Ryzen
- Thread starter time
- Start date
Chewy509
Wotty wot wot.
Looks around the corner... nope, nothing there... time to move on...
Chewy509
Wotty wot wot.
Waiting for actual release with good testing...
sedrosken
Florida Man
I think we're finally in an era where both companies will continue trading blows -- hopefully this continues to lower prices and drive performance and efficiency improvements. I really don't want to end up in another Sandy Bridge era with either company having products head-and-shoulders above the other where they effectively get to set the prices.
I'm not particularly thrilled about Intel's power consumption problems, but I'm also not thrilled about AM5 continuing the use of ASMedia USB hardware -- I've been relatively lucky though I have had some minor issues, but several of my friends have had to RMA AM4 boards because of faulty USB hardware. I'm thinking my next major platform upgrade -- due probably within a couple years, maybe a little more maybe less -- will likely be Intel, hopefully the Alder Lake scheduler from Windows 11 is backported to 10 sometime before that though. It's either that or I'm going to need to keep my old machine around for the games Linux won't run, because I'm not going to be running 11.
But I digress. I concur with Chewy here -- we need actual figures before I can say anything concrete either way. "Leaks" and "reports" can say whatever they damn well please and turn out to be wrong. I'm optimistic for both companies right now because both appear to be back at their A-game. Personally I'm more interested in seeing lower-end Intel GPUs come into use for servers for media transcoding, stuff like what Plex needs.
I'm not particularly thrilled about Intel's power consumption problems, but I'm also not thrilled about AM5 continuing the use of ASMedia USB hardware -- I've been relatively lucky though I have had some minor issues, but several of my friends have had to RMA AM4 boards because of faulty USB hardware. I'm thinking my next major platform upgrade -- due probably within a couple years, maybe a little more maybe less -- will likely be Intel, hopefully the Alder Lake scheduler from Windows 11 is backported to 10 sometime before that though. It's either that or I'm going to need to keep my old machine around for the games Linux won't run, because I'm not going to be running 11.
But I digress. I concur with Chewy here -- we need actual figures before I can say anything concrete either way. "Leaks" and "reports" can say whatever they damn well please and turn out to be wrong. I'm optimistic for both companies right now because both appear to be back at their A-game. Personally I'm more interested in seeing lower-end Intel GPUs come into use for servers for media transcoding, stuff like what Plex needs.
The 5950x at least made the 570x system with that sluggish CPU 3950x more viable. I'd expect at least similar improvement, but hope for more.
I'd rather go with Intel nowadays, if not for the horrors of Win 11. If the performance per core of i9 13th gen is really good I wonder if those weakling cores can be disabled and just the 8 main cores would be fine in Win 10.
I'd rather go with Intel nowadays, if not for the horrors of Win 11. If the performance per core of i9 13th gen is really good I wonder if those weakling cores can be disabled and just the 8 main cores would be fine in Win 10.
sdbardwick
Storage is cool
I expect Ryzen 4 will be my next cpu, but I will wait for a large cache 3D version. I currently run a 5800X3D, and it is snappier (and more thermally efficient) than the i9-11900K in my secondary system. I will wait for testing results as well, but I expect AMD won't pull another Bulldozer and screw up their AVX-512 implementation in R4, unlike FMA in Bulldozer.
sedrosken
Florida Man
The 3D cache is what enabled the 5800X3D to maintain a lead in games over even Alder Lake as I recall. It doesn't help overclocking, but arguably I'd say the era of overclocking is over anyway as these parts come from the factory so close to their voltage, wattage and thermal limits as it is. If you can't even hit your full turbo speed across all cores as it is, what use is it trying to crank an extra 2-300MHz out of a chip? We're not in the Sandy Bridge era anymore where Intel was leaving a whole gigahertz or more on the table.
As far as productive use for the 3D cache, I think it led to healthy bumps across the board in other benchmarks, but as for real-world performance I don't think that was the focus.
Right now AVX-512 isn't used by much of anything to my knowledge. However that stands to change as AMD implements it making it more widespread and people upgrade to both newer AMD and Intel systems supporting it. Kinda the same situation as AVX2 back when Haswell launched -- nothing used it then, but it turned out to be important anyway and it's a major reason why Zen1 and Zen+ aren't viable machines for gaming anymore, their AVX2 implementation was half-assed and it really shows in things like emulation. My performance in Xenia went up something like 75-100% going from my 1600AF to my 5500, on the same socket. Some of this was down to general architecture improvements in Zen3, but a lot of it was the AVX2 implementation being fixed going from Zen1/+ to Zen2.
As far as productive use for the 3D cache, I think it led to healthy bumps across the board in other benchmarks, but as for real-world performance I don't think that was the focus.
Right now AVX-512 isn't used by much of anything to my knowledge. However that stands to change as AMD implements it making it more widespread and people upgrade to both newer AMD and Intel systems supporting it. Kinda the same situation as AVX2 back when Haswell launched -- nothing used it then, but it turned out to be important anyway and it's a major reason why Zen1 and Zen+ aren't viable machines for gaming anymore, their AVX2 implementation was half-assed and it really shows in things like emulation. My performance in Xenia went up something like 75-100% going from my 1600AF to my 5500, on the same socket. Some of this was down to general architecture improvements in Zen3, but a lot of it was the AVX2 implementation being fixed going from Zen1/+ to Zen2.
Last edited:
The forum did something to my post. 
From what I found online the 3D stuff makes the core speeds quite a bit slower. Maybe the Ryzen 4 will have a more coordinated cache system and the extra caches are not as necessary? Why do the games need so much cache anyways; is there no opportunity to improve the software?
From what I found online the 3D stuff makes the core speeds quite a bit slower. Maybe the Ryzen 4 will have a more coordinated cache system and the extra caches are not as necessary? Why do the games need so much cache anyways; is there no opportunity to improve the software?
sedrosken
Florida Man
Yes and no: the tighter tolerances on the cache means that the core simply can't turbo as high. This is also what makes it an especially poor overclocker.
The extra caches were never "necessary." The regular 5800X did just fine in games, but Alder Lake edged it out. The 5800X3D simply does better because I'm willing to bet oftentimes the entire game engine's codebase can fit into cache. The adage with cache was always that more cache was good, faster cache was better -- in lieu of being able to make it appreciably faster they elected to just brute-force it and accomplish the same performance gains by loading up on it. I believe the 5800X3D was made as a PR part to show they had the faster part for games and nothing else, besides being a test-drive of the 3D cache tech as a whole.
I do believe all of Zen4 is set to use 3D cache, however apparently yields have improved to the point that according to a couple sources I've seen -- not that I really believe them until I see the actual specs delivered on release -- the 7950X is set to turbo to 5.85GHz out of the box and hit 6GHz with PBO. I don't even want to know what kind of cooling solution that's going to require much less the power bill for it, but that bodes well for the performance figures for midrange parts too. Again, though, purely because they're still going to be using unreliable Asmedia USB hardware, my next build will be Intel.
The extra caches were never "necessary." The regular 5800X did just fine in games, but Alder Lake edged it out. The 5800X3D simply does better because I'm willing to bet oftentimes the entire game engine's codebase can fit into cache.
The adage with cache was always that more cache was good, faster cache was better -- in lieu of being able to make it appreciably faster they elected to just brute-force it and accomplish the same performance gains by loading up on it. I believe the 5800X3D was made as a PR part to show they had the faster part for games and nothing else, besides being a test-drive of the 3D cache tech as a whole.I do believe all of Zen4 is set to use 3D cache, however apparently yields have improved to the point that according to a couple sources I've seen -- not that I really believe them until I see the actual specs delivered on release -- the 7950X is set to turbo to 5.85GHz out of the box and hit 6GHz with PBO. I don't even want to know what kind of cooling solution that's going to require much less the power bill for it, but that bodes well for the performance figures for midrange parts too. Again, though, purely because they're still going to be using unreliable Asmedia USB hardware, my next build will be Intel.
There's always opportunity to optimize the software and gaming engine but it's very costly, time consuming, and difficult. It's easier for a dev studio to unload the responsibility onto a gamers hardware and let them deal with the issues.
Supposedly up to 29% of a core improvement? 13% is IPC and the rest is speed. Then the all-core performance is even better since the package can handle higher power and is also more efficient. We'll have to the see how this tests at various sites. I'm concerned about heat and don't want to use a radiator with fluids. Will the Nocturnal cooler with two fans be enough?
If you're asking about the new Ryzen 9 7950X and the Noctua NH-D15, it might be tough to manage the thermals. As it is now, I find my NH-D15 can manage my 3950X at 105W TDP when maxing out the CPU but it feels like it's at the edge of managing everything well. If the 7950X is up in the 170W range on a new untested CPU lid, I don't know.
I have the 3950x on the D14 and the 5950x on the D15 and they seem fine to me, but I have that same concern.
There is an interesting read here.
noctua.at
There is an interesting read here.
Noctua’s Standardised Performance Rating (NSPR) and compatibility classification for CPU coolers
Designed in Austria, Noctua's premium cooling components are renowned for their superb quietness, exceptional performance and thoroughgoing quality.
noctua.at
sedrosken
Florida Man
I ran my 3700X with the NH-D15 and it was definitely just on the ragged edge of OK thermals-wise -- my ambient is probably much higher than yours at 78F/25.5C, though, so your mileage may vary. My 5500 runs fine on a bequiet! Pure Rock Slim, which is surprising because 5nm is supposed to be very thermally dense and difficult to cool. I'm definitely going to say that you'll want a watercooling setup with a nice quiet pump and a 240mm rad, possibly larger for the 7950X. I don't know how they're expecting you to properly cool whatever comes this generation for Threadrippers.
That is only a 65W CPU, so I have no idea why a massive heatsink with two fans would not cool it sufficiently unless there are mechanical heat transfer issues or the CPU is being operated improperly. The manufacturer (AMD) sells the CPU with a smaller cooling solution (Prisma Wraith).
The 7950x uses about 3x the power, so that will be challenging. Are there any quiet and more efficient coolers than the NH-D15? I doubt there could be any much larger and still fit in the computer case without blocking everything.
The 7950x uses about 3x the power, so that will be challenging. Are there any quiet and more efficient coolers than the NH-D15? I doubt there could be any much larger and still fit in the computer case without blocking everything.
sdbardwick
Storage is cool
If your 3700X and D15 give bad temps, then there is/was something wrong with your setup.
I run a D15 (single fan) on a Xeon E5-2699v3 (145W TDP) 24/7 with a core package power draw of 137.75W and have core temps under 50C and package temp at 50C. Box runs Prime95 24/7; no bad results yet.
One of the other boxes I am responsible for runs a 3800X with the OEM Wraith cooler also 24/7 Prime95 (when not running some specific programs like QuickBooks) and it has been 100% stable for 19 months.
On the Intel side, a U12S Redux isn't sufficient to keep my i9-11900 from thermal throttling. Not terribly surprising, as that cpu often exceeds 200W at load (125W TDP), but the Redux will work fine on a E5-2699v3, (fine defined as error-free without thermal throttling) as that CPU generally adheres to the 145W TDP.
Redux also seems fine on 5800X3D, but I don't have any hard temp data yet. Runs Prime95 24/7 without error since 5/12/2022.
Quick 5800X3D update: Runs fine on Redux; 67W PPT under Prime95 with less than 62C temps on cores and processor package. I'm probably leaving some performance on the table; haven't done any processor tuning, not even PBO.
Last edited:
sedrosken
Florida Man
Maybe I was just surprised at how high Zen2 and up idle -- 50C felt like too high for an idle temperature of anything to me. It never once thermal throttled or went past 75C under load, but I don't consider that to be particularly amazing either -- I know these things are considered safe up to around 100C but it just feels too high for me. Also remember my ambient temperatures -- if I run my AC any lower than 78F, it's just on 24/7 and my electric bill skyrockets.
Even that idle temp seems high. Too many factors to guess like quality of thermal paste and coverage, case size and ventilation, number of case fans and their speeds/cfm, number of other HDDs, GPU, etc.
I try to keep my dh-15 fans running slow so I can hear them ~700rpm which results in higher idle temps but even so, mine are usually lower than yours. I also run my x570 chip fan low causing it to run hot and my rtx 3080 contributes a bunch of heat.
I try to keep my dh-15 fans running slow so I can hear them ~700rpm which results in higher idle temps but even so, mine are usually lower than yours. I also run my x570 chip fan low causing it to run hot and my rtx 3080 contributes a bunch of heat.
sedrosken
Florida Man
Well, considering I don't have it anymore, I can't really supply much more information outside the fact that I used NT-H1 paste with a Phanteks Enthoo Pro (solid side panels version). I have the 200mm fan in front as intake, the 140mm fan in the rear and a 120mm fan up top as exhaust, and at the time I ran an RX480 GPU. I had tuned my fans such in my UEFI settings that they ramped up to 100% once the temperatures hit 75, and that might be what kept it below 80, I'm not entirely sure.
I traded the 3700X and Noctua cooler to a friend for his old 1070 back when GPU prices were still insane enough that that was a good deal and run that with the 5500 now -- again, the 5500 idles a little high (mid 40s) but never gets past 75 or so at max whack and this is PBO'd to a couple hundred MHz above stock max turbo. This is with a vastly inferior cooler, even. The only common denominator I can think of is the motherboard -- the same B450M-Pro4 I purchased with my 1600AF back in 2020. Maybe something's up with this thing where it reports temperatures wrong...? But then, the CPU temp sensor would be on the CPU, so I don't know.
I traded the 3700X and Noctua cooler to a friend for his old 1070 back when GPU prices were still insane enough that that was a good deal and run that with the 5500 now -- again, the 5500 idles a little high (mid 40s) but never gets past 75 or so at max whack and this is PBO'd to a couple hundred MHz above stock max turbo. This is with a vastly inferior cooler, even. The only common denominator I can think of is the motherboard -- the same B450M-Pro4 I purchased with my 1600AF back in 2020. Maybe something's up with this thing where it reports temperatures wrong...? But then, the CPU temp sensor would be on the CPU, so I don't know.
I don't know either. Maybe just bad silicon luck with that 3700X or it really was just terrible at efficiency. I think my buddy has either the 3700X or 3800X with the same NH-D15 on the same ASRock taichi x570. I'll ping him and see what kind of idle temps he's getting just out of curiosity.
Wccftech: AMD Ryzen 7000 Desktop CPU Delidded: Gold-Plated Zen 4 CCDs & Octopus-Styled IHS For Wider Cooler Compatibility.
wccftech.com
AMD Ryzen 7000 Desktop CPU Delidded: Gold-Plated Zen 4 CCDs & Octopus-Styled IHS For Wider Cooler Compatibility
Steve over at Gamers Nexus recently got the opportunity to go hands-on with a delidded AMD Ryzen 7000 Desktop CPU.
wccftech.com
Mercutio
Fatwah on Western Digital
I have a 3700X that has an actual bad physical core, something I've never seen before. I've been messaging AMD back and forth about it for a little while now. It's the second one of that CPU in particular I've had go bad somehow, where I've historically found bad CPUs to be a once every five years sort of thing.
Newtun
Storage is nice, especially if it doesn't rotate
Wow, that's a bit scary to me, since I've had my Ryzen 7 3700x BOINCing full-time, 24×7×8×2×100%, for almost 1½ years.
It still seems to have 16 threads running OK, no errors reported.
I'm not happy with the temps, though, occasionally getting to 85°-87°.
I'm using the "stock" Wraith Prism cooler, with the damn built-in neon light-show, but I think I installed it OK.
But then again, I'm an amateur old geezer.
It still seems to have 16 threads running OK, no errors reported.
I'm not happy with the temps, though, occasionally getting to 85°-87°.
I'm using the "stock" Wraith Prism cooler, with the damn built-in neon light-show, but I think I installed it OK.
But then again, I'm an amateur old geezer.
Mercutio
Fatwah on Western Digital
I have a spare PC, a Ryzen 5900 that shipped with one of those Wraith coolers, set up in my dining room. Main roommate and emergency backup couch roommate both appreciate having a big boy computer around. Most of my house has Hue color LEDs or other app controlled lighting of some sort or other, and that particular PC happens to be in a Cooler Master 600D chassis with transparent panel, a case I didn't even intend to buy except that it was the cheapest option the day I needed it. I never thought for even one second that I'd like an RGB light in a case but the girls really like playing with it at the same time as all the other lights.
Mercutio
Fatwah on Western Digital
AMD is usually pretty good about having products in the channel after an announcement. Historically, at least. I'm cautiously interested, but I also know I have actual scads of spare DDR4 sitting around and I'm probably going to build systems around that before I start dipping in the DDR5 pool.
I'm kind of interested but also a bit cautious about their new release having the new thermal spreader, BGA packaging & AM5 socket, ddr5, pcie5 and whatever else I've missed. That's a bunch of new changes to get right in a first release.
Mercutio
Fatwah on Western Digital
Zen 4 looks pretty good compared to 12th gen Intel. Higher power than Zen 3 but every CPU has graphics now and the overall TDPs are still lower than what Intel is doing. I only glanced through productivity benchmarks but it looks fairly convincing as a generational improvement.
Single-threaded speed is about +25% which is pretty good. Multi-threaded is dependent on the application and power/cooling but +35% is not difficult to achieve. The 7950x may be something good if it runs Windows 10 and TPM can be decommissioned by March.
I saw adverts from the stores that they are in stock already.
I saw adverts from the stores that they are in stock already.
sedrosken
Florida Man
According to some sources the IHS on Zen4 is too thick and hurts temps. I don't know how true that is.
sedrosken
Florida Man
It's not just overclockers running into this. According to Gamer's Nexus, 95C is just the new load temp now for almost every use case that maxes out the chip. The IHS is so thick primarily to provide backward compatibility with most existing AM4 coolers. Direct-die cooling was always going to be superior, but it seems compromises were made where they maybe shouldn't have been -- that said, a boost clock well into the upper 5GHz range is nothing to be sneezed at either, so there are a variety of factors at play here obviously.I'm sure the engineers at AMD did that for a reason even if the overcaulkers don't like it.
Mercutio
Fatwah on Western Digital
My threadripper tops out at 68C and all of this makes me really excited for what happens when they put these improvements toward the hardware that has enough PCIe lanes to be useful. Seems like there might be both thermal headroom AND gains from the updated architecture.
Apparently AMD thinks everyone with a Zen 4 R9 needs to move to liquid cooling though. I know there are sealed units but every time I've messed with watercooling, it's been expensive and more trouble than it's worth. I am less thrilled about that part.
Apparently AMD thinks everyone with a Zen 4 R9 needs to move to liquid cooling though. I know there are sealed units but every time I've messed with watercooling, it's been expensive and more trouble than it's worth. I am less thrilled about that part.
I refuse to use fluids for cooling. I need a quiet computer since it is next to my area. I can live with the fans at moderate speeds on idle and then noisier under load if it works with the Nocturnal D15. The fluid coolers have to blow out the top which would probably be unbearable.
The whole build would have to be mostly new and that will be a hassle.
The weird thing is that the motherboards have so few PCIe slots. 3x PCIe and 4x M.2 is out of whack. Why don't we get some U.3 or even U.2 ports and not hog the board space? Am I supposed to RAID0 2 or 3 SSDs? Four indepent drives are a PITA.
The whole build would have to be mostly new and that will be a hassle.
The weird thing is that the motherboards have so few PCIe slots. 3x PCIe and 4x M.2 is out of whack. Why don't we get some U.3 or even U.2 ports and not hog the board space? Am I supposed to RAID0 2 or 3 SSDs? Four indepent drives are a PITA.
Mercutio
Fatwah on Western Digital
I think consumer desktop platforms are built on the assumption that everyone wants, at most, a discrete GPU or two and perhaps a NIC. Not everyone has Thunderbolt or USB 3.2 or wants all their expansion to be external. I think it's an actively poor choice for extra drives.
LM, Windows has a function called Libraries and also supports directory symlinks now, so it's a lot easier to find ways to allocate storage locations in a natural way across multiple physical drives. Neither of those things have been pushed as a widely used Windows feature though, but then, most people don't have eight drives in their desktop, either.
LM, Windows has a function called Libraries and also supports directory symlinks now, so it's a lot easier to find ways to allocate storage locations in a natural way across multiple physical drives. Neither of those things have been pushed as a widely used Windows feature though, but then, most people don't have eight drives in their desktop, either.
sedrosken
Florida Man
Those directory redirectors are why I have my data stored on my NAS and it's seamless. Though Windows does like to freak out a little when I don't have a network connection, so I'm assuming I might have just done something slightly wrong. It's a situation that happens so minimally often that I don't really mind when it does, especially now that I'm working on a project laptop I can daily instead of firing up my main desktop and sucking up 150W at idle just to browse the web. Since my main doesn't really perform any server functions anymore I can get away with only turning it on when I really need it.
Back to the topic at hand, yeah, consumer hardware makes those assumptions and usually they're right. It also, as a bonus, lets them budget-segment the hell out of professional hardware that provides the IO people who actually use it really need. U.2/U.3 is a standard that never really made it into the consumer space because at the sizes consumers use, m.2 was plenty sufficient.
On the subject of cooling, liquid cooling has come a long way since I tossed my MasterLiquid Lite 240 for the pump noise. You can get reasonably safe coolers that barely make any noise at all. The companies that make AIOs typically have warranty programs in place that I believe will replace any hardware you had that would have been killed by a leak, but these days I'm pretty sure they're using something that won't kill your hardware as a coolant anyway. I still personally will never choose to do it again, but if a platform all but requires it, it's no longer a reason for me to not consider that platform anymore.
Back to the topic at hand, yeah, consumer hardware makes those assumptions and usually they're right. It also, as a bonus, lets them budget-segment the hell out of professional hardware that provides the IO people who actually use it really need. U.2/U.3 is a standard that never really made it into the consumer space because at the sizes consumers use, m.2 was plenty sufficient.
On the subject of cooling, liquid cooling has come a long way since I tossed my MasterLiquid Lite 240 for the pump noise. You can get reasonably safe coolers that barely make any noise at all. The companies that make AIOs typically have warranty programs in place that I believe will replace any hardware you had that would have been killed by a leak, but these days I'm pretty sure they're using something that won't kill your hardware as a coolant anyway. I still personally will never choose to do it again, but if a platform all but requires it, it's no longer a reason for me to not consider that platform anymore.