UHS-II Cards Reviewed

Started 1 month ago | Discussions thread
ForumParentFirstPrevious
Flat view
RLight Senior Member • Posts: 1,352
UHS-II Cards Reviewed
11

With the new EOS M6 Mark II looming and it's smaller buffer but, UHS-II support, it's time for me to finally make the jump from UHS-I to UHS-II.

However, after I got my recent Prograde Digital 64GB UHS-II v90 card (top of the line) and under certain circumstances got more favorable results with my existing SanDisk Extreme Pro (UHS-I), I quickly realized something was up and a deep dive on the subject was in order

.

Long read ahead, highly technical, but should be informative not just for EOS R owners, future EOS M6 Mark II owners, but also future camera's sporting UHS-II.

.

To start, I did some basic reading on reviews for UHS-II cards, and came across Prograde Digital, they apparently are formed from the ashes of the old Lexar, not to be confused with the new Chinese-owned Lexar. I'd heard good things so I figured I'd give them a shot considering Lexar has been cream of the crop in the past.

.

To my dismay, my first shooting with the Prograde Digital 64GB (indoors), I found it did deliver much faster buffer clearing times than my SanDisk UHS-I card, but, when driving AI-servo with C-RAW, the SanDisk kept chomping away on a buffer-full condition in a few, ka-chunk, ka-chunk, ka-chunk, vs the Prograde paused on full buffer and then after a bit rattled off a burst of high speed and paused again, repeat. I preferred the former and wondered if perhaps another UHS-II card might give me more SanDisk-like performance (more consistent) but faster FPS rate.

.

After a deeper dive online I settled on a couple options:

The "real deal" Lexar 2000x UHS-II (to rule out perhaps the Prograde Digital wasn't so good)

Toshiba "N502" (fastest random write 512KB speed per online benchmarks)

Sony "Tough" SF-G64T/T1 (fastest random write 4K IOPS QD per online benchmarks)

Transcend "700S" 64GB UHS-II (same random write 4K IOPS QD as my existing SanDisk, but, UHS-II)

The Prograde Digital UHS-II V90 64GB I already had

.

I already had a SanDisk SDDR-399 reader which I used exclusively for the testing, even though the Lexar came with a reader, I found this guy was a bit better for whatever reason on my Mac. I did test it a few times, but, did not use any of the data from it for the results.

I'll be testing on my 2014 15" MacBook Pro 2.5GHz 16GB config with the left USB 3.0 port (it's not hanging off the bridge, which the right USB port is, avoiding a potential bottleneck). Latest Mojave update as of the time of this thread.

Lastly, I'll be running this with my EOS R with latest firmware as of the time of this thread.

.

Day One

.

My Transcend arrived before the Lexar Sony or Toshiba. I decided to do my walking and drag along the SanDisk Extreme Pro UHS-I I had, the Prograde Digital and Transcend with my EOS R and RF 35mm.

.

First up was the Transcend in AI-Servo with my usual C-RAW and Small JPEG combination in high speed continuous drive. After some fumbling with settings and different configurations I settled on Fv mode and center cross AF, not that it matters for this test. I ran it through twice and averaged the result.

.

122.5 Frames before full buffer reached (Transcend)

9 seconds to buffer clear

Next I ran it in Single Shot Mode for highest FPS

64.5 Frames before full buffer reached

.

Next I tested on my baseline, the "old" SanDisk Extreme Pro 256GB UHS-I card I had

60 frames before full buffer reached (AI-Servo)

15 Seconds to buffer clear

49 frames (single shot)

.

Clearly UHS-II was delivering here with double the frames before buffer in AI-Servo and a 31.6% jump in single shot (64.5/49). Yay! That alone is very substantial. Clearing the buffer much faster is big potatoes too. Also, I noted I was not getting the consistent ka-chuck, ka-chunk, ka-chunk with the SanDisk today. As it would turn out this is probably due to in fact the outside lighting and detailed subjects producing larger files. The Transcend on the other hand was producing near consistent results in the same conditions except much better, more like, ka-chuck ka-chunk, ka-chunk ka-chuck... ka chuck chunk chunk chunk chunk... Much more responsive than the fast UHS-I SanDisk. Double yay. So, my whole consistency thing was out the window; that was probably due to comparing the Prograde in my backyard (well light) against the Sandisk (indoors)... Not a fair fight and hence I was operating on a false presumption as the SanDisk was dealing with smaller files. But, the Transcend seemed more responsive than the Prograde interestingly enough from the day before which I'd also testing outside.

.

Next, I repeated this on the Prograde during the same loop I was walking:

120.5 Frames before full buffer reached (AI Servo)

9.5 seconds to buffer clear

66.5 (Single Shot)

.

If you note above, the Transcend wins in AI-Servo but looses in Single Shot. I realized after, that the conditions I was shooting shifted just enough that perhaps this wasn't the best way to test these cards IE controlled conditions needed due to how close the results were for a clear winner. However, I did "feel" the Transcend was more responsive (vs the Prograde today) particularly after buffer-full had been reached it gave more consistency with response which troubled/puzzled me as this was the less expensive card.

.

Back at home I did some Black Magic benchmarks as a result to give some synthetic data to explain the real life results, but I'll talk more about those later as that gets into some interesting weeds.

.

Day Two

.

I should mention my kids got the flu of some sort. Housebound for testing today...

However, I made some good use of the time as the Lexar and Sony arrived. The Toshiba was stuck in Ohio and was running a day late. Wouldn't arrive till tomorrow. A shame as I had high hopes for him.

Due to the time intensity of testing, I figured I might as well get started and finish the Toshiba tomorrow huh?

I ran both the Lexar and Sony indoors with AI-Servo with CRAW and Small JPEG only as I figured out that the best benefits of a fast UHS-II card are not at full FPS (31.6% benefit), but at the slightly slower FPS AI-Servo, where you'll likely use it anyways in a long buffer; the slower FPS seemed to give the UHS-II time to pull data away from the buffer to increase the time to full vs the faster single shot, quickly bled away the buffer much faster than the UHS-II cards I had could keep up with a reasonable increase and was more reasonable use case on where I'd use it on the EOS M6 Mark II (in AI-Servo with the real time tracking ala 7FPS).

.

Results (Indoor testing):

.

247 Frames (Lexar)

10 seconds to buffer clear

.

253 Frames (Sony)

7 seconds to buffer clear

.

I did learn my lesson from the day before; this was a controlled test and the Sony didn't just pull ahead, it felt more responsive after buffer full. In fact, it shot in this case in near-real time even after the buffer was full! The Lexar did not. Clearly with the smaller file sizes from indoor shooting coupled with the sheer speed of this card, it was a break even. Pretty amazing result. The Lexar is pretty good too here btw. 247 frames, is a long buffer at 5FPS.

.

I had to pause my testing to care for the kids and run some more benchmarks on the computer which I'll be getting to shortly.

.

Next I ran the Transcend against the Sony directly as I felt these both were the more responsive cards. Different subject, controlled, indoors.

.

163 Frames to full (Transcend)

161 Frames to full (Sony)

.

Truth told, the Transcend got another frame or two in beyond full because the pause was so short to signify full buffer, so you could say it was 161 or 162. That's how responsive these cards are in buffer full condition and hence the breakout of them against each other as it would come down to these two in my estimation.

.

So I haven't been talking about synthetic benchmark data up till this point as all the testing I'd done thus far produced erratic results from Black Magic's tool for MacOS for disk benchmarking. So, I turned to Xbench, and oldie but goodie that gives you more data. However, it uses a fixed much smaller data sampling so I had to run it over and over and over. I collected and aggregated and averaged the data (that took a while) and still had more questions than answers here. The Transcend was scoring higher in Xbench by a bit in weighted score, but, some key metrics it was more around a lot. It was the most reliable of the bunch for producing consistent results in both Black Magic and Xbench though which was notable. This will make more sense a bit later on.

.

It occurred to me looking at the Xbench results and Black Magic that card-warming was influencing results here after some causality analysis and former experiences with doing RAW-video ala my 5D Mark III back in the day where a 10 second warm-up option was available in MagicLantern that would REALLY help with demanding 1080-RAW footage that wanted 90MB/sec, immediately, and the KomputerBay (Lexar knock-off) I had at the time did 100mb/sec, but needed warm-up to hit it. I figured out, warm-up was impacting results. Sure enough, warmer cards performed better. The Sony in particular scored abysmally in Xbench when "cold" but warm, it scored better than the others. It "cooled" faster btw (the Sony) in comparison to other cards. I also noted it's thicker than the other cards and is a tight squeeze in the EOS R so it may have an internal heat sink of sorts I gather. This makes some sense a bit later comparing the Transcend which had similar performance when warmed up but didn't seem to suffer the warm-up as much as the Sony did.

.

Day Three

.

I still wasn't getting good solid results in those synthetic benchmarks and these are really close calls. It was apparent in testing that the Prograde and Lexar were neck and neck in the same vein the Transcend and Sony were neck and neck. I noted, both the Prograde and Lexar had Made in Korea on the cards vs the Transcend and Sony had made in Taiwan on the cards. I'm betting Samsung is providing the NAND for the former and TSMC is providing NAND for the latter and hence the similar results of those pairs.

.

Finally after long wait the Toshiba arrived. I did a quick test on the same indoor scene I did the Transcend vs Sony the day before and got around 163 frames after matching exposure around the same time of day. It felt very responsive though. It felt more like a Transcend or Sony, less like the Prograde or Lexar. And sure enough it's made (Toshiba) in Japan

.

Xbench and BlackMagic REALLY liked the Toshiba though. Unlike the other cards, it just smashed BlackMagic and Xbench out of the ballpark. Rock solid performance. 180MB writes, minimum, time and time again, with usually better. Xbench was 70 or more weighted score with the sequential writes never dipping beneath 150/135mb/sec vs the Sony touched 44mb/sec often when "cold" in Xbench. I noted this card was advertised as being 8K ready, and Black Magic concurred... It's the only card to achieve the 2160/60 rating. Hmmm. Interesting.

.

I still wasn't getting reliable data out of Xbench and Black Magic though. Me being computer oriented for my employment, I decided I'd leverage some command line judo to help here. MacOS comes with the CP command, however, even though the MAN page says -V option (verbose) is supported, the documentation is out of date. It's not in Mojave. Cute (not really). So a bit of Googling brought me to the rsync utility with -ah and --progress options.

However, although I got pretty consistent results, turns out the consistency of the results had nothing to do with the cards, but rather some internal limit of the rsync utility... I wasn't passing 81mb/sec. Things would hit 84.18MB/sec average, 79.07, 70.53, just depending what minute I ran the test I guess. So that was out. It could not deliver the goods. I chuckled to myself internally when I noted in Black Magic's release notes that updated the UI to take advantage of faster cards recently and realized you need code just for this kind of thing.

Well, bit of Google lead me to something I'd never heard of before that sounded promising. AJA System Test Lite. It could do 16GB segments vs 5GB. Surely with that much raw data, whatever fluctuations were going on would be averaged out. And guess what? They were. Now I was getting real, repeatable results that echoed my theory of warm-ups. Warm-up was VERY real. And, I was getting good data now and a good way to warm up my cards for Xbench and Black Magic if I wanted too I might add. It also provided my first thermal throttle too. The Transcend and Sony performed on the same tier, with the Sony edging out the Transcend in RAW speed, but, the Sony took longer to warm up (ha, that's what it seemed like, and it's true, at last, I can prove it!), but, the Sony also had more endurance. Whereas after about 10 minutes of grueling again and again hitting the transcend with 16GB tests, it finally started to dip in performance down to 135mb/sec, the Sony under the same stress kept it's cool, so to speak. Interesting. You have a doubled edged sword; the Sony with what I'd guess to be an internal heat sink making it thicker, takes longer to warm up, is faster to cool down, but, can take the heat vs the Transcend is faster to warm up, slower to cool down, but, after intense sustained activity started to throttle. Interesting. The Lexar and Progrades did not throttle either btw, but, they also didn't hit speeds as fast as the Transcend, Sony or Toshiba either. To the metrics...

.

Benchmark results:

.

Prograde Digital SDXC UHS-II V90 64GB

Xbench:
264.87MB/sec sequential 4K write

181.09MB/sec sequential 256K write

8.20MB/sec sequential 4K read

219.37MB/sec sequential 256K read

.
1.42MB/sec Random 4K write

52.06MB/sec Random 256K write

6.77MB/sec Random 4K read

144.21MB/sec Random 256K read

AJA:
207MB/sec AJA Write

253.5MB/sec AJA Read

Black Magic:
180.6MB/sec BlackMagic Write (Average)

243.8MB/sec BlackMagic Read (Average)
163.1mb sec cold write (5gb BM)

238.5mb sec cold read (5gb BM)

.
195mb/sec write 1GB cold AJA

Made in Korea

.

SanDisk Extreme Pro 256GB SDHC UHS-I

Xbench:

93.46MB/sec sequential 4K write

73.74MB/sec sequential 256K write

7.04MB/sec sequential 4K read

92.43MB/sec sequential 256K read

.

2.58MB/sec Random 4K write

43.28MB/sec Random 256K write

5.40MB/sec Random 4K read

68.35MB/sec Random 256K read

.

AJA:

85MB/sec AJA Write

94MB/sec AJA Read

Black Magic:

84.8MB/sec BlackMagic Write

93.9MB/sec BlackMagic Read

Transcend 64GB SDXC/SDHC 700S

Xbench:
263.58MB/sec sequential 4K write

206.86MB/sec sequential 256K write

9.04MB/sec sequential 4K read

220.28MB/sec sequential 256K read

.
1.62MB/sec Random 4K write

58.50MB/sec Random 256K write

8.61MB/sec Random 4K read

153.87MB/sec Random 256K read

AJA:
223MB/sec AJA Write

260MB/sec AJA Read

Black Magic:
191.6MB/sec BlackMagic Write

245.4MB/sec BlackMagic Read

.
179.0mb sec cold write (5gb BM)

238.7mb sec cold read (5gb BM)

210mb/sec write 1GB cold AJA
Made in Taiwan

Sony Tough High Performance 64GB SDXC UHS-II (SF-G64T/T1)
Xbench:
161.36MB/sec sequential 4K write

126.57MB/sec sequential 256K write

14.20MB/sec sequential 4K read

191.66MB/sec sequential 256K read

.
4.62MB/sec Random 4K write

19.27MB/sec Random 256K write

3.26MB/sec Random 4K read

129.28MB/sec Random 256K read

AJA:
225.5MB/sec AJA Write

261.5MB/sec AJA Read

Black Magic:
198.2MB/sec BlackMagic Write

250.2MB/sec BlackMagic Read
204mb/sec write 1GB cold AJA
Made in Taiwan
Lexar Professional 2000x 64GB SDXC UHS-II

Xbench:
131.45MB/sec sequential 4K write

161.26MB/sec sequential 256K write

9.83MB/sec sequential 4K read

210.55MB/sec sequential 256K read

.
1.40MB/sec Random 4K write

23.67MB/sec Random 256K write

6.48MB/sec Random 4K read

140.67MB/sec Random 256K read

AJA:
198.5MB/sec AJA Write

258.0MB/sec AJA Read

Black Magic:
181.7MB/sec BlackMagic Write

231.2MB/sec BlackMagic Read

.
179.5mb sec cold write (5gb BM)

231.7mb sec cold read (5gb BM)

.
194mb/sec write 1GB cold AJA
Made in Korea
Toshiba SD 64GB Exceria Pro N502 UHS-II

Xbench:
167.48MB/sec sequential 4K write

132.56MB/sec sequential 256K write

13.04MB/sec sequential 4K read

216.31MB/sec sequential 256K read

.
1.55MB/sec Random 4K write

60.94MB/sec Random 256K write

10.85MB/sec Random 4K read

143.46MB/sec Random 256K read

AJA:
209.5MB/sec AJA Write

238MB/sec AJA Read

Black Magic:
221.9MB/sec BlackMagic Write

238.5MB/sec BlackMagic Read

.
214.7mb sec cold write (5gb BM)

238.9mb sec cold read (5gb BM)

.
203mb/sec write 1GB cold AJA
Made in Japan

.

Side note about Xbench results:

I've provided the Xbench results, however, I'd hone in only on things that are obvious like the Sony scores quite higher in the random 4K write and the Toshiba just scores well due to it's consistency. Otherwise I wouldn't put a lot of stock in Xbench data. Even after running it 4-6x and averaging it, the data is just all over the board. But I've made it available nonetheless, just don't put too much stalk in it. I'd have a harder look at the AJA and Black Magic Results which were much more repeatable and more pertinent.

Findings and Recommendations:

Oh boy, that was a lot to read for you, write and test and study for me.

So, let me break this down in a simple fashion:

I did some testing on my EOS R to figure out what it's buffer is, it's around 1GB-ish of RAW data (which matches the Manual's data on page 609). Not including JPEG. In fact, I found when enabling JPEG, I could have 1.5GB of combined RAW+JPEG data vs 1.1GB of RAW data only, and it didn't impact the number of frames captured. It would appear the EOS R (and probably the also DIGIC8, UHS-II equipped EOS M6 Mark II) is RAW-buffer limited and perhaps generates JPEGs after during buffer flush processing? Dunno. That's a guess, but it seems logical. Either way, the buffer confinement is determined solely by RAW data, not RAW+JPEG which reflects in the new EOS M6 Mark II Manual Page 579 and my experiences with testing the EOS R.

.

The EOS M6 Mark II on the other hand appears to have a buffer somewhere around 700-800MB based off the manual's stated data size estimates.

.

Why do I mention this?

If you note, I underlined the AJA performance for 16GB writes when warmed up, and, the 1GB cold writes. These are the key metrics you need to watch for as they govern how the card will operate "cold", that is just firing up the camera, and after having not taken a shot or video for a while, and all the sudden rattling off a full buffer, how fast that card will react to sudden influx of data and impact how long you can shoot before buffer fill. The 1GB metric fit this best since the EOS R is 1GB, and, the EOS M6 Mark II is closest to 1GB. The warmed up 16GB write impacts sustained shooting sports pros will encounter where that card is going to stay warmed up for lots of bursts.

.

The best "cold" performers are the Transcend, Sony and Toshiba in that order for 1GB threshold. I can tell you from watching the AJA and Black Magic benchmarks, the Toshiba hits nearly 180mb out the gate, but, due to it's max speed being less, gets eventually outrun around the 1GB mark by the other two. This is important for the EOS M6 Mark II and other cameras beneath 1GB buffers where the Transcend and Toshiba should be looked at more closely where you need that card to react, fast. Acceleration matters just as much as top speed does.

For cameras with more buffer either now or later like the EOS R and others, the Sony is a clear winner with definite repeatable slightly faster top speed with excellent cold performance, and, notably better endurance. Sports shooters should have a hard look at the Sony, as should long 4K video shooters as I can't remember if it's the EOS M6 Mark II or one of the new PowerShots, but, there's a note about heat issues with the card itself and the camera will throttle itself and even stop if you don't let it cool down. This is a real thing ironically for the card itself.

For others though, that are more like myself that are interested in the EOS M6 Mark II for more prosumer shooting where you may need to catch a fleeting moment and don't plan to do it over and over without some small cool down / pause, the Transcend makes sense with it's slightly faster warmup and slightly slower top speed (vs only the Sony), plus it's much more economical which the EOS M6 Mark II likely will be cheaper than most other UHS-II shooters out there in the future I'd bet.

The Toshiba is worth mentioning for 8K shooters / 4K/60 shooters. It's the only 8k certified card of the bunch, and Black Magic agrees; it just hits the metrics again and again, but, doesn't have quite the top speed of the Sony. It's also the most expensive of the bunch. It doesn't get as hot though in observation so my guess is the NAND in here is a different grade. It may possibly give slightly better results than the Transcend or Sony in the EOS M6 Mark II btw due to it's 700-800MB buffer, but it's going to be really close for 3x the cost of the Transcend and 1.5x the cost of the Sony.

The Prograde is worth a mention if you like the secret sauce for lack of better words of the Lexar of old. It performs like the modern Lexar, but a bit better. You don't have to worry about overheating throttling (I didn't experience any), but, it's not as fast either. It's much cheaper than the slightly faster Toshiba and slightly cheaper than the modern Lexar as of the time of this thread. The seller is American. It came straight from the seller so I can vouch to it.

.

Which will I do?

Oh boy, Transcend and/or Sony. In that order. I haven't decided, even though I have all the facts in front of my screen and did the testing myself. Choices, choices...

 RLight's gear list:RLight's gear list
Canon G1 X III Canon EOS M6 Canon EOS R Canon EF-M 22mm f/2 STM Canon EF-M 11-22mm f/4-5.6 IS STM +3 more
Canon EOS 5D Mark III Canon EOS M6
If you believe there are incorrect tags, please send us this post using our feedback form.
ForumParentFirstPrevious
Flat view
ForumParentFirstPrevious
Keyboard shortcuts:
FForum PPrevious NNext WNext unread UUpvote SSubscribe RReply QQuote BBookmark MMy threads
Color scheme? Blue / Yellow