MSI GeForce RTX 4080 Super Expert

[W1zzard]

MSI introduces a new premium design in their "Expert" lineup, showcasing an impressive all-metal cooler that competes with the Founders Edition looks. Priced at $1150, it offers a triple-slot, dual-fan cooling solution, albeit on the high side cost-wise.

Show full review

 

[wNotyarD]

@W1zzard page 3 has all its text from PNY Verto's review. I can only imagine how much text you would have to write by hand if not pasting from other reviews when doing about 10 different models for the same date.

 

[W1zzard]

Yeah will fix in the next hours, still a bunch of reviews that needs texts. All data and pics should be there though

 

[wNotyarD]

Yeah will fix in the next hours, still a bunch of reviews that needs texts. All data and pics should be there though

Pics are correct, yes.

 

[Deleted member 237269]

MSI GeForce RTX 4080 SUPER 16G GAMING X SLIM​

MSI GeForce RTX 4080 SUPER 16G SUPRIM X​

Reviews ?

EDIT: "Thank you for this review!” "I'd love to see those reviewed as well in the near future if possible"

 

[Dahita]

MSI GeForce RTX 4080 SUPER 16G GAMING X SLIM​

MSI GeForce RTX 4080 SUPER 16G SUPRIM X​

Reviews ?

You misspelled "Thank you for this review!" followed by "I'd love to see those reviewed as well in the near future if possible"

 

[TheDeeGee]

Very expert indeed, high temps, high noise, high price.

 

[Deleted member 237269]

Very expert indeed, high temps, high noise, high price.

Top of the line product.
But it does look good ! Right

 

[MarsM4N]

Dang, expected really better performance from the cooler. The design is really marvelous. Well, for some it might be an argument to dish out the extra $150.

Top of the line product.
But it does look good ! Right

It reminds me more of the Mercedes Diamond Grill.

 

[Chomiq]

Dang, expected really better performance from the cooler. The design is really marvelous. Well, for some it might be an argument to dish out the extra $150.


It reminds me more of the Mercedes Diamond Grill.

I wasn't expecting much based on their last golden card:

 

[lexluthermiester]

• Almost no performance increase over RTX 4080 non-Super
• Large price increase over MSRP
• Weakest cooler of all cards tested today
• Fans could be a little bit quieter

So, to sum up: Costs much more, is ugly as hell, makes more noise and does NOT perform much better. Did I miss anything?

Way to go MSI..

 

[Double-Click]

IDK what is going on with them lately in GPU land...

Didn't expect their lower tier cards to be so bad (Ventus 2x version of my card tanked in review, same w/ Super).
Certainly didn't expect something in this price range to be that bad as well...yikes.

 

[theouto]

 

[Sora]

wouldn't risk one of these with the uPI uP9512R and onsemi combo.

 

[mxthunder]

I thought this one looked cool during the CES videos. Had higher hopes for the cooling/noise. Why pay $150 more for no benefit over FE. Goes to show how hard it is to beat the nvidia design unless you go with one of the gross cheesy triple fan aftermarket cards.

 

[lexluthermiester]

Why pay $150 more for no benefit over FE.

Because the FE is kinda ugly? For the price though, you're not wrong. The PNY, Zotac or Palit cards are the better looking, better performing and better priced offerings.

 

[cbb]

is there a reason no one plots perf vs price in a scatterplot anymore? those used to be common ways to put gpu performance in context (waaay easier than jumping back and forth btwn lists of perf and price, and price/perf is a lot less clear to me for finding value in a price range)? the rel perf list could easily be plotted vs msrp (or, I suppose, one could link to amzn for mkt price but that's got it's own issues and more trouble than it's worth prolly)

Dividing it out, to get a price/perf ratio, is nearly a random list and the neither the price nor the perf is clear. Plotting perf vs price is a lot better, imho as it retains (and makes easily visible) both numbers.

(done kinda hurridly in Libre which still gives me trouble, the 7600 may not be right, 4k perf from above review, prices from newegg today, yadda yadda)

 

[W1zzard]

is there a reason no one plots perf vs price in a scatterplot anymore?

Too much manual work placing the labels. Is there no software that can properly take care of this for me?

 

[theouto]

Too much manual work placing the labels. Is there no software that can properly take care of this for me?

Perhaps one single excel file that is updated over time? A pain in the ass the first time, but after a bit you simply update as needed.

But I would imagine this would be the approach that first comes to mind, so apologies for being captain obvious.

 

[W1zzard]

Perhaps one single excel file that is updated over time? A pain in the ass the first time, but after a bit you simply update as needed.

But I would imagine this would be the approach that first comes to mind, so apologies for being captain obvious.

Not feasible in Excel because it can't figure out the label placement automatically

 

[cbb]

Too much manual work placing the labels. Is there no software that can properly take care of this for me?

both excel (easily) and librecalc autoplace the labels. I used excel since the dark ages, and it was always pretty trivial. I'm new to libre (retired in 2010 so I hardly use calc much now, tbh), but it still did it after about 5m of swearing Turns out it required the x axis to be sorted. I've attached my quick-and-dirty example so you can look at the settings to get it to look that way, but one could easily use make a template and reuse it. I'll try export to xls as well (I no longer have win or office) but idk how the export will look, or even if the forum will let me attach these. Even if not, you can do it, and it's only hard the first time (manually set data columns and data label vector).

yeah, forum blocked the .ods attachment, but the export to xls attached ok (idk if it exported well). The real key is the info: nearly any spreadsheet will do what you want, and autolabel pts correctly, but it takes a little familiarity w/ the program. Sorting by x value is often required or it confuses the graphing routines. It's a good trick to master anyway, imho! I used to use this all the time in the corpo world. Should be doable in 5-15m depending on skillz. Expect some foul language, as it's (exp in libre) a little non-obvious

Not feasible in Excel because it can't figure out the label placement automatically

nope, excel absolutely can autoplace the labels for u. Be sure the array is sorted by x values, and that the label vector is specified to be the fields u want. Did this all the time, for many years (no longer have xcel or win atm tho).

if you can't figure it out, I think my wife might have excel on her machine, I could probably send you an example. But, honestly, knowing it's possible, and the sorting req't, you'll probably figure it out faster yourself.

(my bad; might be sort data by y, not x. Was thinking x but actually did y and it worked w/o paying much attn, out of habit. May vary by program)

just did the same thing in google sheets, it's actually almost easier, although i've never used sheets before and found it kinda opaque and harder to customize. If I remember right, can even do it in gnuplot, but I haven't used that in ~35yrs.

https://docs.google.com/spreadsheets/d/1W_lW2zfqUTrFZmnUyCUBR7oFjr_3iPK-4dLoPtjrgrY/edit#gid=0

(not a fan, but first time rly using it and so may be lack of familiarity)

 

[W1zzard]

I think the label should be the card name, not the price? put in long names and you'll see what i mean with "label placement problem"

 

[cbb]

I think the label should be the card name, not the price? put in long names and you'll see what i mean with "label placement problem"

View attachment 333642

well, w/ unnecessarily long names u simply have more text than will fit if the graph size is small; that's just bad layout. There are several ez ways to mitigate that, several of which make it less work. A trickier-to-figure-out-one, but nice, is to actually get it to treat (as an example) nv and amd as sep series on the same scales (totally doable), so you've got red/green dots instead of more text; that eliminates a bunch of symbols. Clearer, too. And, rly, there's no need for the long name (except, perhaps, for the card being reviewed). U can use a larger plot area, set auto wrap labels on, smaller font (ew!), or manually drag a few labels (there's a setting for auto connector lines). But it's just basic graph layout; you put a fair amount of redundant text on there, and unnecessary text, and a lot of pts, in a small plot; none which is a s/w issue. That's just a bunch of less than ideal user decisions stacking.
Rly, ur gonna have 1 particular model ("Strix Purple Pokimon 9001 OC Smurf Edition!" or whatev) and a bunch of reference pts (and I'd argue "3090 Ti" is clear enough, but if u rly want it could be "4070 12GB"...except all 4070 are 12GB so it's extraneous clutter. There should be no unnecessary symbols on a plot or chart, that's just basic. None of that is a s/w limit, it's just chart design. And you can tweak the offset and other defaults, and manual drag a problematic label or two (like the aforementioned silly example card name), but I'd argue if ur doing it right you won't need to do any of that.

Yes, model should be label, my export may not have come thru intact, u can see my example above has model as label.

You don't need RTX, ever; the model numbers are clear to anyone who reads this site or cares. You don't need the ram, except in the case where you're exploring ram variant sufficiency in which case only a couple models will need the extra text. You can treat the mfgs as seperate series, although that's slightly trickier the first time you do it. Etc. Rly, this is a discussion of "how to present data" not an excel issue. And, if u want that many pts, prolly make it a half page or full page graph (I err on the side of clarity after a long career presenting data to execs and other children )

 

[W1zzard]

well, w/ unnecessarily long names u simply have more text than will fit if the graph size is small; that's just bad layout. There are several ez ways to mitigate that, several of which make it less work. A trickier-to-figure-out-one, but nice, is to actually get it to treat (as an example) nv and amd as sep series on the same scales (totally doable), so you've got red/green dots instead of more text; that eliminates a bunch of symbols. Clearer, too. And, rly, there's no need for the long name (except, perhaps, for the card being reviewed). U can use a larger plot area, set auto wrap labels on, smaller font (ew!), or manually drag a few labels (there's a setting for auto connector lines). But it's just basic graph layout; you put a fair amount of redundant text on there, and unnecessary text, and a lot of pts, in a small plot; none which is a s/w issue. That's just a bunch of less than ideal user decisions stacking.
Rly, ur gonna have 1 particular model ("Strix Purple Pokimon 9001 OC Smurf Edition!" or whatev) and a bunch of reference pts (and I'd argue "3090 Ti" is clear enough, but if u rly want it could be "4070 12GB"...except all 4070 are 12GB so it's extraneous clutter. There should be no unnecessary symbols on a plot or chart, that's just basic. None of that is a s/w limit, it's just chart design. And you can tweak the offset and other defaults, and manual drag a problematic label or two (like the aforementioned silly example card name), but I'd argue if ur doing it right you won't need to do any of that.

yeah .. what you're actually saying is that there is no automatic label placement It's a well-known problem in computer science: http://www.eecs.harvard.edu/shieber/Biblio/Papers/tog-final.pdf

 

[cbb]

yeah .. what you're actually saying is that there is no automatic label placement It's a well-known problem in computer science: http://www.eecs.harvard.edu/shieber/Biblio/Papers/tog-final.pdf

nah, I'm saying it's unreasonable to ask software to compensate for arbitrarily unwise user choices. If you make halfway decent choices, consistent with sensible practices on how to present data, it works just fine. I'm trying to find a polite way to say it's not the software's fault that a user may not know how to present information, or how to use the software. W/ reasonable choices, auto labeling has worked well enough to be useable w/ little or no tweaking for ages. Can you exceed its limits? absolutely. No one who would try to put all those *very obviously* extraneous chars on a plot is in a position to opine on this. Even if it managed to fit all those pointless chars on the page, I'd have sent you back to redo it in a work env. Training new employees on the utter basics of presenting data always took the first month, but it is entirely a teachable skill. Every time I got a new employee I'd have to train them in basics of chart-making (and research, and a lot of other topics). This isn't a software issue, imho.