Team Group Vulcan G 1 TB SSD

[W1zzard]

can you test sn850?

I would love to, but can't seem to find a decent contact at WD

 

[InVasMani]

I wonder if TechPowerUp has considered testing how storage sensitive performance impacts micro hitching across a variety of games and how alarming a FPS impact it can have on actual game play on a cold boot depending on storage utilized HDD/SSD/NVME would even be good to test a x16 quad M.2 to a x4 M.2 as well. I would think also bar SIZE and resolution would impact things as well. I think you've got a situation of higher random I/O versus higher sequential data as you scale resolution from lower to higher. I would be curious to see how much FPS you lose on average and at 99% percentile and 0.1% percentile over a given time period within the first like 5 minutes to 15 minutes of game play perhaps in certain games?

I'm just curious how much potential performance on average on a cold boot you're losing by utilizing slower storage in some of these storage sensitive games. In fact I would think the stronger the GPU/refresh rate at the lower the resolution the more compounded the issue. I also would think the same holds true of SLI/CF in regard to lower resolution. The way I see it is that the higher the ceiling the further it can drop quickly and quick sweeping frame rate inconsistency can be major problem.

In terms of resolution I think what you've got is a random I/O versus sustained performance balance shift of storage resource usage as you scale resolution and frame rate in a given direction. If a higher overall refresh rate and frame rate allows you to travel from one section of game environment into a different one more quickly that leads to more random I/O access from storage. On the other hand higher resolution leads to slightly higher sequential data requirements per frame, but because of the overhead lower frame rates and/or refresh means you probably won't shift around locations as rapidly so overall random I/O access is a bit reduced at least to a point.

Consider this single GPU 60FPS/60Hz versus dual GPU 120FPS/120Hz or alternatively 240FPS/240Hz versus 60FPS 4K. Now that in mind does the storage itself complicate the matter more in one instance over the other!? Would a stronger GPU trying to maintain a higher peak refresh rate and FPS target relative to it's inherent strengths not falter more than the reverse from weaker storage? I see no reason why SLI/CF shouldn't be more sensitive to storage performance realistically if you set the bar higher on refresh rate and frame rate target.

I think the issue is when you raise the performance ceilings on FPS and refresh rate and especially true at lower resolutions to larger degree a storage hiccup in game that is random I/O sensitive a stall in performance can cause a larger drop off in performance on a stronger GPU. The bigger they are the harder the fall idiom comes to mind. When it comes to storage performance impact and assuming the same storage performance on a stronger system will have a greater drop off in performance than a weaker system due to the flow of data I/O from storage to the GPU. I think it's a subject really worth exploring just how much could your storage be holding back your GPU depending upon how the game loads and utilizes data.

 

[lexluthermiester]

I wonder if TechPowerUp has considered testing how storage sensitive performance impacts micro hitching across a variety of games and how alarming a FPS impact it can have on actual game play on a cold boot depending on storage utilized HDD/SSD/NVME would even be good to test a x16 quad M.2 to a x4 M.2 as well. I would think also bar SIZE and resolution would impact things as well. I think you've got a situation of higher random I/O versus higher sequential data as you scale resolution from lower to higher. I would be curious to see how much FPS you lose on average and at 99% percentile and 0.1% percentile over a given time period within the first like 5 minutes to 15 minutes of game play perhaps in certain games?

Actually, those are very interesting points. These answers are already known. The effect you describe can be observed on DRAMless SSD's and HDD's that are slow and have small DRAM cache. Ironically, modern HDD's with DRAM caches above 64MB, HybridHDD's that have both a DRAM cache and an SLC SSD cache and SSD's with DRAM are either unaffected by stutters in game due to drive access or the effect is so minimal as to be unobservable.

 

[InVasMani]

What I'd be interested in how long a time period it takes for a solid state device of different types to render 60 more frames in total than a HDD. I think it would be rather neat to know for types of games that are storage sensistive or prone you might say. Even in the case of as little as 24 frames it would be intriguing. Perhaps a 3 metrics 24 frames/30 frames/ and 60 frames with HDD being a base line it measures against.

 

[pexxie]

maybe the on/off toggle of their caches aren't exposed to the OS

The SLC cache is part of the SSD, and invisible to the OS

Actually I was looking into this again, and found this really pleasant explanation of that Windows setting: https://www.game-debate.com/news/25...-how-windows-write-caching-could-save-the-day

I don't have an SSD with actual proper dedicated SLC cache to verify it, but I just bought a Seagate BarraCuda 120 250GB, and writes are abysmal with write caching off. Although @W1zzard I saw you did mention a pseudo-SLC cache for the BarraCuda 120 1TB you reviewed. So perhaps this pseudo-SLC cache IS exposed to Windows, while the normal SLC cache in other SSDs you have reviewed aren't.

I guess it's NAND performance I'm actually looking at. I suppose the cheaper one makes things; the more corners one needs to cut.

 

[InVasMani]

Just get Primo Cache and and it'll be as good or better than Samsung Magician for all your storage drives. Avoid the lazy writes though. Writes speeds are always questionable. When you get a SSD/NVME focus on the random write speed I/O and queue depth performance. The read speed is easier and safer to cache and speed up. I highly recommend NTFS unit allocation size of a 2048bytes with compression enabled.

 

[W1zzard]

Actually I was looking into this again, and found this really pleasant explanation of that Windows setting: https://www.game-debate.com/news/25...-how-windows-write-caching-could-save-the-day

I don't have an SSD with actual proper dedicated SLC cache to verify it, but I just bought a Seagate BarraCuda 120 250GB, and writes are abysmal with write caching off. Although @W1zzard I saw you did mention a pseudo-SLC cache for the BarraCuda 120 1TB you reviewed. So perhaps this pseudo-SLC cache IS exposed to Windows, while the normal SLC cache in other SSDs you have reviewed aren't.

I guess it's NAND performance I'm actually looking at. I suppose the cheaper one makes things; the more corners one needs to cut.

Windows disk cache is not pSLC cache. Of course if you turn off the Windows write caching, performance will suffer, but it has nothing to do with pSLC.

"SLC cache" and "pseudo SLC" and "pSLC" are the exact same thing, just slightly different wording

 

[pexxie]

Windows disk cache is not pSLC cache. Of course if you turn off the Windows write caching, performance will suffer, but it has nothing to do with pSLC.

How do you know this? Do you know of any bedtime reading I could check out? According to this article by MS; that disk write cache policy is based on a query of the device itself: https://docs.microsoft.com/en-gb/wi...storage/querying-for-the-write-cache-property
I don't think there's such a thing as Windows disk cache, but I know there's Windows file caching, and can see that with RAMMap. This file caching is forced, regardless of the disk write cache policy setting. But the write cache policy looks to me to be device-dependent, which means it's controlling something on the device itself.

 

[W1zzard]

pSLC cache works on Linux and all other OS's too

https://www.i.u-tokyo.ac.jp/edu/training/ss/lecture/new-documents/Lectures/15-CacheManager/CacheManager.pdf

Read Ahead and Write Behind | Microsoft Windows Internals (4th Edition): Microsoft Windows Server 2003, Windows XP, and Windows 2000

Microsoft Windows Internals (4th Edition): Microsoft Windows Server 2003, Windows XP, and Windows 2000,2004, (isbn 0735619174, ean 0735619174), by Russinovich M., Solomon D.

flylib.com

 

[pexxie]

pSLC cache works on Linux and all other OS's too

https://www.i.u-tokyo.ac.jp/edu/training/ss/lecture/new-documents/Lectures/15-CacheManager/CacheManager.pdf

Read Ahead and Write Behind | Microsoft Windows Internals (4th Edition): Microsoft Windows Server 2003, Windows XP, and Windows 2000

Microsoft Windows Internals (4th Edition): Microsoft Windows Server 2003, Windows XP, and Windows 2000,2004, (isbn 0735619174, ean 0735619174), by Russinovich M., Solomon D.

flylib.com

Thanks, but I don't see any mention of the disk write cache property setting. I know in Windows it's found in device manager, and in Linux you can see it with hdparm (-W). Never mind though - I'll keep probing the web for clarity.

EDIT: This aint the holy grail, but sure is the clarity I was looking for: https://www.postgresql.org/docs/current/wal-reliability.html
It talks about the disk write-back caching from just above the first bullet point list, describing how to toggle the disk's on-board cache feature.

 

[W1zzard]

Very nice, just to clarify, this is not the SLC cache. The SLC cache is non-volatile. Also check out differences of NVMe vs the ATA/SCSI commands you listed

 

[Scour]

Like W1zzard said, WindowsWrite Cache, which works also with other kind of memory like USB-Flashdrives, SD cards etc., is completely different to the pseudo SLC.

pseudo SLC is a thing of the SSD-controller and the Firmware.

I have different SSDs with different kinds of Cache, the kind of Cache is on every system/Windows the same, so it´s built in in the SSD, not in Windows.