|
I thought that only pertained to DX10/11, and DX12 was a clean-sheet design without all that AMD legacy cruft?
|
# ? Mar 31, 2017 14:21 |
|
|
# ? May 21, 2024 10:20 |
|
JnnyThndrs posted:I thought that only pertained to DX10/11, and DX12 was a clean-sheet design without all that AMD legacy cruft? I haven't seen it extensively tested, but Hardware Unboxed did notice that AMD still had higher CPU overhead in Doom/Vulkan: There might be some inefficiency in the backend of the driver that's common to all APIs.
|
# ? Mar 31, 2017 14:40 |
|
Huh, that's interesting, maybe, like you said, it's intrinsic to GCN. Definitely hitting the CPU harder.
|
# ? Mar 31, 2017 14:46 |
|
Update on the Ashes benchmark discrepancy: it looks like the game silently scales the detail level based on the number of cores available. And Oxide didn't think to tell anyone this or make it configurable http://www.pcgameshardware.de/commoncfm/comparison/clickSwitch.cfm?id=138531 i'm sure hexa/octa-core owners will be thrilled that oxide has blessed them with particle rendering from 500 miles away at the low low cost of half their fps repiv fucked around with this message at 18:55 on Mar 31, 2017 |
# ? Mar 31, 2017 18:12 |
|
Jesus Christ.
|
# ? Mar 31, 2017 20:15 |
|
Another benchmark showing 30% DX12 gains: https://www.youtube.com/watch?v=QBf2lvfKkxA
|
# ? Mar 31, 2017 20:48 |
|
repiv posted:Update on the Ashes benchmark discrepancy: it looks like the game silently scales the detail level based on the number of cores available. And Oxide didn't think to tell anyone this or make it configurable
|
# ? Mar 31, 2017 21:01 |
|
Anime Schoolgirl posted:lmao who the gently caress didn't tell them that ryzen was a 4+4 core complex That has little to do with the stupidity of having hidden detail settings, regardless the Intel 6/8 core models actually do worse.
|
# ? Mar 31, 2017 23:07 |
|
Edit: Totally the wrong thread apologies.
rex rabidorum vires fucked around with this message at 01:41 on Apr 1, 2017 |
# ? Apr 1, 2017 00:43 |
|
Tbf is ashes actually played by anyone
|
# ? Apr 1, 2017 01:27 |
|
It would be hilarious if Ashes were being kept afloat solely by people buying it for benchmarking purposes.
|
# ? Apr 1, 2017 01:30 |
|
With all the RNG and hidden detail settings for Ashes, isn't it kinda not work benchmarking? Unless you're doing like a thousand runs of it?
|
# ? Apr 1, 2017 01:44 |
|
Shame it can't scale details up or down depending on your refresh rate.
|
# ? Apr 1, 2017 02:08 |
|
Except you can't even run it a thousand times, because _the benchmarks get better over time_ because of the branch prediction learning thing.
|
# ? Apr 1, 2017 02:18 |
|
Where can I get an Intel bracket for the wraith max CPU cooler?
|
# ? Apr 1, 2017 02:54 |
|
SwissArmyDruid posted:Except you can't even run it a thousand times, because _the benchmarks get better over time_ because of the branch prediction learning thing. Is it really that good/powerful/effective or just worth a few percentage points in benchmarks? GRINDCORE MEGGIDO posted:Shame it can't scale details up or down depending on your refresh rate. That would be way rad.
|
# ? Apr 1, 2017 03:32 |
|
I don't know how much it is. It was mentioned on a recent PCPer podcast in passing.
|
# ? Apr 1, 2017 04:44 |
|
Yeah Wendell mentioned it on L1, apparently AMD advised running a bench 3-4 times before taking results but I have no idea if anyone actually did that.
|
# ? Apr 1, 2017 04:55 |
.... People seirously don't? That's kinda the whole point of experimentation. You want to test for precision!
|
|
# ? Apr 1, 2017 05:19 |
|
So I don't know if it was brought up or not, but ECC memory does actually work with Ryzen. Hardware Canucks did ECC memory testing and was able to get partial memory correction working in both linux and Windows. It did successfully recover from all single-bit errors, but not all ECC features were available.
SlayVus fucked around with this message at 05:47 on Apr 1, 2017 |
# ? Apr 1, 2017 05:26 |
|
Sata and nvme storage benchmarks http://www.tweaktown.com/articles/8073/amd-ryzen-ssd-storage-performance-preview/index.html
|
# ? Apr 1, 2017 11:05 |
|
GRINDCORE MEGGIDO posted:Sata and nvme storage benchmarks I loving knew it. The platform is poo poo. NVMe performace is nearly HALF Intel at low queue depths.. which is exactly what most desktop pcs will be seeing. loving AMD! Same poo poo, different platform.
|
# ? Apr 1, 2017 15:07 |
|
redeyes posted:I loving knew it. The platform is poo poo. NVMe performace is nearly HALF Intel at low queue depths.. which is exactly what most desktop pcs will be seeing. loving AMD! Same poo poo, different platform. AMD CPU and Platfrom Discussion:
|
# ? Apr 1, 2017 16:02 |
|
redeyes posted:I loving knew it. The platform is poo poo. NVMe performace is nearly HALF Intel at low queue depths.. which is exactly what most desktop pcs will be seeing. loving AMD! Same poo poo, different platform. Isn't there a lot of room for optimization driver-side?
|
# ? Apr 1, 2017 20:13 |
|
Its not like amd has had 4 years to get this stuff right.
|
# ? Apr 1, 2017 20:27 |
|
I wonder how much time they spent shoveling money into the bulldozer shaped hole.
|
# ? Apr 1, 2017 20:40 |
|
redeyes posted:I loving knew it. The platform is poo poo. NVMe performace is nearly HALF Intel at low queue depths.. which is exactly what most desktop pcs will be seeing. loving AMD! Same poo poo, different platform. I assume you're talking about this? quote:With CDM, there is very little difference in sequential performance between the two platforms; Intel still holds a slight edge. Intel holds a distinct advantage at 4K QD1 write, where it is 50% faster than AMD. I don't think you read it correctly. AMD is not 50% as fast as Intel, Intel is 50% faster than AMD. The numbers are 200MB/s for AMD and 300MB/s for Intel, which means AMD is two thirds as fast as Intel. Not saying your wrong to be disappointed, I don't really know the implications of this benchmark.
|
# ? Apr 1, 2017 20:46 |
|
Low queue depth is a thing for home users, but I'm not sure how often you'd be handling tons of tiny 4K files like that. I don't know either. Be interesting to see how it's handled on other motherboards too. Does NVME performance vary much on different Intel boards? GRINDCORE MEGGIDO fucked around with this message at 20:52 on Apr 1, 2017 |
# ? Apr 1, 2017 20:49 |
|
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.
|
# ? Apr 1, 2017 21:17 |
|
Do you have a source? I'm not doubting, I'd just like to know more about that.
|
# ? Apr 1, 2017 21:51 |
|
How exactly can you get NVMe performance wrong? Isn't it just shoveling data back and forth over PCIe?ufarn posted:Another benchmark showing 30% DX12 gains: Combat Pretzel fucked around with this message at 22:33 on Apr 1, 2017 |
# ? Apr 1, 2017 22:30 |
|
Combat Pretzel posted:How exactly can you get NVMe performance wrong? Isn't it just shoveling data back and forth over PCIe? That's what I thought. But I guess it's got to get from the bus to the cache and vice versa? Is that how it works?
|
# ? Apr 1, 2017 22:35 |
|
I would assume its still a driver issue at this point since they didn't even SATA drivers in that article to test with. If after a month or 2 of driver updates down the road performance is still lagging for NVMe SSD's then AMD probably screwed up somewhere in the hardware. That being said performance, while quite a bit worse than Intel at the moment, isn't actually bad for desktops. That is from a practical typical user perspective you might possibly see boot and program load times as maybe several seconds or so slower than on a modern Intel platform right now. That isn't good but calling it poo poo doesn't seem correct either. Its still a huge step up from their older socket AM3/FM3 platform SATA and I/O performance. Personally I've only messed with 1 AM4 system and while I didn't do any benchmarking it sure "felt" as fast and as smooth as any modern Intel system when loading up CS:GO, PoE, Libreoffice 5.3, and booting Windows 10 Pro.
|
# ? Apr 2, 2017 04:20 |
|
PC LOAD LETTER posted:I would assume its still a driver issue at this point since they didn't even SATA drivers in that article to test with. If after a month or 2 of driver updates down the road performance is still lagging for NVMe SSD's then AMD probably screwed up somewhere in the hardware. Didn't intel help develop the nvme standard, if so i would assume they would have better drivers for it.
|
# ? Apr 2, 2017 04:34 |
|
wargames posted:Didn't intel help develop the nvme standard, if so i would assume they would have better drivers for it. That's an excuse for 5 years ago when NVMe drives were first released. It's not an excuse that makes sense now. Even the latest revision in use is nearly 3 years old at this point.
|
# ? Apr 2, 2017 05:06 |
|
wargames posted:Didn't intel help develop the nvme standard, if so i would assume they would have better drivers for it. Given the way BIOS improvements are already making a impact on overclocking and memory support, as well as bug fixes, I think there is good reason to believe AMD is serious about and able to work out the current issues. Especially since the server platform Zen's should be launching within several months.
|
# ? Apr 2, 2017 06:18 |
|
Another comparison of Ryzens game performance when paired with AMD or NV GPUs: https://www.youtube.com/watch?v=nLRCK7RfbUg
|
# ? Apr 2, 2017 13:42 |
|
The NVMe thing mystifies me. I can't make sense of how it's possible. It's not like there's CPU/Platform-specific drivers for that, right? It should just be talking standard PCIe on that side. Like, is it possible AMD just has a general PCIe IO latency problem that has gone unnoticed? (That could also explain why games show worse performance than other kinds of applications...)
|
# ? Apr 2, 2017 16:45 |
|
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. SSDs have been getting faster every year, but it's made zero difference after the first year or so, because their main advantage over older disks isn't the raw speed, but the lack of seek time.
|
# ? Apr 2, 2017 16:49 |
|
|
# ? May 21, 2024 10:20 |
|
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 17:16 |