some new stuff into the light-spot.

Cheers!

I think you'd be surprised what can be achieved through incremental updates. A while back when the FX-8000 series released, all the benchmarks had it about 10-15% behind the Sandy Bridge i5's. Next year, it was only about 5% behind. Fast forward two years, it's 5-10% ahead in the same game against the same SB i5. One of the major YT reviewers dug up that info.

I'd like to see the review pt 2 as well.

Not really

 

[cdrkf]

Quick question: Considering benchmarks - Out of all R7's, best bang for the buck as a workstation (3ds Max, Vray), with secondary role in gaming and streaming, is the R7 1700 with aftermarket cooler? Am I wrong?

I'd actually suggest sticking with the stock cooler supplied with the 1700 as it's a 'Wraith Spire' which is actually pretty capable (you can overclock the 1700 to the 3.8 - 3.9ghz on all cores range using the Spire without issue).

Link to a youtube review of the cooler:
https://www.youtube.com/watch?v=LzIIIoSveU0

Essentially AMD are bundling fairly high performance coolers with their cpu's these days so unless you are trying to push extreme overclocks the stock cooler is ample.

 

[jdwii]

I'm extremmly picky about noise ha ha so i'd still go after market if you want a real quiet system but as cdrkf said its not needed.

 

[-Fran-]

I'm extremmly picky about noise ha ha so i'd still go after market if you want a real quiet system but as cdrkf said its not needed.

I second this. The style/design AMD chose for their Wraith coolers, I don't think it's the best sound-wise.

Cheers!

 

[jaymc]

The first Ryzen patch (Dota 2)...
https://www.reddit.com/r/Amd/comments/60p19s/dota_2_first_ryzen_game_patch/

Also check this out rumour going around that MS released an update to Windows Schedular:
https://www.reddit.com/r/Amd/comments/60g5cj/windows_10_update_on_16_mar_have_partially_solved/

 

[Redneck5439]

Quick question: Considering benchmarks - Out of all R7's, best bang for the buck as a workstation (3ds Max, Vray), with secondary role in gaming and streaming, is the R7 1700 with aftermarket cooler? Am I wrong?

If you are truly using your rig as a workstation first computer with a secondary role of gaming then you are going to want the eight cores and 16 threads of the R7, so there isn't a compelling reason to wait for benchmarks of the 6 core 12 thread R5. With that said, which R7 comes down to one simple question - are you comfortable with overclocking? and are you going to overclock? If the answer to both is yes then the best value right now is the R7 1700. The R7 1700 in all testing at this stage can overclock to 3.9 to 4.0Ghz just like its bigger brother the R7 1800X. If however you are not going to overclock then the R7 1800X will outperform the R7 1700, however the R7 1800 X costs $500.

At this point in time, and at least until the process matures, the R7 1700 and R7 1800X both overclock to the same 3.9 - 4.0Ghz range, making the R7 1700 the best bang for your buck for overclockers.

 

[dgothi]

I am using Ryzen 1700 since about three weeks. I notice windows 10 running is not always smooth or good performance. I assume Ryzen CPU is too new... Maybe Windows 10 is not ready for AMD's new Ryzen and wait for microsoft provide patch for Ryzen? Ryzen's new SMT is issue with windows 10.

 

[juanrga]

I am using Ryzen 1700 since about three weeks. I notice windows 10 running is not always smooth or good performance. I assume Ryzen CPU is too new... Maybe Windows 10 is not ready for AMD's new Ryzen and wait for microsoft provide patch for Ryzen? Ryzen's new SMT is issue with windows 10.

https://community.amd.com/community/gaming/blog/2017/03/13/amd-ryzen-community-update?sf62107357=1

 

[juanrga]

Recent info about the new X399 platform coming from CanardPC. Clocks in second gen engineering sample are 3.1GHz base and 3.6GHz turbo. But those engineering samples are using B1-grade silicon; therefore don't expect improvements from better silicon. The marketing TDP is 180W. The real TDP is expected to be higher than 200W.

 

[salgado18]

Recent info about the new X399 platform coming from CanardPC. Clocks in second gen engineering sample are 3.1GHz base and 3.6GHz turbo. But those engineering samples are using B1-grade silicon; therefore don't expect improvements from better silicon. The marketing TDP is 180W. The real TDP is expected to be higher than 200W.

I missed something: is this the 16c/32t Zen? If so, I think the clocks are similar to the R7 1700, but TDP is way higher than double (65W x 2 = 130W). Or is it 32c/64t?

 

[Embra]

Recent info about the new X399 platform coming from CanardPC. Clocks in second gen engineering sample are 3.1GHz base and 3.6GHz turbo. But those engineering samples are using B1-grade silicon; therefore don't expect improvements from better silicon. The marketing TDP is 180W. The real TDP is expected to be higher than 200W.

I missed something: is this the 16c/32t Zen? If so, I think the clocks are similar to the R7 1700, but TDP is way higher than double (65W x 2 = 130W). Or is it 32c/64t?

Those are very good clocks if indeed 16c/32t.

 

[jaymc]

Dota 2 before and after Ryzen patch:

https://hardforum.com/threads/dota-2-update-increases-performance-for-ryzen.1927957/

 

[cdrkf]

Dota 2 before and after Ryzen patch:

https://hardforum.com/threads/dota-2-update-increases-performance-for-ryzen.1927957/

That looks quite a nice boost- 30fps, although lets not forget Dota 2 is pretty easy to run so it's probably more important to look at the % difference. This does at least show that at least in some circumstances there are things that can be done to better utilize ryzen on the developers end (I'd be curious to know what they did precisely). My guess is this type of optimisation is likely to benefit Ryzen in situations where it's performance doesn't make sense- e.g. games like AOTS that are shown to scale well on higher core count Intel parts but don't show scaling on Ryzen, or the few cases where Ryzen shows a drop in fps with SMT on but Intel either stays the same or gains performance thanks to SMT.

Recent info about the new X399 platform coming from CanardPC. Clocks in second gen engineering sample are 3.1GHz base and 3.6GHz turbo. But those engineering samples are using B1-grade silicon; therefore don't expect improvements from better silicon. The marketing TDP is 180W. The real TDP is expected to be higher than 200W.

Does this mean we think the 16c / 32t Ryzen 'extreme' platform is real?! I must admit I thought this was just wishful thinking but I'd be happy to be proven wrong! As for the TDP, to push those speeds on that many cores looks pretty good even at around 220W (it is double an R7 after all)- I'm guessing there will be a lower speed version with a much lower TDP (down to 130 - 140W range at say 2.5ghz) that's more in line with the server parts for those who don't want to deal with the higher heat dissipation.

I am using Ryzen 1700 since about three weeks. I notice windows 10 running is not always smooth or good performance. I assume Ryzen CPU is too new... Maybe Windows 10 is not ready for AMD's new Ryzen and wait for microsoft provide patch for Ryzen? Ryzen's new SMT is issue with windows 10.

That's strange, Windows 10 runs smoothly on almost all hardware. Can you be a bit more specific about A: what is your overall specification- if you are running on an old HDD for example that would hold back performance? B: What power profile are you using- there is a known issue with Ryzen that currently Windows 10 tends to park cores when it shouldn't resulting in a delay waking them back up- switching from 'balanced' to 'high performance' stops this behaviour (and there is an update due to Windows to sort that out)... Final thing, what exactly is 'not always smooth or good performance'? I would have thought the OS itself would always be smooth on such a powerful cpu, heck my old fx 8320 runs Windows 10 very nicely, is it a specific application that is causing you issues?

As for the SMT issue, that has been shown to be working as intended (essentially Windows just treats it the same as it does Intel's Hyper-threading, the only possible issue is Windows might move threads between the two CCX core blocks when ideally it wouldn't but that is a fairly limited issue that mainly affects games).

Final thought: I saw this video (there's also another by Linus which tends to agree with this)- a 'simulated' Ryzen 5 benchmark (achieved by disabling cores on an R7):
https://www.youtube.com/watch?v=O5cqOtWz5sU

This is exactly what I was hoping for- for gaming the cut down hex core and quad core parts both exhibit almost the exact same performance profile as the more expensive R7 at a lower price. No they aren't faster (only a higher clocked single CCX Ryzen part would theoretically be quicker for gaming), however given the R7 isn't actually a *bad* performer (the benchmarks on average place it neck and neck with the i7 4790k and just ahead of the i5 7600k which are both regarded as powerful gaming cpu's after all) the issue is just the R7 is a bit expensive if looking purely at gaming. The R5 parts bring this cost down to a level that makes sense for the performance on offer imo.

 

[truegenius]

Nope 1151 said :
Is anyone willing to benchmark a X6 Phenom against a 1600?

i have 1090t, if you have or can simulate 1600 then bring it on :-8 ( though i only have capped mobile internet thus have limited benchmarks software )
i will suggest to create a thread having list of benches and settings so that others can join in too

 

[dgothi]

I am using Ryzen 1700 since about three weeks. I notice windows 10 running is not always smooth or good performance. I assume Ryzen CPU is too new... Maybe Windows 10 is not ready for AMD's new Ryzen and wait for microsoft provide patch for Ryzen? Ryzen's new SMT is issue with windows 10.

https://community.amd.com/community/gaming/blog/2017/03/13/amd-ryzen-community-update?sf62107357=1

"Power Plans
Users may have heard that AMD recommends the High Performance power plan within Windows® 10 for the best performance on Ryzen, and indeed we do. We recommend this plan for two key reasons: "

Yeah,I just set to High Performance power plan, it is running much better. Problem solved. Thank you for share this link with us. Thumb up! I am not surprise they will work on patch for windows 10.

-DG

 

[juanrga]

Recent info about the new X399 platform coming from CanardPC. Clocks in second gen engineering sample are 3.1GHz base and 3.6GHz turbo. But those engineering samples are using B1-grade silicon; therefore don't expect improvements from better silicon. The marketing TDP is 180W. The real TDP is expected to be higher than 200W.

I missed something: is this the 16c/32t Zen? If so, I think the clocks are similar to the R7 1700, but TDP is way higher than double (65W x 2 = 130W). Or is it 32c/64t?

Yes this is the 16C/32T Zen. The disparity in TDPs is because the 1700 has a real TDP of 90W (as confirmed by several reviews and own CanardPC measurements). Two dies are 180W and extra 20W from the package give the ">200W" that claims CanardPC.

Does this mean we think the 16c / 32t Ryzen 'extreme' platform is real?! I must admit I thought this was just wishful thinking but I'd be happy to be proven wrong! As for the TDP, to push those speeds on that many cores looks pretty good even at around 220W (it is double an R7 after all)- I'm guessing there will be a lower speed version with a much lower TDP (down to 130 - 140W range at say 2.5ghz) that's more in line with the server parts for those who don't want to deal with the higher heat dissipation.

Yes it is real. CanardPC confirmed it time ago when they got their hands on the first gen engineering sample, and now got their hands on a second generation engineering sample.

That is the top SKU. I would expect 12 core models as well, from combining a pair of six-core dies.

 

[Rookie_MIB]

Just a quick note on some things with my 1700/RAM/motherboard. There still appear to be some issues with getting the RAM to run at top speed.

My core:
Ryzen 7 1700
8GB DDR4 Corsair Vengeance LPX (2x4GB) 3000mhz (capable).
Asus B350M-A motherboard.

With the original BIOS (0405) - no matter what I did, the RAM would only run 2133mhz. Anything higher would fail to POST. Updated to the new BIOS (0502), tried running 2400/2633/2966 - no good. So - I set DOCP to enabled which jumped it to 2933mhz and then DOWNCLOCKED the speed manually to 2400mhz and - it works. Then... I ran out of time for this morning. I'm going to try and see if I can get the speed of the ram up another notch or two.

Quick geekbenching (geekbench 3). Fully stock, 24200. @2400mhz - 24,900 or about a 3% increase.

------------------------ edit --------------------

Got home and tried 2666mhz with DOCP enabled - and it locked. There's really something that needs ironing out in the memory systems. So - I reset to 2400mhz and went with the multiplier:

Stock: 3.4ghz. GB3: single/multi 3481 / 24274
OC1 : 3.4ghz GB3: single/multi 3617 / 24840
OC2 : 3.5ghz GB3: single/multi 3528 / 26848
OC3 : 3.6ghz GB3: single/multi 3602 / 27560
OC4 : 3.7ghz GB3: single/multi 3676 / 28250

No matter what I do though, I can't get that RAM any faster. 2400 mhz is the most that I can get out of it right now. Hopefully another BIOS update will allow some additional unlocking of some RAM speed out of it.

 

[Nope 1151]

Nope 1151 said :
Is anyone willing to benchmark a X6 Phenom against a 1600?

i have 1090t, if you have or can simulate 1600 then bring it on :-8 ( though i only have capped mobile internet thus have limited benchmarks software )
i will suggest to create a thread having list of benches and settings so that others can join in too

I am several years behind the curve in tech to speak for personal reasons. The only thing I have right now is a 8350 .
I do remember seeing Hardware Unboxed doing a simulated 6c/12t.

I can try to bench a [strike]housefire[/strike] 90xx serries... But I Iive where it is 90 Degrees F right now (38/39c) so I'd rather not burn my house down.

 

[aldaia]

Check this "Simulation" of Ryzen 5
http://www.techspot.com/review/1360-amd-ryzen-5-1600x-1500x-gaming/

It's quite interesting, a 1800x is used to emulate 1600x and 1500x by disabling cores. and comapred to i7 i5 and i3. All 6 overclocked: AMD to 4 GHz, intel to 4.8 GHz. All 6 are tested with 1080Ti, 1070 and 1060.

With GTX 1060, GPU is the botleneck, so all cpus behave similarly (we already knew that). So for budget gaming (GPU < $300) go with the cheapest CPU, or if parallel performance is important for other reasons, go with the more cores.

With 1080Ti, GPU is almost never the botleneck. There is a big diference in performance between the 3 intel's, however the diference between the 3 AMDs is rather small. That suggests most games are optimized for 8 threads at most. Seems like 4C 8T ryzen will be an interesting proposition for gaming.

Finally with GTX 1070 it's a mixed bag. There are some games where things get reversed vs 1080Ti and Ryzens beat the Intels, suggesting GPU is not the bottleneck, or at least not all the time.

If the real R5's behave like that, gaming performance will actually be very good in most real use cases (except for hardcore gamers with top GPU).

 

[juanrga]

Dota 2 before and after Ryzen patch:

https://hardforum.com/threads/dota-2-update-increases-performance-for-ryzen.1927957/

The OP claims the patch improved performance by 23% on RyZen. The i7 7700k was a huge 50--60% ahead the R7 1800x before the patch

http://www.phoronix.com/scan.php?page=article&item=nvidia-1080ti-ryzen&num=2

 

[Rookie_MIB]

Check this "Simulation" of Ryzen 5
http://www.techspot.com/review/1360-amd-ryzen-5-1600x-1500x-gaming/

It's quite interesting, a 1800x is used to emulate 1600x and 1500x by disabling cores. and comapred to i7 i5 and i3. All 6 overclocked: AMD to 4 GHz, intel to 4.8 GHz. All 6 are tested with 1080Ti, 1070 and 1060.

With GTX 1060, GPU is the botleneck, so all cpus behave similarly (we already knew that). So for budget gaming (GPU < $300) go with the cheapest CPU, or if parallel performance is important for other reasons, go with the more cores.

With 1080Ti, GPU is almost never the botleneck. There is a big diference in performance between the 3 intel's, however the diference between the 3 AMDs is rather small. That suggests most games are optimized for 8 threads at most. Seems like 4C 8T ryzen will be an interesting proposition for gaming.

Finally with GTX 1070 it's a mixed bag. There are some games where things get reversed vs 1080Ti and Ryzens beat the Intels, suggesting GPU is not the bottleneck, or at least not all the time.

If the real R5's behave like that, gaming performance will actually be very good in most real use cases (except for hardcore gamers with top GPU).

The thing is with this though, is that I wonder if they would put out a single CCX based Ryzen based CPU. Part of the problem that is nearly guaranteed is that there are some issues when trying to access data in the L3 cache on the other CCX, or migrating a thread across a CCX. It's slower. If all the hex and quad core parts are based on the same dual CCX design, with either one or two bad cores (and disabling the matching core on the other CCX) determining the quad or hex core status - they will all have the same issue with the slower fabric connecting the CCXs.

A single CCX design for the quad parts would eliminate that issue and I would think would provide a higher level of performance AND a lower TDP design.

Regardless - the thing that's going to make the difference in the quad/hex parts is flat out the price/performance ratio for gaming. The $200-250 Ryzen parts are nearly equivalent to the $350-400 Ryzen parts in many games, and close to the i7-7700k parts, while being MUCH cheaper. At that point you're at about 80-90% of the performance for 60% of the price.

 

[cdrkf]

Dota 2 before and after Ryzen patch:

https://hardforum.com/threads/dota-2-update-increases-performance-for-ryzen.1927957/

The OP claims the patch improved performance by 23% on RyZen. The i7 7700k was a huge 50--60% ahead the R7 1800x before the patch

http://www.phoronix.com/scan.php?page=article&item=nvidia-1080ti-ryzen&num=2

True, but lets be honest Dota 2 is a *very* light workload based on an ancient game engine that really only uses 1 core. It scales with clock speed, ram speed and ipc- all 3 areas where the 7700k is ahead. The point is though the frame rates we are getting are very high on all parts, and a 23% improvement with identical hardware is very good. Anyone who thinks that Ryzen patches will allow it to topple the 7700k in gaming are going to be disappointed, but it does show that improvements are there.

The one exception to that is I do think in game engines that *actually scale well to 8 cores* could see major boosts where Ryzen is behind where it probably should be (e.g. situations where a 6900k is ahead of a 7700k like AOTS but the R7s aren't). Those are the situations patches like this will potentially shine. That said there are only a handful of cases where that is applicable so it's not going to change the overall landscape all that much.

 

[aldaia]

The thing is with this though, is that I wonder if they would put out a single CCX based Ryzen based CPU. Part of the problem that is nearly guaranteed is that there are some issues when trying to access data in the L3 cache on the other CCX, or migrating a thread across a CCX. It's slower. If all the hex and quad core parts are based on the same dual CCX design, with either one or two bad cores (and disabling the matching core on the other CCX) determining the quad or hex core status - they will all have the same issue with the slower fabric connecting the CCXs.

A single CCX design for the quad parts would eliminate that issue and I would think would provide a higher level of performance AND a lower TDP design.

Definitely a single CCX design will be great, and cheaper to produce. If AMD has not produced one (yet) is because it's a small company and has not enough resources to design several dies. I think AMD has been very clever when designing the zeppelin die. The same die is used for PC's scaling from 4C8T to 8C16T and for servers scaling up to 32C64T. Also the CCX design is a clever idea, since they can easily design APU dies with a relatively small effort just by reusing parts and adding a bit of glue logic.

Regardless - the thing that's going to make the difference in the quad/hex parts is flat out the price/performance ratio for gaming. The $200-250 Ryzen parts are nearly equivalent to the $350-400 Ryzen parts in many games, and close to the i7-7700k parts, while being MUCH cheaper. At that point you're at about 80-90% of the performance for 60% of the price.

Agreed, the good thing about ryzen is not that they are a match for the intel counterparts, it's that the price/performance is really good.

 

[8350rocks]

Recent info about the new X399 platform coming from CanardPC. Clocks in second gen engineering sample are 3.1GHz base and 3.6GHz turbo. But those engineering samples are using B1-grade silicon; therefore don't expect improvements from better silicon. The marketing TDP is 180W. The real TDP is expected to be higher than 200W.

Looks like up to 16 cores and quad channel memory was accurate...