|
So what about ryzen 2 huh
|
# ? May 17, 2018 05:31 |
|
|
# ? Jun 8, 2024 06:15 |
|
If AMD isn't able to get clocks much higher in the next couple years because of process problems, I wonder if they should go all in on the slower but wider idea. Add more execution units to each core and put in 4-way SMT instead of 2-way. I know IBM made some Power chips that do that, and I think there's an ARM server chip that also has 4-way SMT.... it could potentially be an easy way to get a lot better in some server workloads.
|
# ? May 17, 2018 05:54 |
|
IIRC there are POWER and SPARC chips with 8-way SMT. 2-way has only been the standard for x86 because consumer applications haven't been in a position to gain any dramatic performance increases from anything more until now. Possibly even now.
|
# ? May 17, 2018 08:25 |
|
The case where you gain in scaling beyond 2-way SMT is a workload that does not max out any one individual thread, any two threads do not max out a single core, and the workload scales close to linearly with core count. Which very much approaches a "manycore" processor model. The difference between "multicore" and "manycore" being that "manycore" is designed from the beginning with explicit parallelism in mind. Just like a GPU! At any rate, that's not a situation that you would see in any kind of desktop workload ever, where applications are free to ramp up and devour an entire core's worth of resources. I think we're more likely to see AMD put to use the tech they were parading around a few years back, where threads are allocated dynamically across the full number of cores, allocating a non-integer number of cores >= 1 to a single thread. I'll remember what it was called soon enough, just give me a minute. edit: Remembered: VISC architecture. https://techreport.com/news/27259/cpu-startup-claims-to-achieve-3x-ipc-gains-with-visc-architecture Unfortunately, my completely forgetting that they had existed means that I had missed that they had been bought out by Intel back in 2016. As penance, have an article with claims by Intel that have *not* aged well. https://www.cnet.com/news/intel-pledges-80-cores-in-five-years/ SwissArmyDruid fucked around with this message at 11:29 on May 17, 2018 |
# ? May 17, 2018 08:48 |
|
VostokProgram posted:If AMD isn't able to get clocks much higher in the next couple years because of process problems, I wonder if they should go all in on the slower but wider idea. Add more execution units to each core and put in 4-way SMT instead of 2-way. I know IBM made some Power chips that do that, and I think there's an ARM server chip that also has 4-way SMT.... it could potentially be an easy way to get a lot better in some server workloads. Khorne fucked around with this message at 13:45 on May 17, 2018 |
# ? May 17, 2018 13:41 |
|
VostokProgram posted:If AMD isn't able to get clocks much higher in the next couple years because of process problems, I wonder if they should go all in on the slower but wider idea. Add more execution units to each core and put in 4-way SMT instead of 2-way. I know IBM made some Power chips that do that, and I think there's an ARM server chip that also has 4-way SMT.... it could potentially be an easy way to get a lot better in some server workloads. EPYC can do 4 way smt right?
|
# ? May 17, 2018 14:50 |
|
wargames posted:EPYC can do 4 way smt right? Nope, not yet at least. Some of the new ARM architectures (Cavium ThunderX 2) have 4-way SMT currently.
|
# ? May 17, 2018 14:59 |
|
A GF fab guy said that they're expecting 5Ghz range chips with their 7nm process although he didn't specify which chip exactly he was talking about its generally accepted he meant Zen2 FWIW.
|
# ? May 17, 2018 15:17 |
|
PC LOAD LETTER posted:A GF fab guy said that they're expecting 5Ghz range chips with their 7nm process although he didn't specify which chip exactly he was talking about its generally accepted he meant Zen2 FWIW. They're taking about Vega 7nm obviously. It's the only way to get it competitive with nVidia's new chips. (Subscription to home LN2 delivery service required, not available in all areas.)
|
# ? May 17, 2018 16:51 |
|
PC LOAD LETTER posted:A GF fab guy said that they're expecting 5Ghz range chips with their 7nm process although he didn't specify which chip exactly he was talking about its generally accepted he meant Zen2 FWIW. gently caress
|
# ? May 17, 2018 17:16 |
|
A threadripper part at 5GHz is basically the dream, so if they can deliver that....
|
# ? May 17, 2018 17:26 |
|
PC LOAD LETTER posted:A GF fab guy said that they're expecting 5Ghz range chips with their 7nm process although he didn't specify which chip exactly he was talking about its generally accepted he meant Zen2 FWIW. Oh gently caress I stand corrected. GloFo making a good process is but I'm hyped
|
# ? May 17, 2018 17:34 |
|
Does it really count as GloFo when they brought in the IBM team years ago?
|
# ? May 17, 2018 17:55 |
|
FaustianQ posted:Does it really count as GloFo when they brought in the IBM team years ago? No True GloFlo
|
# ? May 17, 2018 18:51 |
|
FaustianQ posted:Does it really count as GloFo when they brought in the IBM team years ago? The entire 7nm coalition was basically 'this is too expensive to do ourselves, team up so we can steal Intel's lunch, cool?"
|
# ? May 17, 2018 23:04 |
|
Methylethylaldehyde posted:The entire 7nm coalition was basically 'this is too expensive to do ourselves, team up so we can steal Intel's lunch, cool?" Turns out they didn't have to intel TMG shat in their own lunch, multiple times.
|
# ? May 17, 2018 23:30 |
|
Crossposting from the Parts thread, would y'all choose the 1700 or 2600 if they were the same price?
|
# ? May 17, 2018 23:34 |
|
Well. 1700 is 8-core, 2600 is 6-core, but the 2600 has better clocks.... I'd say 2600, unless you're using all eight cores. In which case, look for a price where the 1700 is cheaper than the 2600.
|
# ? May 17, 2018 23:43 |
|
ItBreathes posted:Crossposting from the Parts thread, would y'all choose the 1700 or 2600 if they were the same price? Workstation heavy 1700. Gaming heavy 2600. I have not noticed any difference since running my 1700 at stock clocks, most of my workloads benefit from having more cores. I've even started twitch streaming since doing it. Anarchist Mae fucked around with this message at 23:47 on May 17, 2018 |
# ? May 17, 2018 23:44 |
|
ItBreathes posted:Crossposting from the Parts thread, would y'all choose the 1700 or 2600 if they were the same price? Based on your use case, the 2600 or an i5-8400. Overclocked, with at least CL14 3200 memory the 2600 will beat the i5-8400, it needs to be overclocked and it needs fast memory. So do a price comparison taking system memory into account, not just CPU+Mobo. I'd spring for the 2600 personally, as the 3000 series is going to be a major jump early next year, AM4 will support up to at least the 4000 series maybe the 5000 series depending on how long it takes DDR5 to available for consumers. B360 and Z370 are already dead platforms and Intels follow up processors on 10nm are moving to late 2019 and do not look as promising. EmpyreanFlux fucked around with this message at 23:47 on May 17, 2018 |
# ? May 17, 2018 23:45 |
|
Thanks y'all. Follow up question: somehow, all of the builds I've ever done have had ASRock Pro4 motherboards, and I expect this to continue, any reason not to? Notably, Microcenter only has B350 in mATX. On the memory front I have CL16 3000 and selling & replacing it sounds like , though I'll probably try to gently caress with the timings. Beating the 8400 isn't a priority, and I'll likely be GPU limited anywhere it would matter anyways.
|
# ? May 18, 2018 03:53 |
|
ItBreathes posted:Thanks y'all. Follow up question: somehow, all of the builds I've ever done have had ASRock Pro4 motherboards, and I expect this to continue, any reason not to? Notably, Microcenter only has B350 in mATX. the only reason would be better power for future zen processors: the 2xxx series is boosting itself well above it's normal TDP if it has enough heat dissipation & power delivery. next zen will probably be even more of that. However I'm not sure there exists a mATX with actually good power. Probably you'd have to pay twice the money for one of the more high-end ones from gigabyte or asus. That asrock you're looking at is adequate even though the VRM is 3 phase that's badly faking a 6. It'll at least deliver power without overheating since it has a heatsink, unlike some of the other mATX boards.
|
# ? May 18, 2018 19:38 |
|
MaxxBot fucked around with this message at 19:58 on May 18, 2018 |
# ? May 18, 2018 19:46 |
|
Most of the testing I've seen on AM4 more or less shows that you're going to be limited by the chip before the VRM, unless you are doing crazy LN2 stuff or obsessed with eeking out the last 10mhz. For example, AT recently tested the Asrock mITX X370 board with a 3x2 VRM and hit the same overclocks that the rest of the boards had been able to hit with their same sample chip. https://www.anandtech.com/show/12569/the-asrock-x370-gaming-itx-ac-motherboard-review e: so i guess what im sayin' is if you arnet OCing, I wouldnt worry about it at all, and if you are... like i guess it matters a little maybe but i wouldnt spend much for it Cygni fucked around with this message at 20:12 on May 18, 2018 |
# ? May 18, 2018 19:57 |
|
Cygni posted:Most of the testing I've seen on AM4 more or less shows that you're going to be limited by the chip before the VRM, unless you are doing crazy LN2 stuff or obsessed with eeking out the last 10mhz. The funny part is nobody really gave a crap about VRMs back in the FSB OCing days when chips are actually worth OCing with 50+% headroom
|
# ? May 19, 2018 13:38 |
|
Palladium posted:The funny part is nobody really gave a crap about VRMs back in the FSB OCing days when chips are actually worth OCing with 50+% headroom Vrm are the only things mobo manufacturers can differentiate on and are dirt cheap for the profit they can extract
|
# ? May 19, 2018 14:15 |
|
https://www.techpowerup.com/reviews/AMD/Ryzen_7_2700X/20.html 2700X OCing: Only 0.8% better overall in performance, worse gaming performance and 27W more power draw versus stock do-nothing XFR2. 11/10 will OC again because its not pointless yet
|
# ? May 19, 2018 16:33 |
|
Palladium posted:https://www.techpowerup.com/reviews/AMD/Ryzen_7_2700X/20.html BLCK overclocking works with XFR2. Besides, Ryzen gets crazy performance increases from memory clocks and timings so I wouldn't proclaim overclocking as dead just yet
|
# ? May 19, 2018 20:22 |
|
Cygni posted:e: so i guess what im sayin' is if you arnet OCing, I wouldnt worry about it at all, and if you are... like i guess it matters a little maybe but i wouldnt spend much for it People watch Buildzoid videos (GamersNexus features him a lot and people love that channel) and see him crap on motherboards for their VRMs and don’t understand that the guy they’re listening to likes to overclock chips well past what most goons would. If you’re overclocking only up to the point where mainstream review sites say, “past this there’s a significant jump in heat with minimal performance gains, and we don’t recommend it”, almost any board will do for you. You’re listening to a guy who wants to run extreme benchmarks without breaking anything; because unlike Derbauer or whoever he isn’t well-sponsored enough to take a $400 motherboard and a $600 processor and go glory run and if the thing broke oh well. Once you understand that context, it makes more sense. He even admits he hates most low end boards, but that’s because he plans to run half a gHz higher than the point of diminishing returns for home users.
|
# ? May 19, 2018 21:43 |
|
Palladium posted:The funny part is nobody really gave a crap about VRMs back in the FSB OCing days when chips are actually worth OCing with 50+% headroom cpus consumed less power & were way less sensitive to fluctuations back then, and the OC limiter was more the primitive heatsinks of the day and quality of your FSB clock. Hell, I worked as a tech back during the big capacitor plague year. you know what the amazing thing was? the computers still mostly worked. Like, they'd crash or even hard restart once an hour or more, but still, you could turn it on and boot to windows despite the caps barfing their guys all over the inside of the machine! Kinda amazing when I think about it. No way would a modern cpu work under those conditions. Craptacular! posted:People watch Buildzoid videos (GamersNexus features him a lot and people love that channel) and see him crap on motherboards for their VRMs and dont understand that the guy theyre listening to likes to overclock chips well past what most goons would. I have a MSI mobo so obviously I'm not taking his reviews as blind gospel. The guy asked if there was any reason not to use the Pro4 and that's the reason I could think of. As CPUs get these more sophisticated power management features that effectively are self-OCing, I think VRMs are gonna matter more. Obviously not to the extent that Buildzoid does where he calls anything that can't do continuous 250W "crap". But transient spikes in that ballpark -- which people have measured the 2700X doing right now -- put more stress on those components. Which for normal people like myself is more about longevity than performance. Caps and regulators wear out, running hotter wears them out faster, overspecced VRM runs cooler. If this was intel and I knew my mobo was 100% disposable I wouldn't care. But with AMD's promise to keep AM4 going until DDR4 or something else forces incompatibility, I'm actually thinking twice about this stuff. I will likely put a Zen 2 CPU in this board, and easily might keep using it for 5 or 6 years -- if it keeps working that long.
|
# ? May 19, 2018 23:12 |
|
Has anyone played with very high speed memory in their Ryzens? 3800 or faster mahz? Stuff like this? https://www.newegg.com/Product/Prod...ICE&PageSize=36 Haven't had much luck finding any performance comparisons or how often these speeds are achievable.
|
# ? May 20, 2018 00:55 |
|
LRADIKAL posted:Has anyone played with very high speed memory in their Ryzens? 3800 or faster mahz? Stuff like this? You need golden samples but someone has hit 4000Mhz, CL14. FaustianQ posted:In crazy batshit news, someone has been able to push DDR4 4000 CL14 with Ryzen 2. Latency and clockspeed both matter, and Ryzen shows quickly diminishing returns beyond CL14 3600Mhz or CL12 3200Mhz. I'd guess 4000Mhz CL14 is ~6% faster than CL14 3600Mhz. Most of those are CL19 and at best CL17, there might be no change in performance or even a regression due to the latency.
|
# ? May 20, 2018 02:25 |
|
FaustianQ posted:You need golden samples but someone has hit 4000Mhz, CL14. These are all numbers I can't afford so
|
# ? May 20, 2018 05:54 |
|
If you're strapped for cash go for the DDR4 2933/3000. Gets most of the gains of DDR4 3200 and usually doesn't cost much more than DDR4 2666. Eyeballing newegg real quick shows it be as little as a $1 difference for some of the cheaper 2x8GB kits. For DDR4 2400 vs DDR4 3000 its a $10 difference. So if you can afford to buy 8-16GB of DDR 2400/2666 you can deffo afford to buy DDR4 2933/3000. Bang for the buck is absolutely worth it. For the DDR4 3200+ low latency kits...not so much.
|
# ? May 20, 2018 06:55 |
|
Pinnacle's IMC is also much better so even getting CL16/3000 is okay since tweaking it to CL14 or 3200 shouldn't be too much of a task.
|
# ? May 20, 2018 07:33 |
|
I, for one, am going to be super glad when overclocking is dead, because buying a chip that can do 37ghz but running it at 35 just feels like leaving money on the table for no reason, but on the other hand I don't want to spend 6 hours figuring out what freq it'll top out at so if the chipmaker does that bit for me that's better as far as I'm concerned.
|
# ? May 20, 2018 10:51 |
|
Leaving performance on the table is something companies do when they are getting complacent about their market position. When companies feel competitive pressure they push each piece of silicon close to its limit, and TDP suffers. The amount of OC headroom that a company leaves is probably the best platonic indication of how much competitive pressure they're feeling. Intel's TDP is about to jump up, and their headroom is going to nosedive, going against 7nm Ryzen. Paul MaudDib fucked around with this message at 11:16 on May 20, 2018 |
# ? May 20, 2018 11:09 |
|
Sony's devs are working on LLVM compiler improvements for the current Zen architecture. This should be good news down the road for all gamers as it seems exceedingly likely that the PS5 will launch with some sort of Ryzen CPU/APU. The PS4 launched with 8x1.6 GHz at Piledriver IPC, even a basic 8x3.0 Ghz Zen would be 3-4x the CPU performance. https://www.phoronix.com/scan.php?page=news_item&px=Sony-LLVM-Ryzen-Improvements
|
# ? May 20, 2018 13:41 |
|
Craptacular! posted:People watch Buildzoid videos (GamersNexus features him a lot and people love that channel) and see him crap on motherboards for their VRMs and don’t understand that the guy they’re listening to likes to overclock chips well past what most goons would. If you’re overclocking only up to the point where mainstream review sites say, “past this there’s a significant jump in heat with minimal performance gains, and we don’t recommend it”, almost any board will do for you. The point of quality VRMs on AM4 is longevity. In two senses, one your motherboard is more likely than your CPU to crap out in the lonterm. And the second, AM4 still has ~2 generations of CPUs left on it and it's likely they'll be power hungry.
|
# ? May 20, 2018 13:54 |
|
|
# ? Jun 8, 2024 06:15 |
|
Paul MaudDib posted:Leaving performance on the table is something companies do when they are getting complacent about their market position. When companies feel competitive pressure they push each piece of silicon close to its limit, and TDP suffers. Pentium 4 debacle all over again.
|
# ? May 20, 2018 14:24 |