I saw this article today and it looks promising for folks looking for small form factor AMD solutions. AMD's apparently comparing its performance to the RX 480, but hopefully it's substantially lower power. I'm praying for low-profile dual-slot for a hackintosh build soon, since the 1650 isn't an option and the RX 560 is a bit underpowered for my taste. Anyway, I suppose we'll see Friday.
Rumor RX 5500 Might Be Coming Soon-Ish
- Thread starter reasons4reasons
- Start date
I'm curiously excited by these cards - very interested in seeing how a small RDNA part like that (with just a 128-bit memory bus!) would perform. Also, that die might be small enough to fit onto an AM4 package. Not that I think that would happen, but it would be sweet if AMD did that. And as you say, this should have the potential for an LP card, at least in a cut-down, low-clocked form. RX 480/570 levels of performance in an LP form factor would be amazing.
AMD announces Radeon RX 5500 Series with GDDR6 memory | VideoCardz.com
Featured, Radeon RX 5500, Radeon RX 5500 XT AMD Radeon RX 5500 gets 7nm Navi and GDDR6 memory Today AMD announces a new mid-range solution for desktop and mobile markets. The RX 5500 series will
videocardz.com
Fingers crossed that the cards are 75W
First of all, fingers crossed that anything in that post is correct. Beyond that, if they can reach those levels of performance at 75W, I'm sold.AMD announces Radeon RX 5500 Series with GDDR6 memory | VideoCardz.com
Featured, Radeon RX 5500, Radeon RX 5500 XT AMD Radeon RX 5500 gets 7nm Navi and GDDR6 memory Today AMD announces a new mid-range solution for desktop and mobile markets. The RX 5500 series will
Fingers crossed that the cards are 75W
AMD Radeon RX 5500(M): Navi-GPU für FHD‑Gaming mit Desktop und Notebook
AMD stellt mit Radeon RX 5500 (Desktop) und Radeon RX 5500M (Notebook) neue Grafikkarten der Mittelklasse für Spiele in 1080p-Auflösung vor.
www.computerbase.de
Hope it will be released soon.
So:
Have to say I'm disappointed, even if I see how this makes sense as a clear mass-market, money-making play from AMD. OEM systems far outsell the BYO retail market - but that doesn't help me as an enthusiast. Still, there's a long history of OEM parts and retail parts being specced differently even with the same name.
So my hope is that the retail RX 5500 series a) adjust the TBP to a more reasonable number closer to reality; b) is offered in a low-clocked 75W variant (RX 5400?); c) launches relatively soon, and d) is priced reasonably (= <$150).
- Current launch is OEM only, with no retail launch date given beyond "Q4"
- Total Board Power is reportedly 150W, though that sounds completely bonkers given that the same number for the RX 5700 is 180W
- On the other hand, their own slides list 12% more performance over the RX 480 at 30% lower power consumption - and the RX 480 was also a 150W TBP card
- 30% below 150W equals around 105W
- Though both power and performance may vary as OEMs can adjust clock speeds and similar specs.
- TBP for the RX 5500M is 85W with slightly lower clocks than the OEM desktop RX 5500.
- The currently presented "RX 5500 series" cards are not using a full enabled die. In other words the die has more than 22 CUs. Possiblity of non-OEM 5500XT with 24-26?
- Unlike the GTX 1650 the video encode/decode block is intact from the higher end cards in the series.
Have to say I'm disappointed, even if I see how this makes sense as a clear mass-market, money-making play from AMD. OEM systems far outsell the BYO retail market - but that doesn't help me as an enthusiast. Still, there's a long history of OEM parts and retail parts being specced differently even with the same name.
So my hope is that the retail RX 5500 series a) adjust the TBP to a more reasonable number closer to reality; b) is offered in a low-clocked 75W variant (RX 5400?); c) launches relatively soon, and d) is priced reasonably (= <$150).
Last edited:
It's up on AMD's site now, and (thankfully) it looks like it's a 110 watt TDP instead of 150. Maybe board partners can coax that down to 75 watts, but I'm not super hopeful. Makes some decisions about my next case a bit easier, at least.
It's not 110 watts instead of 150. As said by Valantar, 150W is TBP (Typical Board Power), not TDP (which is 110W).
You simply were not speaking of the same thing
Edit : my bad, the T in TBP stands for Typical, not Total, woops
Last edited:
Appreciate the correction! Less good news in that case, but I suppose we were never really counting on AMD to have good performance per watt in the first place.
Nevermind. Anandtech now says 110w TBP and AMD says 150w TBP... did they reverse?
if true 150w TBP is not good for a 7nm chip + GDDR6 with the performance they are stating. That's a regression compared to RX 5700 (180w TBP) and puts it behind Pascal & Turing in power efficiency. I'm betting there was a spec mix up / typo.
if true 150w TBP is not good for a 7nm chip + GDDR6 with the performance they are stating. That's a regression compared to RX 5700 (180w TBP) and puts it behind Pascal & Turing in power efficiency. I'm betting there was a spec mix up / typo.
Last edited:
Don't know, seems a lot for me too, compared to the 5700 ?
The AMD page really says 150W for TBP : https://www.amd.com/en/products/graphics/amd-radeon-rx-5500
AMD has a long history of having different SKUs and configurations for OEMs with naming very close if not identical to retail cards. As such, we can expect retail RX 5500 (and possibly 5500 XT) cards to arrive as promised, but CU count, clock speeds, memory amounts and speeds, and so on will undoubtedly vary. AnandTech also says AMD says
Which doesn't quite seem to align with the TBP numbers. Likely given the OEM-only nature of these cards they're a) playing it safe, and b) leaving OEMs plenty of rooms to configure these cards as they please.AnandTech said:
1.6x perf/W with a 12% average performance increase over the RX 480 means (going by TechPowerUp's 163W measurement for "Typical Gaming") typical gaming power consumption of ~110W. On the other hand AMD's slides specifically say these efficiency/performance scores are for the GPU (as in the chip only, not the whole card) which might mean the faster GDDR6 consumes more power than the GDDR5 of the 480 or some such. Still, this difference ought to be negligible. Also, the move from a 256-bit GDDR5 bus to a 128-bit GDDR6 bus ought to save quite a bit of power, really. 8GB of 256-bit GDDR5 consumes ~40W if not more, and bus width matters more for power consumption than the amount of memory.
Still, 150W for this card doesn't add up. Even if the GPU (chip only) TDP is 110W, TDPs are calculated in a lot of different ways, and does not necessarily reflect in-use power draw. And a slightly cut-down 75W card (or just a downclocked one) should be entirely within reach. Personally I would love a low-and-slow 22-ish CU RX 5400 for medium-to-high 1080p gaming at 75W and $100-120.
Edit: after reading more about the 85W RX 5500M I'm slightly more hopeful about this, even if that might be that I want this to turn out more efficient than it is. Still, 85W for a 4GB card with a 1448MHz game clock and purported 88% of the full-fat card's performance is not bad at all, even if these are dice binned for efficiency and low leakage. That could quite easily translate to a 95W desktop card with less intensely binned dice, or a 75W card with just a few less CUs. Could be interesting given that the stock RX 5500 is supposed to beat the GTX 1650 by 30-35%.
Last edited:
My bad, I don't mean Anandtech having a typo, I mean I think AMD's page is incorrect. 150w TBP is beyond conservative and doesn't line up with their message about improved efficiency. We'll see if they revise it or if 150w TBP is what they're sticking to for whatever reason. Regardless I think the actual cards will prove to draw much less in testing ?
Point is, would a downclocked Navi 14 even reach a 1650 Ti/Super? That's, on paper, today's most performing 75W GPU. An on par (both in performance and price) AMD RX 5400 is only useful for AMD enthusiasts and hackintosh (?) projects, and a less performing one is good only for the latters.
Also, considering the 7 nm architecture, you would expect better performance per Watt on Navi side, compared to Turing's 12 nms…
Which also points out the excellent job Nvidia has done on efficiency with Turing, if things stay as they seem.
Nvidia's efficiency lead hasn't changed much since Maxwell, so nothing new there. One would only expect AMD to lead on efficiency with their process advantage if one knew nothing about the architectures involved. Still, Navi does gain on Nvidia beyond the process advantage - which is about time, really.
As for performance matching a 1650 Ti/super: First off, the 1650 itself is 75W (and runs hotter/consumes more power than the 1050/1050 Ti), so Nvidia needs to cut clocks on an upcoming Super if it has more CUDA cores. Or they might just stick faster memory on it and leave everything else. Either way, there's not much Nvidia can do within 75W. Beyond that: The mobile RX 5500M purportedly performs at 88% of the desktop part, at 85W. On the other hand, the desktop part is purportedly 30-35% faster than the 1650, or to flip that around for comparable numbers, the 1650 is 74-77% the performance of the desktop RX 5500. At the very least a 75W version of this die should be as fast as the 1650, but I wouldn't be surprised if they could (marginally) beat it.
Considering how in AMD they compares their RX 5500 with Nvidia GTX 1650 and glorify the performance per Watt compared to Polaris, one would expect a 75W GPU performing substantially better than the GTX 1650. When they can really compare RX 5500 with GTX 1650 only because they will both sit on the same price range. While GPU is more on the range of GTX 1660 when you look at performance per Watt.
And I want to point out. It may seem I'm criticizing AMD with what they are doing with NAVI platform. Which is not. I'm actually planning my new PC to fit into soon-to-be-delivered Sentry 2.0. I've already settled on a Ryzen 3700X as the CPU of choice and I'm largely oriented towards RX 5700 XT/5800 XT, unless Nvidia Ampere will offer better performance per dollar (which I doubt, given Nvidia pricing over AMD) or will allow smaller GPUs with comparable performance, all in a relatively small time frame, won't wait a year or so…
Last edited:
I don't think you sound overly critical, I just disagree with some of your interpretations. I don't see how they're glorifying the perf/W compared to Polaris - after all, the RX 5700 series has shown just how much more efficient Navi is than Polaris, with the RX 5700 reaching roughly the same efficiency levels as the RTX 2080 and 2070S. The GTX 1660, -Ti and 1650 are 5-7% better in perf/W than those, which would seem to align with AMD's claim that the RX 5500 will deliver 1.6x more perf/W than the RX 480 compared to 1.5x for the RX 5700 series. These are reasonable numbers that align with third-party benchmarks, so "glorifying" is too harsh IMO. On the other hand, TPU's tables show the RX 5700 being 2x the efficiency of the RX 570 and 580 (the 400 series has sadly dropped off the table), which doesn't align well with AMD's 1.5x number. Still, expecting a small increase in perf/W over the RX 5700 series seems reasonable.
As for the comparison to the GTX 1650, I think you're right that price is the main reason for the comparison - with a 30-35% performance advantage some increase in power draw is a necessity. Navi isn't that efficient. Given the relatively high advertised clocks of the desktop RX 5500 (and the very reasonable clocks for the 85W mobile part) I don't see how AMD would have a problem making a downclocked 75W part, but given that they are just reaching GTX 16-series efficiency levels, it's hard to see them beating Nvidia at this wattage level. On the other hand, they might beat them in price, which is still a partial win, even if it's less significant.
Guess it's just me being disappointed because I set the expectation way too high with AMD releasing a 75W GPU with better performance than 1650/future 1650 Ti/Super (however they will call it) and same (or lower?) price point. ?
Similar threads
-
- Article
