|
Truga posted:hot take: who gives a poo poo about 200 or 300 megs per second random access to 4k sized files? That's still thousands of files per second, you're not going to notice a difference in any real world scenario ever. its 200-300 megs out of what 2000 because nvme drives are insane on speed right?
|
# ? Apr 2, 2017 17:18 |
|
|
# ? May 21, 2024 15:30 |
|
Yea, it's not so noticeable for me but system boot is faster noticeably. And I do copy a lot of fairly big files around, at which NVME absolutely flies at. a lot of programs won't load noticeably faster GRINDCORE MEGGIDO fucked around with this message at 18:19 on Apr 2, 2017 |
# ? Apr 2, 2017 17:47 |
|
GRINDCORE MEGGIDO posted:A lot of programs won't load noticeably faster as they're compressed. Do you mean that the executable is compressed? Could you give an example of a major program that does this?
|
# ? Apr 2, 2017 17:56 |
|
Truga posted:hot take: who gives a poo poo about 200 or 300 megs per second random access to 4k sized files? That's still thousands of files per second, you're not going to notice a difference in any real world scenario ever. Anyone who'd be buying an NVMe drive for a desktop in the first place? On laptops sure its sometimes the only form of storage available. But if you're buying any of the current systems that use a Ryzen CPU, it's a normal size desktop that also has SATA ports where you can buy cheaper drives that move data slower over the SATA connection if you didn't care about the speeds.
|
# ? Apr 2, 2017 17:58 |
|
Truga posted:hot take: who gives a poo poo about 200 or 300 megs per second random access to 4k sized files? people looking at zen servers, if the NMVe results also extend to other PCIe ssd devices. desktop users, not so much if this only applies to the 4kb result and larger IOs are even with intel. Combat Pretzel posted:So even if general IOs are chunkier than 4KB, mix in fragmentation, and there's potential of it being split up in more actual IOs. NTFS cluster size is 4KB.
|
# ? Apr 2, 2017 18:06 |
|
Subjunctive posted:Do you mean that the executable is compressed? Could you give an example of a major program that does this? Reading more, I could be absolutely wrong about that being a general thing sorry. But app loads are often not much faster, I guess cos the read speeds advertised are at high queue depths?
|
# ? Apr 2, 2017 18:09 |
|
Klyith posted:I'm pretty sure SSDs don't work like that though. fragmentation into clusters is a spinning disk thing. Fragmentation is a filesystem characteristic. Drives deal in (virtual) regions, not files. Fragmentation isn't as much of a problem with SSDs, which is why defrag generally isn't worth the cost in drive wear, but loading a contiguous megabyte will be faster than if it's split over 250 4K pieces.
|
# ? Apr 2, 2017 18:16 |
|
All the files are 100% "fragmented" all the time on every decent SSD due to internal wear balancing shenanigans I thought?
Truga fucked around with this message at 18:29 on Apr 2, 2017 |
# ? Apr 2, 2017 18:27 |
|
Truga posted:All the files are 100% "fragmented" all the time on every decent SSD due to internal wear balancing shenanigans.
|
# ? Apr 2, 2017 18:29 |
|
Truga posted:All the files are 100% "fragmented" all the time on every decent SSD due to internal wear balancing shenanigans. I don't buy it. Not an expert on this by any means, but I think you're confusing two levels of fragmentation. SSDs have no issue with internally fragmented data, as you said, but there's another layer of fragmentation, at the OS/filesystem layer where reading one chunk of data can be represented either as one request of "give me this 1 MB contiguous block" or 1000 requests of "give me this 1 KB block" That fragmentation can be detrimental to performance as there is overheard involved with creating, transmitting and servicing such requests, I imagine.
|
# ? Apr 2, 2017 18:32 |
|
Reading say 1MB will always end up being 256 requests at 4KB regardless. The problem is when the data is scattered all over the place. As I've just mentioned, the internal block size is rather large. If all these requests happen to be located on a single of these 2MB slabs (say a sequental read or winning the block allocation lottery), it'll get read into RAM once and the SSD can service all the requests cached. If a lot of these 256 blocks are all over the place, it has to keep reading 2MB slabs to get all the data. The absolute worst case is that it'd need to read 512MB of data from the NAND to service that 1MB request from the app (256 requests of 2MB internal slabs).
|
# ? Apr 2, 2017 18:36 |
|
Oh yeah, I wasn't thinking about the filesystem. Yeah that'll be slower if it's fragmented.
|
# ? Apr 2, 2017 18:37 |
|
redeyes posted:4k Q1 is where nearly all home systems sit in terms of i/o load. ATTO on their benchmark indicated half the sustained reads that Intel gets. No, not really. Typical IO size for a desktop is going to float around 24-32kb/op. If you break down the throughput test on crystalmark to ops you'll see the AMD chipset is pulling 16k read and 51k write which is exceeding the specs for samsung SSD they're using. Intel is probably doing some kind of ram buffering/caching to get those numbers, which is concerning since it increases the risk of data loss. And that is magnitudes more IO performance than most home users will ever need, the impact in any real world use case is going to be minimal. Maybe if you're running a large SAP dev instance on your home computer you should think about it but for everyone else: doesn't matter.
|
# ? Apr 2, 2017 18:48 |
|
That buffering thing would only be valid for someone using the Intel RST drivers. If you're running MSAHCI, for all intents and purposes, the Intel just sees the OS communicating to something via PCIe.
|
# ? Apr 2, 2017 19:29 |
|
RST buffers to another block device, so that doesn't make sense here. It's something in the driver and using some kind of ram cache or write combining scheme seems most likely. You don't just magic 50% over the OEM's write OP spec out of nowhere.
|
# ? Apr 2, 2017 19:33 |
|
RST (Rapid Storage Technology) buffers in RAM. SRT (Smart Response Technology) caches to an SSD up to 64GB.
|
# ? Apr 2, 2017 19:52 |
|
Supposedly a BIOS update for MSI motherboards fixes performance in Rise of the Tomb Raider: https://www.youtube.com/watch?v=HX_WEvEzR64 Now RedGamingTech does a lot of rumour videos, and is fairly new to this benchmarking thing as far as I can tell, so take it with a grain of salt.
|
# ? Apr 2, 2017 21:48 |
|
Here's a video that explains the difference in approach between AMD and nVidias DX11 drivers, and explains how that results in worse performance for nVidia on Ryzen in some games: https://www.youtube.com/watch?v=nIoZB-cnjc0 Basically this is a more in depth look at what AdoredTV found in Rise of the Tomb Raider.
|
# ? Apr 3, 2017 12:30 |
|
I haven't seen this linked yet, a video purporting to be an R5 1400 vs a 7400 http://hexus.net/tech/news/cpu/104167-early-amd-ryzen-5-1400-review-comparison-leaks/
|
# ? Apr 3, 2017 12:45 |
|
In case you thought that we *weren't* headed towards our chips being assembled piecemeal going forward: http://wccftech.com/intel-kaby-lake-g-hbm2-gpu-multi-die/
|
# ? Apr 3, 2017 20:02 |
|
Well... Ain't that something.
|
# ? Apr 3, 2017 21:36 |
|
Is AMD's new APU going to be full Zen, a single CCX, with HBM? Does it have SMT? E- yea looks that way. Wonder what the HBM will do for min framerates with discrete cards. GRINDCORE MEGGIDO fucked around with this message at 01:35 on Apr 4, 2017 |
# ? Apr 4, 2017 01:32 |
probably close to what crystalwell did
|
|
# ? Apr 4, 2017 02:50 |
|
Watermelon Daiquiri posted:probably close to what crystalwell did It could be a really great chip for min frames. As Zen loves memory bandwidth it'll be interesting to see.
|
# ? Apr 4, 2017 03:06 |
|
Gigabyte has some beta bios if you are are adventurous, some users have bricked their boards testing (probably amateurs, they are all dual bios) http://forum.gigabyte.us/thread/886/am4-beta-bios-thread
|
# ? Apr 4, 2017 16:09 |
|
SwissArmyDruid posted:In case you thought that we *weren't* headed towards our chips being assembled piecemeal going forward:
|
# ? Apr 4, 2017 16:26 |
|
NUC sized boxes actually capable of playing games? YES PLEASE.
|
# ? Apr 4, 2017 17:43 |
|
Anime Schoolgirl posted:I'm going to chortle heartily if they use mini-vega for this this at least opens up the option for them to do so.
|
# ? Apr 4, 2017 18:55 |
|
Seamonster posted:NUC sized boxes actually capable of playing games? YES PLEASE. Laptops with real good battery life also.
|
# ? Apr 5, 2017 03:02 |
|
My motherboard finally arrived, everything is working. I'm experiencing none of the issues that I had with my old motherboard so it's making me feel justified in upgrading. I haven't had the time to do any real fun testing, and I'm too tired to. I'm just happy I didn't break the cpu via thermal paste or pin bending. It's my first time seeing pins and not LGA so that was interesting. Toalpaz fucked around with this message at 06:34 on Apr 5, 2017 |
# ? Apr 5, 2017 06:06 |
|
Early R5 1600 review, albeit in Spanish.
|
# ? Apr 5, 2017 12:56 |
|
Have we found out what they're using that THREADRIPPER trademark for yet?
|
# ? Apr 5, 2017 15:31 |
|
Toalpaz posted:It's my first time seeing pins and not LGA so that was interesting. Way to make me feel ancient
|
# ? Apr 5, 2017 15:33 |
|
It's been ages since I've taken a heatsink off and it conveniently pulls the chip out the socket as well. Good times.
|
# ? Apr 5, 2017 15:37 |
|
FuturePastNow posted:Have we found out what they're using that THREADRIPPER trademark for yet? Hopefully it's the new opteron
|
# ? Apr 5, 2017 15:38 |
i thought that was smt
|
|
# ? Apr 5, 2017 16:28 |
|
I thought it was just to keep people from writing Victorian AU bodice-ripper fanfic involving AMD faces. ....no? Just me? *discreetly closes notepad*
|
# ? Apr 5, 2017 19:01 |
|
Ohh.. Lisa, Raj, I didn't expect to see you in my salon at such a late hour, I'll call you a carriage home. >No, Lady Cinebench, we came for you. We heard your nT benchmarks needed a little massaging My... My... My points are so large! You have so many cores! And your threads... OH MY GOD YOUR THREADS ARE RIPPING ME APART We're lucky they didn't also trademark ScoreVore™ to describe whatever magic technology they're using for their next big thing. Or maybe unlucky. Who knows.
|
# ? Apr 5, 2017 19:12 |
|
FuturePastNow posted:Have we found out what they're using that THREADRIPPER trademark for yet?
|
# ? Apr 5, 2017 19:55 |
|
|
# ? May 21, 2024 15:30 |
|
Can somebody help me with something some guy on another forum is saying about single/dual rank memory with Ryzen? I've got a machine on order and the supplier is waiting for his supplier to deliver a bunch of 350M boards before dispatching my order all together instead of piecewise. My system is set to be Ryzen 1700/ MSI B350M Gaming Pro mobo / 2x16GB 3200Mhz DDR4, one stick of ram for each slot for 32GB of hotness. The guy is recommending that: quote:If you still want to buy a Ryzen 7 CPU for CFD here is a hint: use dual-rank DIMMs. You will not be able to get the same memory frequencies as you would with single-rank DIMMs. But as benchmarks have shown, Ryzen 7 performs better in memory-intensive applications (7-zip for example) with DDR4-2666 dual-rank than it does with DDR4-3200 single-rank and identical memory timings. Which is what I've already got, right? He doesnt provide a link to this benchmark, but from my own digging around I found this: http://www.legitreviews.com/amd-ryzen-single-rank-versus-dual-rank-ddr4-memory-performance_192960/5 Which says to me that my future setup is A-Ok. Have I got that right? OR I have missed something pretty fundamental? Plasmafountain fucked around with this message at 22:44 on Apr 5, 2017 |
# ? Apr 5, 2017 22:41 |