|
Dr. Video Games 0031 posted:
Golden edge makes it look like that whole thing plugs into another, much larger, motherboard. A grandmotherboard.
|
#
?
Jun 2, 2022 02:32
|
|
|
|
| # ? Jun 6, 2024 03:32 |
|
|
Not specifically an AMD question, but I'm looking to grab another 16gb of RAM before EOFY as a WFH tax write off. Currently have 2 sticks of Vengeance Pro 3200 CL16, so I was looking at grabbing another 2x8 kit of the same - but the 3600 CL18 seems to be more readily available with faster shipping etc for around the same price. As I understand it, it's probably the same RAM with different clock speed and timings and maybe a higher bin on clock speed on the 3600, and I would probably be able to manually set the timings in Bios to have it run at the same timings as the CL16, and possibly vice versa, but I do like being able to just enable XMP and go. My current ram has two different XMP profiles if I recall, but Corsair don't seem to advertise them anywhere. I have an MSI Mortar-M B550 board and a Ryzen 5600 with a GTX980 so there's basically not gonna be a noticable performance difference either way, just looking for a recommendation on whether I should wait longer and get the exact same RAM, or get the 3600 and either downclock it with the same timings as the 3200, or OC the 3200 a bit?
|
|
#
?
Jun 2, 2022 03:28
|
|
|
Don Dongington posted:Not specifically an AMD question, but I'm looking to grab another 16gb of RAM before EOFY as a WFH tax write off. Currently have 2 sticks of Vengeance Pro 3200 CL16, so I was looking at grabbing another 2x8 kit of the same - but the 3600 CL18 seems to be more readily available with faster shipping etc for around the same price. The best rule of thumb is to have exactly matched RAM, but downclocking the 3600 to match the 3200 would probably work. Personally I would do my damnedest to get exactly matched RAM.
|
|
#
?
Jun 2, 2022 03:36
|
|
|
repiv posted:Intel bought Killer two years ago, at this point I'd assume their newer products are just Intel NICs with gamerified drivers They are. The Killer E3100 is the same chip family as the I225/I226. Specifically, it looks to be a rebranded I225-LM with a different PCI device ID. e: Found it; it's exactly that. Vendor ID 0x8086 as expected, device ID 0x3100. Linux treats it just like a regular I225. Kazinsal fucked around with this message at 03:46 on Jun 2, 2022 |
|
#
?
Jun 2, 2022 03:38
|
|
|
CaptainSarcastic posted:The best rule of thumb is to have exactly matched RAM, but downclocking the 3600 to match the 3200 would probably work. Personally I would do my damnedest to get exactly matched RAM. Thanks for this. I grabbed the 3200mhz stuff, hopefully it arrives before June bloody 15th given that it's shipping from Amazon AU.
|
|
#
?
Jun 2, 2022 04:05
|
|
|
Don Dongington posted:As I understand it, it's probably the same RAM with different clock speed and timings and maybe a higher bin on clock speed on the 3600 This is not a good assumption when there's been a time spread between the old stuff and the new stuff. I had my own ram adventure recently where I thought my ram had gone bad. I bought 32gb of some cheapish corsair 3600-C18. It has a very hard time reducing latency even with lower clock speed. So it effectively performs worse than the 16gb 3000-C15 (which I ran at 3200-C16) that dates back to 2018.
|
|
#
?
Jun 2, 2022 17:00
|
|
|
Don Dongington posted:Thanks for this. I grabbed the 3200mhz stuff, hopefully it arrives before June bloody 15th given that it's shipping from Amazon AU. I'm assuming Corsair mostly sources memory chips from whatever supplier costs the least to hit MT/s, CL and XMP/AMD optimization now, even for middling Vengeance sticks (unlike the high-end stuff where they have to use the same top performers as everyone else). Corsair has this reputation for their LPX sticks already, they come in a million different versions and the only common feature is (mostly) hitting the advertised clocks and latency on specified platforms. Even if someone had V4.20 that ran perfectly and even overclocked a little, there's zero guarantee a V8.69 DIMM would overclock the same or even run at XMP together with the older sticks of the same rating. Running different versions of different speed/latency kits might work for 90% of people but if it doesn't work for you, have fun and good luck!
|
|
#
?
Jun 2, 2022 17:49
|
|
|
orcane posted:Good decision IMO. The opposite company is gskill- their alphabet soup of product codes and UPCs identifies the modules used and the bin.
|
|
#
?
Jun 2, 2022 22:51
|
|
|
how significant is the binning factor on b-die? There is Samsung b-die ECC but it is only available in 2666C19 sticks (Samsung M391A2K43BB1-CTD). But of course that's 1.2v and I assume if you run 1.35v or 1.4v then they'll juice right up to something a little more competitive? One thing that's slightly wack is apparently that's cheaper than a lot of the non-ECC binned gaming kits, the G-skill Trident Z Neo 3600C16 kit that's the current cheapest is $230 for 2 sticks and the Samsung is running around $100 a stick... obviously doesn't work on Intel chips without a C-series chipset (or at least would not have ECC functionality enabled - actually with most Intel chips shipping with ECC support now, it might actually boot but just the board might not enable ECC) but everyone is buying B-die kits for Ryzen and unless it's an APU then ECC should work fine... seems like you are paying more and losing ECC in return for RGB lighting and a factory voltage bump? again, unless you're running an ECC-locked part it seems like the Samsung would be the better choice if you're willing to punch in some voltages and manual timings edit: given that description I sent a question to a few vendors asking if they actually had the part or if this was an “equivalent” part Paul MaudDib fucked around with this message at 05:25 on Jun 3, 2022 |
|
#
?
Jun 2, 2022 23:31
|
|
|
Kazinsal posted:They are. The Killer E3100 is the same chip family as the I225/I226. Specifically, it looks to be a rebranded I225-LM with a different PCI device ID. Is it actually different silicon, or just different drivers on top of the same Foxville die?
|
|
#
?
Jun 3, 2022 04:38
|
|
|
movax posted:Is it actually different silicon, or just different drivers on top of the same Foxville die? I'm pretty sure it's like those terrible HWinfo/mobo temp monitoring software skins some Mobo vendors (*Cough* asus *cough*) would provide. Back in the CGI tits on the graphics card box era of computing.
|
|
#
?
Jun 3, 2022 04:41
|
|
|
movax posted:Is it actually different silicon, or just different drivers on top of the same Foxville die? That's an electron microscope question to be 100% sure, but based on the datasheet and exposed capabilities they appear to be identical chips with different PCI device IDs in the EEPROM.
|
|
#
?
Jun 3, 2022 05:27
|
|
|
orcane posted:Good decision IMO. Yeah when I built this system originally I used LPX 3200 C16, and it went bad in about 6 months. At that point it was a pretty typical goonbox with the understated MSI Tomahawk mobo (with the RGB LEDs set to steady red or off), the stock 2600 cooler and a mix of black and noctua brown fans. No RGB. Hadn't touched the stuff since the days when Antec cases shipped with tricolor fans. The LPX ended up getting caught up in an RMA battle with the fair trade ombudsman, so I grabbed the only other 3200mhz ram I could get at the time - the Vengeance Pro RGB stuff. Zen2 had just dropped, 3600CL18 was the new hotness, but because Zen+ had some issues with higher clocked memory iirc, I wanted to stay with what worked. 3 months later I had a 360mm coolermaster RGB AIO in there and since then, I've been fighting against the urge to replace the rest of my system fans, add some ARGB strips and swap out the gross corsair RM power supply leads with pretty ones. It's a bug that bites hard.
|
|
#
?
Jun 3, 2022 05:52
|
|
|
So basically Intel bought out a competitor just to destroy a product that was threatening them. I feel like that's probably an anti-trust violation, but it happens so often in this industry that it seems like everyone's just stopped keeping track.
|
|
#
?
Jun 3, 2022 05:53
|
|
|
Don Dongington posted:Yeah when I built this system originally I used LPX 3200 C16, and it went bad in about 6 months. At that point it was a pretty typical goonbox with the understated MSI Tomahawk mobo (with the RGB LEDs set to steady red or off), the stock 2600 cooler and a mix of black and noctua brown fans. No RGB. Hadn't touched the stuff since the days when Antec cases shipped with tricolor fans. The LPX ended up getting caught up in an RMA battle with the fair trade ombudsman, so I grabbed the only other 3200mhz ram I could get at the time - the Vengeance Pro RGB stuff. Zen2 had just dropped, 3600CL18 was the new hotness, but because Zen+ had some issues with higher clocked memory iirc, I wanted to stay with what worked. My pair of "3200C16" LPX bought in early 2017 needed insane IMC voltages to even get the advertised speed stable on a 8700K.
|
|
#
?
Jun 3, 2022 14:28
|
|
|
I stopped buying Corsair LPX entirely. Just too many sticks were failing or not hitting advertized speeds. G.Skill and Crucial have been good. And I do use B die on my personal boxes. Since changing to B die, stability over many months of no reboots has been near perfect.
|
|
#
?
Jun 3, 2022 14:31
|
|
|
poo poo I am using a 2x32GB kit of LPX 3600/18, what killed you guys' RAM do you think? I'm running at most power settings stock but voltage at about 1.38V to tighten timings and run in 1T instead of the default 2T, am I boned? I haven't had any practical issues so far and it's passing memtests and extended stability testing tweaked at 3600/16-18-18-34, in a less than 2 month old build. 
Agreed fucked around with this message at 19:04 on Jun 3, 2022 |
|
#
?
Jun 3, 2022 17:55
|
|
|
Anecdotal evidence is anecdotal. Personally I never had XMP RAM break, but I also never had any beyond DDR4-3200 on Intel platforms, and never overclocked them except for setting the BIOS XMP profiles. The most common complaint I know is that the XMP versions didn't work at rated speeds in Ryzen systems for a lot of people, especially with Zen/Zen+ CPUs (they made "for Ryzen" versions of LPX RAM which should have worked better, but you had to actively look for those). Sometimes RAM dies and it's hard to pinpoint why. I ran a set of four DDR3-2133 sticks at XMP settings with an i7-4790k for over two years without any issues, well cooled case etc. then one morning Windows would BSoD and as I tested the RAM, memtest showed that one of the sticks was suddenly faulty.
|
|
#
?
Jun 3, 2022 18:53
|
|
|
and sometimes they won't throw errors unless you run memtest on them, one stick at a time, which is a massive pain in the rear end when you're under time crunch to diagnose why the gently caress this computer.
|
|
#
?
Jun 3, 2022 19:38
|
|
|
Agreed posted:poo poo I am using a 2x32GB kit of LPX 3600/18, what killed you guys' RAM do you think? I'm running at most power settings stock but voltage at about 1.38V to tighten timings and run in 1T instead of the default 2T, am I boned? I haven't had any practical issues so far and it's passing memtests and extended stability testing tweaked at 3600/16-18-18-34, in a less than 2 month old build. I have this exact same set and I have to run 1.36 manual voltage on top of my XMP just to get the PC to boot stable. I can't even tighten the timings, any change to timings or voltage beyond 1.38 causes the whole system not to post. That was a head-scratcher when I put my pc together and turned XMP on and the pc wouldn't boot. I've confirmed the memory is stable at 1.36v with 8+ hours of synthetic testing on four different stress tests. Mine were Micron b-die, and low binned. The best I could do was change powerdown, geardown, and command rate. Anything you've got stable is better than not running at the rated XMP at the rated voltage haha
|
|
#
?
Jun 4, 2022 06:38
|
|
|
Negative_Kittens posted:I have this exact same set and I have to run 1.36 manual voltage on top of my XMP just to get the PC to boot stable. I can't even tighten the timings, any change to timings or voltage beyond 1.38 causes the whole system not to post. That was a head-scratcher when I put my pc together and turned XMP on and the pc wouldn't boot. I've confirmed the memory is stable at 1.36v with 8+ hours of synthetic testing on four different stress tests. Hint: auto XMP is a lovely consumer 'spec' that doesn't guarantee stability at the 'rated' speed, which is made even worse by all the stealth overvolts the mobo makers throw in when enabled to make RAM OCing look easy to unsuspecting newbies.
|
|
#
?
Jun 5, 2022 00:55
|
|
|
Palladium posted:Hint: auto XMP is a lovely consumer 'spec' that doesn't guarantee stability at the 'rated' speed, which is made even worse by all the stealth overvolts the mobo makers throw in when enabled to make RAM OCing look easy to unsuspecting newbies. If its also on the motherboard manufacturer's QVL then it is "guaranteed" at that speed and should be returned if its not possible.
|
|
#
?
Jun 5, 2022 01:04
|
|
|
Palladium posted:Hint: auto XMP is a lovely consumer 'spec' that doesn't guarantee stability at the 'rated' speed, which is made even worse by all the stealth overvolts the mobo makers throw in when enabled to make RAM OCing look easy to unsuspecting newbies. It doesn't always kick in either. Learned this the hard way when my 'xmp enabled' memory was still set to 1.2v and wouldn't boot
|
|
#
?
Jun 5, 2022 09:23
|
|
|
what are people's preferred tools for figuring out the best undervolt curves for their cpu's? i'm at -25 out of -30 on all cores so far on the 5800X3D with pbo2 tuner, which, as nice as it sounds, is bound to have some edge cases i'm not detecting so far i've used occt with sse instructions per core and corecycler. i'll try to keep corecycler running for longer, but maybe it's down to not stress testing each core for several hours per setting
|
|
#
?
Jun 7, 2022 21:14
|
|
|
kliras posted:what are people's preferred tools for figuring out the best undervolt curves for their cpu's? I'm waiting for better info to come out, it's great as is. I did gain way better temps by undervolting my 3080 FE
|
|
#
?
Jun 7, 2022 22:07
|
|
|
kliras posted:what are people's preferred tools for figuring out the best undervolt curves for their cpu's? AMD has the ability to allow only negative per core offsets in AGESA or ryzen master and deny options like positive offsets and PBO. There is likely a reason for denying negative offsets. …I will be joining you in finding out that reason tonight. To answer your actual question, core cycler is good, it may be over kill. I used prime95 blend on a 5600x because there’s a brief pause per thread between each test and that seems very stable. I did add 5 to the offsets just to be safe after finishing that though. (Ie, moved -25 to -20)
|
|
#
?
Jun 7, 2022 22:30
|
|
|
https://twitter.com/videocardz/status/1534996038641147904 Interesting strategy. I haven’t seen any leaks of a Zen4 Vcache part at launch, so might be a strange situation for gamers.
|
|
#
?
Jun 9, 2022 21:40
|
|
|
There's a lot more detail in this article. One thing that doesn't seem to add up is AMD's claims of Zen 4 having an 8-10% IPC increase and 15%+ single thread performance improvement when the clocks are expected to be a lot higher than Zen 3. Zen 5 in 2024. It's expected to be a bigger leap than Zen 4, improvements more akin to Zen 3's. Given HBM is being used in an increasing number of products, is it likely to get cheap enough due to economies of scale to make it to consumer platforms?
|
|
#
?
Jun 9, 2022 22:13
|
|
|
Cygni posted:Interesting strategy. I haven’t seen any leaks of a Zen4 Vcache part at launch, so might be a strange situation for gamers. https://twitter.com/greymon55/status/1534539721464025090 but more than just a single chip is expected
|
|
#
?
Jun 9, 2022 23:06
|
|
|
 It would have to be v-cache, there aint loving nowhere they can stick in any more chiplets. https://www.techpowerup.com/295642/amds-upcoming-zen-4-cpu-delidded-by-overclocker
|
|
#
?
Jun 9, 2022 23:13
|
|
|
it's probably just they haven't committed to using 3D as the branding yet for the Zen 4 parts yet
|
|
#
?
Jun 9, 2022 23:45
|
|
|
lih posted:it's probably just they haven't committed to using 3D as the branding yet for the Zen 4 parts yet Yeah, the way I took it is V-cache is the technology that, at least for Zen3, is branded as "3D" for Ryzen and "-X" for Epyc.
|
|
#
?
Jun 10, 2022 05:08
|
|
|
If raptor lake really is being delayed to next year, then AMD is saving v-cache for that and the 5800X3D will be stocked alongside the regular Zen 4 CPUs for at least a few months, maybe at a discount.
|
|
#
?
Jun 10, 2022 05:21
|
|
|
Dr. Video Games 0031 posted:If raptor lake really is being delayed to next year, then AMD is saving v-cache for that and the 5800X3D will be stocked alongside the regular Zen 4 CPUs for at least a few months, maybe at a discount. i would disagree about AMD, the V cache/3d stuff is just whatever is left over from the server market, if they have any leftovers its going to consumers, make them dollars.
|
|
#
?
Jun 10, 2022 05:33
|
|
|
8-10% higher ipc, the bulk of performance gain is achieved by the good ol' crowbar which has gained popularity lately. Ryzen 7000 TDP is going from 105W to 170W, max PPT from 142W to 230W. https://www.computerbase.de/2022-05/amd-ryzen-7000-raphael-mit-zen-4-soll-singe-core-ueber-15-prozent-zulegen/
|
|
#
?
Jun 10, 2022 08:28
|
|
|
sauer kraut posted:8-10% higher ipc, the bulk of performance gain is achieved by the good ol' crowbar which has gained popularity lately. urge to open wallet dead
|
|
#
?
Jun 10, 2022 11:19
|
|
|
sauer kraut posted:8-10% higher ipc, the bulk of performance gain is achieved by the good ol' crowbar which has gained popularity lately. 170W TDP is the maximum the socket is rated for, they haven't said anything about specific SKUs
|
|
#
?
Jun 10, 2022 13:39
|
|
|
Arzachel posted:170W TDP is the maximum the socket is rated for, they haven't said anything about specific SKUs Robert said on Reddit that Ryzen 7000 will launch with 170w TDP SKUs, it's not just the socket limit. We do know that the 5.5GHz gaming benchmark was running below 170w, however.
|
|
#
?
Jun 10, 2022 14:32
|
|
|
BurritoJustice posted:Robert said on Reddit that Ryzen 7000 will launch with 170w TDP SKUs, it's not just the socket limit. We do know that the 5.5GHz gaming benchmark was running below 170w, however. The 5950x probably could’ve been a 170W sku if they wanted to say that. I wouldn’t read too much into that tidbit.
|
|
#
?
Jun 10, 2022 14:38
|
|
|
|
| # ? Jun 6, 2024 03:32 |
|
|
sauer kraut posted:Ryzen 7000 TDP is going from 105W to 170W, max PPT from 142W to 230W.  and here i thought the 5800X was stupid chip to cool on air
|
|
#
?
Jun 10, 2022 19:52
|
|
