|
Anime Schoolgirl posted:it's going to eat poo poo in floating point artificials (that nobody but HPC people actually use, and they prefer POWER over anything x86 by a long shot) Who cares about floating point on cpu when you have gpu compute?
|
#
?
Feb 9, 2017 02:36
|
|
|
|
| # ? Oct 14, 2024 06:00 |
|
|
Kazinsal posted:(16 THREADRIPPERS? you don't trademark a name like that and never loving use it wargames posted:Who cares about floating point on cpu when you have gpu compute?
|
|
#
?
Feb 9, 2017 02:36
|
|
)
|
Anime Schoolgirl posted:this has gotta be a loving thing People who hate trying to code poo poo in CUDA?
|
|
#
?
Feb 9, 2017 04:00
|
|
|
2017 is shaping up to be a kick rear end year, if these rumors are true
|
|
#
?
Feb 9, 2017 04:26
|
|
|
I want Ryzen benchmarks already gat damnit.
|
|
#
?
Feb 9, 2017 04:34
|
|
|
teagone posted:I want Ryzen benchmarks already gat damnit. This. I am dying for actual independent benchmarks. I want to upgrade my emulation HTPC from a Haswell i3 and a 4 or 6 core box would be great for other duties like Plex. The latent AMD fan boy in me is cautiously optimistic about Ryzen and 2017.
|
|
#
?
Feb 9, 2017 05:36
|
|
|
http://www.cnbc.com/2017/02/08/trump-meets-intel-ceo-brian-krzanich.html yeah intel is trash
|
|
#
?
Feb 9, 2017 05:46
|
|
|
See, I don't mind letting meeting with Kim Don-Un, I mind that Krzanich let BB-Hate take credit for something that Putin's Pumpkin Puppet clearly had nothing to do with.
|
|
#
?
Feb 9, 2017 06:10
|
|
|
SwissArmyDruid posted:See, I don't mind letting meeting with Kim Don-Un, I mind that Krzanich let BB-Hate take credit for something that Putin's Pumpkin Puppet clearly had nothing to do with. https://www.youtube.com/watch?v=Hn7B5FdwBK8
|
|
#
?
Feb 9, 2017 07:10
|
|
|
Yeah, you're right. This isn't the place for politics.
|
|
#
?
Feb 9, 2017 07:36
|
|
|
https://www.custompcreview.com/news/amd-ryzen-7-1700-1700x-1800x-cpu-pricing-leaked-coming-sub-500-pricepoints/37559/
|
|
#
?
Feb 9, 2017 08:08
|
|
|
It's getting increasingly hard not to get excited. $490 for the top 8C/16T model would be a complete game changer if they can manage Haswell level IPC. I am a bit worried that the aggressive pricing is a hint at lower performance but we'll have to wait for benchmarks to find out. Unlike with Bulldozer the time is ripe for a bump in mainstream core count, particularly since many games these days are optimised for the 8 cores in the current gen consoles. Looking at Vega/Polaris performance I'm fairly sure that this would also lock down AMD's APUs for the next console generation. Opteron based Mac Pros with more than 16C/32T would also be in the cards.
|
|
#
?
Feb 9, 2017 08:21
|
|
|
wargames posted:Who cares about floating point on cpu when you have gpu compute? There are probably enough latency sensitive workloads that offloading to GPU or even APU doesn't always make sense. Too bad Intel kneecapped AVX by disabling it on Pentiums/Celerons for some reason.
|
|
#
?
Feb 9, 2017 08:36
|
|
|
If the 8C/16T model is $490 and has a similiar Haswell level IPC then I might have to jump ship from my I7 6700K and eat the costs of selling it on Craigslist. The used CPU market is going to get destroyed though.
|
|
#
?
Feb 9, 2017 08:41
|
|
|
looking forward to the 8c16t haswell xeon 70-90 dollar specials sooner than a couple years if ryzen makes the landing
|
|
#
?
Feb 9, 2017 08:44
|
|
|
Arzachel posted:There are probably enough latency sensitive workloads that offloading to GPU or even APU doesn't always make sense. Too bad Intel kneecapped AVX by disabling it on Pentiums/Celerons for some reason. How many nano/mili seconds does it take to offload work?
|
|
#
?
Feb 9, 2017 08:44
|
|
|
wargames posted:How many nano/mili seconds does it take to offload work? Tens of microseconds.
|
|
#
?
Feb 9, 2017 08:56
|
|
|
Platystemon posted:Tens of microseconds. I am curious what workloads are that sensitive?
|
|
#
?
Feb 9, 2017 09:07
|
|
|
RyuHimora posted:https://www.custompcreview.com/news/amd-ryzen-7-1700-1700x-1800x-cpu-pricing-leaked-coming-sub-500-pricepoints/37559/ See, if those are distributor prices, then we'll be paying more than this. Also, just for fun: https://www.forbes.com/sites/greats...e/#4cf272d06c6f PerrineClostermann fucked around with this message at 09:18 on Feb 9, 2017 |
|
#
?
Feb 9, 2017 09:14
|
|
|
wargames posted:How many nano/mili seconds does it take to offload work? Best case scenario would be ~5 microsecond round trip just for the PCI-E transfer with any communication overhead on top of that. That's 10 000+ cycles before you factor in the actual execution. Arzachel fucked around with this message at 09:34 on Feb 9, 2017 |
|
#
?
Feb 9, 2017 09:23
|
|
|
isn't that the very reason why HPC uses POWER8/9?
|
|
#
?
Feb 9, 2017 09:24
|
|
|
Anime Schoolgirl posted:isn't that the very reason why HPC uses POWER8/9? PCI-E bandwidth is what tends to bottleneck HPC applications, which is why stuff like NVLink is being developed. Not having to rewrite everything in CUDA also helps.
|
|
#
?
Feb 9, 2017 09:51
|
|
|
Arzachel posted:PCI-E bandwidth is what tends to bottleneck HPC applications, which is why stuff like NVLink is being developed. Not having to rewrite everything in CUDA also helps. Are they actually running out of bandwidth with 3.0 x16 because titan xps don't get limited?
|
|
#
?
Feb 9, 2017 09:55
|
|
|
If you look at the top500 list, these days it's pretty much all Intel x86_64 followed by POWER (4.5%), AMD and SPARC (1.5%). Even if POWER is theoretically better, I think most people have ended up better off spending their money on more mainstream Intel gear. In the domain I've used a HPC, Computation Fluid Dynamics, a lot of the COTS programs only come in x86 versions.
Pablo Bluth fucked around with this message at 10:17 on Feb 9, 2017 |
|
#
?
Feb 9, 2017 10:14
|
|
|
Is there any word on quad channel memory or ECC versions?
|
|
#
?
Feb 9, 2017 10:40
|
|
|
Second Sun posted:Is there any word on quad channel memory or ECC versions? 2) only if the mobo manufacturers provide traces for it
|
|
#
?
Feb 9, 2017 10:44
|
|
|
edit: Oops, poo poo, wrong thread.
SwissArmyDruid fucked around with this message at 21:29 on Feb 9, 2017 |
|
#
?
Feb 9, 2017 11:21
|
|
|
For the first time in 9 months, I feel a twang of regret buying a Skylake.
|
|
#
?
Feb 9, 2017 13:41
|
|
|
Nam Taf posted:For the first time in 9 months, I feel a twang of regret buying a Skylake. You should probably wait for benchmarks before you feel any regret. It wouldn't be the first time that everything looks good on paper and then womp womp, big wet fart.
|
|
#
?
Feb 9, 2017 13:47
|
|
|
One great exciting thing about all this is if >4 threads becomes really mainstream, and games take advantage of that across the board. I still hope AMD or Intel stick a big fat cache on a chip soon, though, 5775c style. I'm interested in how Zen's memory controller copes with fast RAM speeds, to help min FPS. GRINDCORE MEGGIDO fucked around with this message at 14:22 on Feb 9, 2017 |
|
#
?
Feb 9, 2017 14:19
|
|
|
GRINDCORE MEGGIDO posted:One great exciting thing about all this is if >4 threads becomes really mainstream, and games take advantage of that across the board. I'm pretty sure they've been doing that for a while out of pure necessity, seeing as the current gen consoles have 8-core netbook class CPUs
|
|
#
?
Feb 9, 2017 14:22
|
|
|
GRINDCORE MEGGIDO posted:One great exciting thing about all this is if >4 threads becomes really mainstream, and games take advantage of that across the board. The CPU performance benefit from the L4 cache was miniscule and it took up more die space than the cores themselves did, I don't think we'll see it used for anything else than boosting iGPU bandwidth.
|
|
#
?
Feb 9, 2017 14:47
|
|
|
Arzachel posted:The CPU performance benefit from the L4 cache was miniscule and it took up more die space than the cores themselves did, I don't think we'll see it used for anything else than boosting iGPU bandwidth. I'm happy with it being aimed at iGPU, if it's accessibly by the CPU also - as the improvement in min framerates with discrete graphics with the 5775c seemed really worthwhile. I suppose as memory controllers get faster though and provide less latency it's less of a thing. /armchair speculation HalloKitty posted:I'm pretty sure they've been doing that for a while out of pure necessity, seeing as the current gen consoles have 8-core netbook class CPUs This is a good point, I am now less excited. GRINDCORE MEGGIDO fucked around with this message at 16:43 on Feb 9, 2017 |
|
#
?
Feb 9, 2017 14:53
|
|
|
PC watch has more details on Zen, i.e. 3.4 Ghz+ Base clock, per-core voltage regulation, "AMD asserts that it will be able to compete with Intel's Skylake generation CPU core at a level equal to that of a single-threaded performance and performance-efficient design.", die shot, etc. machine translated link here eames fucked around with this message at 16:42 on Feb 9, 2017 |
|
#
?
Feb 9, 2017 16:38
|
|
|
If it turns out they actually managed to get near Skylake performance then they'll have succeeded better than anyone (including me) expected. I think Kanter is still expecting Haswell-ish performance for non-HPC work loads and normally he is spot on. That expectation makes a lot more sense given the benches we have so far too. Really I'd be perfectly fine with a 8C/16T 3.5Ghz+ chip for less than $400 with that level of performance but actually getting on par with Skylake would explain lots about those rumors of Intel doing some sort of panic release to try and combat Zen in the near term.
|
|
#
?
Feb 9, 2017 16:47
|
|
|
eames posted:PC watch has more details on Zen, i.e. 3.4 Ghz+ Base clock, per-core voltage regulation, "AMD asserts that it will be able to compete with Intel's Skylake generation CPU core at a level equal to that of a single-threaded performance and performance-efficient design.", die shot, etc. Didn't we already know about the minimum base clock and fine grain power regulation?
|
|
#
?
Feb 9, 2017 17:35
|
|
|
PerrineClostermann posted:Didn't we already know about the minimum base clock and fine grain power regulation? Yeah, it was in the official reveal a while back
|
|
#
?
Feb 9, 2017 17:37
|
|
|
Good, not losing my mind here then. Also I wouldn't start " regretting" buying Skylake. The big news is AMD getting close to or matching Intel, not obsoleting Intel. Your 6700k is still going to be a great CPU.
|
|
#
?
Feb 9, 2017 17:42
|
|
|
That's true, but I was just so underwhelmed by the upgrade from 2500k to 6700k, I noticed my new fan being quieter a lot more than any performance gains. Framerates barely budged: I thought I might be CPU limited on a 980 in games like Arma 3, nope not at all. Why did I spend all this money again? After five years of waiting feeling that underwhelmed was really just a bummer. If AMD can double my cores or boost IPC more than a dinky Intel 5% every two years or whatever then everyone is going to jump to AMD immediately. gently caress Intel.
|
|
#
?
Feb 9, 2017 17:45
|
|
|
|
| # ? Oct 14, 2024 06:00 |
|
|
Pryor on Fire posted:That's true, but I was just so underwhelmed by the upgrade from 2500k to 6700k, I noticed my new fan being quieter a lot more than any performance gains. Framerates barely budged: I thought I might be CPU limited on a 980 in games like Arma 3, nope not at all. Why did I spend all this money again? I understand that sentiment, I suppose. I'm on a 2600k. I said I wouldn't upgrade until I could get double the cores (with per core improvements) and double the RAM. Right now double the RAM costs about the same as my ram did, but CPUs didn't get too far ahead... I've upgraded pretty much everything I can other than CPU/RAM/Mobo. Added SSDs, moved to a proper case, went full water cooling loop, added a new GPU, got an Ultrawide. If Zen ends up not being too compelling, at least it might get Intel moving in computing performance again.
|
|
#
?
Feb 9, 2017 17:55
|
|