Hardware & Technical Ryzen Pricing looks promising ...
- Thread starter GuruNot
- Start date
Robert Maynard
Volunteer Moderator
An interesting report regarding overclocking of i7-7700Ks (or not, as the case may be....): http://wccftech.com/intel-i7-7700k-owners-flood-forums-with-overheating-complaints/
Are you using the new beta of Prime95 (29.2)? The current stable build (28.10) doesn't load Ryzen correctly and isn't a good stability test for them.
Intel has never sanctioned overclocking on any of their parts by default, even the unlocked parts. Indeed, it immediately voids the standard warranty to do so and only the purchase of a Performance Tuning Plan allows one to overclock while retaining any support at all.
Due to the type and thickness of TIM used in Intel's mainstream part, the only way to get substantial, 24/7 stable, OCs out of the higher-clocked samples is to delid them and replace said TIM with a thinner layer of higher thermal conductivity material. Doing this will immediately void any Intel warranty, including the Tuning Plan.
Anyway, no manufacturer condones running their parts significantly out of spec in any parameter.
Robert Maynard
Volunteer Moderator
.... yet they sell their K versions at a premium....(as do AMD, of course)
Indeed, but touching that unlocked multiplier still voids your warranty.
On the Intel to OC or not to OC question, following is what I wrote elsewhere. (it originally referenced pcgamer, and not surprisingly it is spread around the usual tech sites now)
I think there is quite a bit of overclocking hype, that those who don't properly understand it see lots of jigahurts being reported and then cry when their sample can't match it. Personally I don't even come close to others since I need my systems to run Prime95 equivalent loads 24/7 with 100% stability and not go into meltdown. I don't do barely bench stable OCs outside of the rare occasion I actually am dabbling with competitive OC.
To bring it back to topic, I had another quick go at OC on my R7 1700 earlier today. Previously I determined it would boot at 3.7 GHz 1.2v (voltage randomly chosen to keep power low) and had stability tested it at 3.6 to allow for extra headroom. That was at 2133 ram clocks. What I was most interested in was if I can find any ram I had spare that would get above 2666. Alas, it was not to be the case but I did repeat the stability testing with ram at 2666, and all was fine. 1 hour realbench, 1 hour prime95 29.1 blend, and about 20 minutes of aida64 to fill in a little slack time. I don't find aida64 to be useful to prove stability, but only usually use it for the temperature history chart.
28.xx doesn't detect Ryzen, and uses old AMD code without AVX.
29.1 updates CPU detection method, and now recognises Ryzen and enables AVX2 on that (same code as for Intel). The benchmark window has been updated nicely so no more manual editing of files needed in most cases.
29.2 only exists for testing alternate code to determine what is the optimal configuration for various scenarios. Future versions based on this could have better throughput on modern systems, but worse on old systems.
If you have Ryzen and want to use Prime95, get 29.1 from http://www.mersenneforum.org/showthread.php?t=22141
Note that as Ryzen doesn't have strong AVX2 performance, Prime95 might not be the most stressful for the architecture. I found realbench to be better as a stability test in cases like this.
I think there is quite a bit of overclocking hype, that those who don't properly understand it see lots of jigahurts being reported and then cry when their sample can't match it. Personally I don't even come close to others since I need my systems to run Prime95 equivalent loads 24/7 with 100% stability and not go into meltdown. I don't do barely bench stable OCs outside of the rare occasion I actually am dabbling with competitive OC.
To bring it back to topic, I had another quick go at OC on my R7 1700 earlier today. Previously I determined it would boot at 3.7 GHz 1.2v (voltage randomly chosen to keep power low) and had stability tested it at 3.6 to allow for extra headroom. That was at 2133 ram clocks. What I was most interested in was if I can find any ram I had spare that would get above 2666. Alas, it was not to be the case but I did repeat the stability testing with ram at 2666, and all was fine. 1 hour realbench, 1 hour prime95 29.1 blend, and about 20 minutes of aida64 to fill in a little slack time. I don't find aida64 to be useful to prove stability, but only usually use it for the temperature history chart.
28.xx doesn't detect Ryzen, and uses old AMD code without AVX.
29.1 updates CPU detection method, and now recognises Ryzen and enables AVX2 on that (same code as for Intel). The benchmark window has been updated nicely so no more manual editing of files needed in most cases.
29.2 only exists for testing alternate code to determine what is the optimal configuration for various scenarios. Future versions based on this could have better throughput on modern systems, but worse on old systems.
If you have Ryzen and want to use Prime95, get 29.1 from http://www.mersenneforum.org/showthread.php?t=22141
Note that as Ryzen doesn't have strong AVX2 performance, Prime95 might not be the most stressful for the architecture. I found realbench to be better as a stability test in cases like this.
Thanks for the advice! I'm leaning towards the 1600. It's a bit cheaper than the i5-7600K, comes with a decent cooler, performs better in productivity, performs comparably in gaming, etc.
With my usage, assuring 100% stability 24/7 is not absolutely essential, as there is very little I do at home that could be considered mission critical. That said, I do like a stable system. I will poke around to get an idea of what others are seeing as far as overclocks, and see what the extreme edge cases being reported are, and then shoot for something approaching those. As for stability goes, I will throw the normal benchmarks, stress tests at it for a while, watch for crashes, check for errors, and after a couple of hours of messing around, if all is good, I call it a day.
I got my 3.8 on the system pictured above with a vcore of 1.18. I guess I am one of the lucky one whose memory is running a 3200 out of the box simply using the XMP. All verified by HWinfo, and CPU-z. Some real world testing...
My system is a Ryzen 1700@3.8 ghz, GTX 1080ti stock, 16gb g.skill trident z 3200 cl16, 3440x1440 21:9 60hz monitor, 20c ambient temps...
Elite Dangerous everything on ultra for 6 hours sat pegged at 60fps regardless of what I was doing and CPU temps hit 25c, GPU temps hit 24c. I did end up lowering quality to high settings because I was seeing some of the same things others have reported here with using the ultra presets namely stuttering while framerate stays constant. CPU load sat around 25% and GPU load around 75%. This is with Voice Attack, EDDI, Captains Log, and EDDiscovery running as well.
Planet Coaster, arguably a more taxing game...I loaded up some of the parks that have been used for performance testing and benchmarking, and let them run. The 10k visitor parks were getting between 20-30 frames per second running at fast speed and as high as 35 FPS when paused. My temps hit 32 on the CPU and 35 on the GPU with my fans staying at about 40%, so pretty much silent. CPU load bounced around a bit from 50-80%, and the GPU load sat up near 100%.
All that with no crashes, no errors, except for the usual crashes in ED. So I'm calling it good.
Last edited:
Robert Maynard
Volunteer Moderator
Things may be about to get more interesting: http://wccftech.com/amd-computex-event-radeon-rx-vega-x399-hedt/
This competition with Intel / Nvidia is excellent for consumers
Robert Maynard
Volunteer Moderator
More details of upcoming Ryzen 9: http://wccftech.com/amd-ryzen-9-lineup-threadripper/
Good old WCCFTech, taking a 1000 words to say the same thing as the table slightly lower down on the same page.
They don't address some things though. Are the new CPUs a new die or multi-chip in one package? Given what we currently know about CCX so far in that each module needs to be balanced, I wonder how those 10 cores would be split. Either the balance isn't really needed, or this is some new structure. I don't think it the latter as CCX was designed to be scaleable by having more of them.
Their comments on TDP falls exactly into AMDs marketing plan. Intel TDP tends to be higher as the highest TDP scenarios involve AVX2, and Zen was not designed to be high performance in that area. In other workloads they're pretty much comparable. With AVX512 being introduced Intel could once again be 4x the IPC compared to AMD in that area.
Also I'd take the 3200 ram claim with a pinch of salt. It is different between officially supporting it, and offering it and you're on your own luck. Pretty much I don't see it being different than the situation now. You have a lower official supported speed, but you are effectively ram interface overclocking if you want to go higher. If AMD really do offer 3200 as an officially supported speed (as opposed to OC speed) I think they would need a ram certification scheme to ensure it works. Right now on both Intel and AMD, even if the mobo claims to support it, running at 3200 is far from guaranteed.
Still, it will be a really interesting time once R9 and i9 are announced and we have another wave of benching to look forward to.
On a slight tangent, I built my 2nd Ryzen system on the weekend. Slightly lower spec than 1st, with a R5 1600 currently overclocked to 3.6 GHz paired with 2666 ram. I'm running the bundled Wraith Spire cooler while I wait for my free Noctua AM4 mount kit to arrive and allow me to use something nicer. Both the Wraith Spire coolers, one each from 1700 and 1600, have noise problems. The one from 1700 clicks when it rotates, with nothing visible outside so it must be internal. I suspect I got a bad sample and it isn't worth returning since I replaced it anyway. The one from 1600 is fine under normal use, but under sustained intense loading (even when not OC), if fan speed goes above 2000 rpm a whine starts which is really annoying. This isn't just air noise but something else going on. Not likely to affect casual users to be fair. From a performance point of view, I'd still look to get decent after market cooler for anything other than very tight budget builds.
They don't address some things though. Are the new CPUs a new die or multi-chip in one package? Given what we currently know about CCX so far in that each module needs to be balanced, I wonder how those 10 cores would be split. Either the balance isn't really needed, or this is some new structure. I don't think it the latter as CCX was designed to be scaleable by having more of them.
Their comments on TDP falls exactly into AMDs marketing plan. Intel TDP tends to be higher as the highest TDP scenarios involve AVX2, and Zen was not designed to be high performance in that area. In other workloads they're pretty much comparable. With AVX512 being introduced Intel could once again be 4x the IPC compared to AMD in that area.
Also I'd take the 3200 ram claim with a pinch of salt. It is different between officially supporting it, and offering it and you're on your own luck. Pretty much I don't see it being different than the situation now. You have a lower official supported speed, but you are effectively ram interface overclocking if you want to go higher. If AMD really do offer 3200 as an officially supported speed (as opposed to OC speed) I think they would need a ram certification scheme to ensure it works. Right now on both Intel and AMD, even if the mobo claims to support it, running at 3200 is far from guaranteed.
Still, it will be a really interesting time once R9 and i9 are announced and we have another wave of benching to look forward to.
On a slight tangent, I built my 2nd Ryzen system on the weekend. Slightly lower spec than 1st, with a R5 1600 currently overclocked to 3.6 GHz paired with 2666 ram. I'm running the bundled Wraith Spire cooler while I wait for my free Noctua AM4 mount kit to arrive and allow me to use something nicer. Both the Wraith Spire coolers, one each from 1700 and 1600, have noise problems. The one from 1700 clicks when it rotates, with nothing visible outside so it must be internal. I suspect I got a bad sample and it isn't worth returning since I replaced it anyway. The one from 1600 is fine under normal use, but under sustained intense loading (even when not OC), if fan speed goes above 2000 rpm a whine starts which is really annoying. This isn't just air noise but something else going on. Not likely to affect casual users to be fair. From a performance point of view, I'd still look to get decent after market cooler for anything other than very tight budget builds.
It seems that market share is evolving for AMD with Ryzen as flagship product
News of AMDs market gain is sloshing around the internet, but if that's based on PassMark data which most articles seem to be referencing, take it with a pinch of salt. I don't know if anyone outside AMD knows the actual sales numbers yet. It is almost certain AMD will be growing now, but how much is much less clear.
Yes, I agree. Only AMD can communicate on this subject