• Welcome to TechPowerUp Forums, Guest! Please check out our forum guidelines for info related to our community.

AMD Readies Three RX 5500 Series and Two RX 5300 Series SKUs Based on "Navi 14"

Joined
Jul 9, 2015
Messages
1,956 (1.22/day)
System Name My all round PC
Processor i5 750
Motherboard ASUS P7P55D-E
Memory 8GB
Video Card(s) Sapphire 380 OC... sold, waiting for Navi
Storage 256GB Samsung SSD + 2Tb + 1.5Tb
Display(s) Samsung 40" A650 TV
Case Thermaltake Chaser mk-I Tower
Power Supply 425w Enermax MODU 82+
Software Windows 10
Where's big Navi?
Makes no sense to develop big Navi on DUV, and EUV isn't mature enough.
AMD has scarce resources and it shouldn't waste them on BS.
People into $1000+ GPUs can take a deeper breath.
 
Joined
Dec 29, 2014
Messages
749 (0.42/day)
I've seen reported power figures down at 55W (so total board power <65W) from a 5700XT when the GPU load is low enough that it can afford to clock down to sub 1200MHz speeds.
Wondering what the performance hit is. Not many people willing to pay $$$ for fast hardware only to make it slow to save power.
 

bug

Joined
May 22, 2015
Messages
6,718 (4.08/day)
Processor Intel i5-6600k (AMD Ryzen5 3600 in a box, waiting for a mobo)
Motherboard ASRock Z170 Extreme7+
Cooling Arctic Cooling Freezer i11
Memory 2x16GB DDR4 3600 G.Skill Ripjaws V (@3200)
Video Card(s) EVGA GTX 1060 SC
Storage 500GB Samsung 970 EVO, 500GB Samsung 850 EVO, 1TB Crucial MX300 and 3TB Seagate
Display(s) HP ZR24w
Case Raijintek Thetis
Audio Device(s) Audioquest Dragonfly Red :D
Power Supply Seasonic 620W M12
Mouse Logitech G502 Proteus Core
Keyboard G.Skill KM780R
Software Arch Linux + Win10
Makes no sense to develop big Navi on DUV, and EUV isn't mature enough.
AMD has scarce resources and it shouldn't waste them on BS.
People into $1000+ GPUs can take a deeper breath.
Are you saying one node advantage is not enough for AMD to equal Nvidia on performance? :roll:
 
Joined
Feb 20, 2019
Messages
299 (1.08/day)
System Name PowerEdge R730 DRS Cluster
Processor 4x Xeon E5-2698 v3
Cooling Many heckin screamy bois
Memory 480GB ECC DDR4-2133
Video Card(s) Matrox G200eR2
Storage SD Card. Yep, really no other local storage.
Display(s) It's probably a couple of boring Dell Ultrasharps and a sacrificial laptop.
Case 39U 6-rack server room with HEVC and 44KVA UPS
Mouse Maybe
Keyboard Yes!
Software ESXi 6.5 U3
Benchmark Scores I once clocked a Celeron-300A to 564MHz on an Abit BE6 and it scored over 9000.
Wondering what the performance hit is. Not many people willing to pay $$$ for fast hardware only to make it slow to save power.
None. There is no performance hit.

I'm talking about when the GPU utilisation is only 50%, like for instance when you are running with vsync and the card is comfortably delivering frames at the maximum refresh rate of your monitor with plenty of time spare between each frame.

In this case, the GPUs throttle down to much lower clockrates, and I was saying that below about 1200MHz, Navi runs at very low voltage (750mv) and is an efficiency marvel. It's only once you start pushing beyond about 1600MHz that the power usage truly skyrockets.


YMMV but I think most AMD cards since Polaris have been very good candidates for undervolting for huge power savings. You can choose whether to spend those savings on higher clocks (because most AMD cards are power-limited) or keep clocks the same and choose the reduced heat/fan noise.
 
Joined
Dec 29, 2014
Messages
749 (0.42/day)
I was saying that below about 1200MHz, Navi runs at very low voltage (750mv) and is an efficiency marvel. It's only once you start pushing beyond about 1600MHz that the power usage truly skyrockets.
You were saying AMD could make 35w TDP laptop cards by utilizing this scheme. If you *keep* the max clock at <1200MHz you aren't utilizing the card's potential and are limiting it for power reasons. That's an inefficient way to acheive low power vs what Nvidia does.
 
Joined
Aug 8, 2019
Messages
64 (0.60/day)
You were saying AMD could make 35w TDP laptop cards by utilizing this scheme. If you *keep* the max clock at <1200MHz you aren't utilizing the card's potential and are limiting it for power reasons. That's an inefficient way to acheive low power vs what Nvidia does.
Isn't that exactly what NV does? They limit clock speed, voltage, and often snip off parts of the core to make mobile versions and they really do it on the Max-Q versions.

Even on the Quadro versions they gimp the GPUs with cuts to core complexity and clock speed.
 
Joined
Feb 20, 2019
Messages
299 (1.08/day)
System Name PowerEdge R730 DRS Cluster
Processor 4x Xeon E5-2698 v3
Cooling Many heckin screamy bois
Memory 480GB ECC DDR4-2133
Video Card(s) Matrox G200eR2
Storage SD Card. Yep, really no other local storage.
Display(s) It's probably a couple of boring Dell Ultrasharps and a sacrificial laptop.
Case 39U 6-rack server room with HEVC and 44KVA UPS
Mouse Maybe
Keyboard Yes!
Software ESXi 6.5 U3
Benchmark Scores I once clocked a Celeron-300A to 564MHz on an Abit BE6 and it scored over 9000.
Isn't that exactly what NV does? They limit clock speed, voltage, and often snip off parts of the core to make mobile versions and they really do it on the Max-Q versions.

Even on the Quadro versions they gimp the GPUs with cuts to core complexity and clock speed.
Yep. This man gets it.

I'm simply speculating mobile TDPs based on existing data of Navi operating at low clock, low voltage.
 
Joined
Dec 29, 2014
Messages
749 (0.42/day)
Isn't that exactly what NV does? They limit clock speed, voltage, and often snip off parts of the core to make mobile versions and they really do it on the Max-Q versions.
But not nearly as drastically. The mobile versions (not MaxQ) generally take ~10% hit. The example you used was a 5700XT which typically boosts to 1980Mhz running at 1200MHz. That's a huge performance hit.
 
Joined
Feb 20, 2019
Messages
299 (1.08/day)
System Name PowerEdge R730 DRS Cluster
Processor 4x Xeon E5-2698 v3
Cooling Many heckin screamy bois
Memory 480GB ECC DDR4-2133
Video Card(s) Matrox G200eR2
Storage SD Card. Yep, really no other local storage.
Display(s) It's probably a couple of boring Dell Ultrasharps and a sacrificial laptop.
Case 39U 6-rack server room with HEVC and 44KVA UPS
Mouse Maybe
Keyboard Yes!
Software ESXi 6.5 U3
Benchmark Scores I once clocked a Celeron-300A to 564MHz on an Abit BE6 and it scored over 9000.
But not nearly as drastically. The mobile versions (not MaxQ) generally take ~10% hit. The example you used was a 5700XT which typically boosts to 1980Mhz running at 1200MHz. That's a huge performance hit.
The 1980MHz boost is a marketing number, nothing close to the actually the real-world gaming clock of about 1750MHz for me with my stock reference card. It depends on the load, but the game clock is entirely limited by the power draw at 225W with stock Wattman settings on a 5700XT.

But you're still missing the point: I am not talking about strictly limiting the GPU to 1200MHz. I'm saying that the GPU will clock itself down to 1200MHz automatically as part of normal behaviour via the driver whenever you're not GPU bottlenecked. If you don't understand how GPU load affects frequency and voltage on AMD cards going way back to Polaris 10, then you have three years of basic understanding to catch up on before I can explain the current topic to you.

Even if you can't get your head around that, it's irrelevant because I was only using the instantaneous power draw and clockspeed at a low-load moment as the basis to speculate the TDP of a smaller, 1200MHz Navi 14 part; Nobody would buy a 5700XT to lock it at 1200MHz all the time - that's plain stupid.
 
Last edited:
Joined
Dec 29, 2014
Messages
749 (0.42/day)
Nobody would buy a 5700XT to lock it at 1200MHz all the time - that's plain stupid.
My point exactly, which is why I don't understand why you used it as an example of what AMD can acheive (low TDP and high performance) by low-clocking and low-volting mobile processors. They aren't going to take a big performance hit to get to 35W TDP, so if they even do it will be a weak card with modest clock reduction. I don't believe they'll be beating Nvidia in the mobile space with this generation, but they should be close. On Nvidia's side they have a 1050 (edit 1650) MaxQ that is 35W TDP ~2x faster than a MX150.
 
Last edited:
Joined
Feb 20, 2019
Messages
299 (1.08/day)
System Name PowerEdge R730 DRS Cluster
Processor 4x Xeon E5-2698 v3
Cooling Many heckin screamy bois
Memory 480GB ECC DDR4-2133
Video Card(s) Matrox G200eR2
Storage SD Card. Yep, really no other local storage.
Display(s) It's probably a couple of boring Dell Ultrasharps and a sacrificial laptop.
Case 39U 6-rack server room with HEVC and 44KVA UPS
Mouse Maybe
Keyboard Yes!
Software ESXi 6.5 U3
Benchmark Scores I once clocked a Celeron-300A to 564MHz on an Abit BE6 and it scored over 9000.
Jesus! Did you even read the article? That's exactly what they ARE doing!

The reason I mentioned Navi at 1200MHz is because the article clearly states the 5300M will run at just 1181MHz. I'm telling you that Navi 10 runs at those speeds at just 750mv already, as well as the power draw at those speeds.

The only thing I'm speculating is how much power will be saved by using a smaller chip.

/facepalm
 
Last edited:

bug

Joined
May 22, 2015
Messages
6,718 (4.08/day)
Processor Intel i5-6600k (AMD Ryzen5 3600 in a box, waiting for a mobo)
Motherboard ASRock Z170 Extreme7+
Cooling Arctic Cooling Freezer i11
Memory 2x16GB DDR4 3600 G.Skill Ripjaws V (@3200)
Video Card(s) EVGA GTX 1060 SC
Storage 500GB Samsung 970 EVO, 500GB Samsung 850 EVO, 1TB Crucial MX300 and 3TB Seagate
Display(s) HP ZR24w
Case Raijintek Thetis
Audio Device(s) Audioquest Dragonfly Red :D
Power Supply Seasonic 620W M12
Mouse Logitech G502 Proteus Core
Keyboard G.Skill KM780R
Software Arch Linux + Win10
Jesus! Did you even read the article? That's exactly what they ARE doing!

The reason I mentioned Navi at 1200MHz is because the article clearly states the 5300M will run at just 1181MHz

/facepalm
Maybe, but they won't include 2300+ stream processors, 14Gbps VRAM or a 256bit memory interface. It's not simply a downclocked 5700.
 
Joined
Feb 20, 2019
Messages
299 (1.08/day)
System Name PowerEdge R730 DRS Cluster
Processor 4x Xeon E5-2698 v3
Cooling Many heckin screamy bois
Memory 480GB ECC DDR4-2133
Video Card(s) Matrox G200eR2
Storage SD Card. Yep, really no other local storage.
Display(s) It's probably a couple of boring Dell Ultrasharps and a sacrificial laptop.
Case 39U 6-rack server room with HEVC and 44KVA UPS
Mouse Maybe
Keyboard Yes!
Software ESXi 6.5 U3
Benchmark Scores I once clocked a Celeron-300A to 564MHz on an Abit BE6 and it scored over 9000.
Maybe, but they won't include 2300+ stream processors, 14Gbps VRAM or a 256bit memory interface. It's not simply a downclocked 5700.
For the third time, that is only the basis for my speculation, which is why I think the 5300M will have a pretty tiny TDP.

I'm counting on the fact that Navi 14 with fewer transistors will consume much less at 1181MHz than Navi 10 does.

Obviously I'm using Navi 10 figures because those are the only Navi cards we have to pull the power consumption numbers from at the moment....
 
Joined
Dec 29, 2014
Messages
749 (0.42/day)
Jesus! Did you even read the article? That's exactly what they ARE doing!
"Radeon RX 5300 ("Navi 14 XL"), with a peak frequency of 1448 MHz, and its mobile sibling, the Radeon RX 5300M ("Navi 14 XLM"), with lower clock speeds of 1181 MHz. "

They are taking a chip that maxes at 1448 MHz and clocking it down to 1181 MHz. That's not the same as taking a 1980MHz chip and dropping it down to that level, is it? You won't get anywhere near the same efficiency gains. Do you think the 5300M will be designed to run at 750mv? I don't know if it will be a 35W card (definitely possible), but I doubt it will be as fast as a 1050 (edit 1650) MaxQ if it is. Still, it will be good to see AMD being competitive in laptops for a change.

I have zero problem with you posting what you did as some random speculation based on very little, I just thought there might be more to it.
 
Last edited:
Joined
Feb 20, 2019
Messages
299 (1.08/day)
System Name PowerEdge R730 DRS Cluster
Processor 4x Xeon E5-2698 v3
Cooling Many heckin screamy bois
Memory 480GB ECC DDR4-2133
Video Card(s) Matrox G200eR2
Storage SD Card. Yep, really no other local storage.
Display(s) It's probably a couple of boring Dell Ultrasharps and a sacrificial laptop.
Case 39U 6-rack server room with HEVC and 44KVA UPS
Mouse Maybe
Keyboard Yes!
Software ESXi 6.5 U3
Benchmark Scores I once clocked a Celeron-300A to 564MHz on an Abit BE6 and it scored over 9000.
"Radeon RX 5300 ("Navi 14 XL"), with a peak frequency of 1448 MHz, and its mobile sibling, the Radeon RX 5300M ("Navi 14 XLM"), with lower clock speeds of 1181 MHz. "

They are taking a chip that maxes at 1448 MHz and clocking it down to 1181 MHz. That's not the same as taking a 1980MHz chip and dropping it down to that level, is it? You won't get anywhere near the same efficiency gains. Do you think the 5300M will be designed to run at 750mv? I don't know if it will be a 35W card (definitely possible), but I doubt it will be as fast as a 1050 MaxQ if it is. Still, it will be good to see AMD being competitive in laptops for a change.

I have zero problem with you posting what you did as some random speculation based on very little, I just thought there might be more to it.
Your thinking is good, but you're missing the fact that lower cost cards don't get the fancy VRMs, nor as many phases - so the 1448MHz upper range is a side effect of that inferior regulation, and desire to keep component costs lower - not because the silicon can't clock that high.

The architecture and manufacturing process are the same as the 5700 series, so clock/voltage curves should be the same, if not slightly better thanks to tweaks made from lessons learned with Navi 10 - at least within the range of clocks that Navi 14 can run at given the cheaper components.

This is still just speculation that could easily be wrong, but I believe Lisa is serious about cracking the mobile market. Nvidia and Intel have dominated there for, well... forever.
 
Last edited:
Joined
Mar 9, 2016
Messages
49 (0.04/day)
Location
Frankfurt am Main
System Name A.M.D
Processor Ryzen 3600
Motherboard MSI B450M Mortar MAX
Cooling EK Fluid Gaming A240
Memory ADATA XPG Spectrix D80 16GB 3200Mhz
Video Card(s) WIP
Storage Samsung 960 Evo 500GB
Display(s) AOC G2460Fq
Case Phanteks Enthoo Evolv MATX
Power Supply Corsair RMx 550
Mouse Razer Basilisk
Keyboard Razer Blackwindows Tournament Edition Chroma Stealth
Software Win 10 64bit
RX 5500, 5500 XT, 5600. Where are you? Black Friday and Christmas are here, Nvidia with 1660s and 1650s have a market advantage. AMD?
 
Joined
Aug 8, 2019
Messages
64 (0.60/day)
But not nearly as drastically. The mobile versions (not MaxQ) generally take ~10% hit. The example you used was a 5700XT which typically boosts to 1980Mhz running at 1200MHz. That's a huge performance hit.
My P2000 mobile would like a word, it's basically cut down by 33% compared to the desktop version.

Desktop P2000: 1024 Cores, 5GB GDDR5 160bit memory for 160GB/s, 3.0 Tflops FP32 and 75W
1076mhz boost 1480mhz core and 7008mhz mem
4.4million transistors
Equal to an RX570


Vs

Mobile P2000: 768 Cores, 4GB GDDR5 128bit memory for 96GB/s, 2.4 Tflops FP32 and 50W
1557mhz boost 1607mhz core and 6008mhz mem
3.3million transistors
23% slower than a RX570

That's not 10%...

That's 23% in perf alone.

94.7Gflops vs 77.1Gflops of FP64 calculation power. Hmm I can't seem to find where it's only 10% less...
 
Joined
Feb 20, 2019
Messages
299 (1.08/day)
System Name PowerEdge R730 DRS Cluster
Processor 4x Xeon E5-2698 v3
Cooling Many heckin screamy bois
Memory 480GB ECC DDR4-2133
Video Card(s) Matrox G200eR2
Storage SD Card. Yep, really no other local storage.
Display(s) It's probably a couple of boring Dell Ultrasharps and a sacrificial laptop.
Case 39U 6-rack server room with HEVC and 44KVA UPS
Mouse Maybe
Keyboard Yes!
Software ESXi 6.5 U3
Benchmark Scores I once clocked a Celeron-300A to 564MHz on an Abit BE6 and it scored over 9000.
<example of dramatically underclocked mobile GPU>
<examples of barely underclocked mobile GPU>
You're both right. The issue is that Nvidia don't need to downclock their mobile GPUs because they've always been pretty good at choosing the sweet spot on the performance/voltage curves of their silicon.

AMD, on the other hand, typically push their silicon to the limits of their voltage and clockspeeds. If you bolt a fat cooler onto it, that works great for desktops, but not for mobile.

We'll have to see what AMD do with their mobile parts, but if they want to compete with Nvidia right now, they can't be too aggressive with clocks and voltages. The 1181MHz indicates that they're aiming for a low TDP, and at a guess from Navi drivers, that is the highest speed the 7nm silicon will go whilst operating at its minimum voltage*. Beyond that, more voltage is required which exponentially increases power use in order to tap into the performance offered by higher clockspeeds.

More importantly, it's likely that RX 5300M solutions will be in the market when Nvidia makes the transition to 7nm, so (and I'm REALLY guessing now), I'd think Lisa is actually trying to make something that is perhaps not the most profitable thing in the world, but that offers a low-enough TDP to be appealing to OEMs for design wins even once Nvidia's mobile 7nm parts are on the table.

* - (to clarify, the GPU cores can run at lower voltage than the driver-exposed minimum voltage, but I believe power savings beyond the exposed minimum are minimal because the VRAM won't operate at sensible frequencies at all below a certain voltage. On desktop where there are no power concerns at idle states, there is no point even exposing these lower voltages to the user. You get 750mv as the baseline because that's what the GDDR6 needs to operate, and if the GPU draws 7W instead of 6W at idle because of this, nobody is going to care - especially not in a machine with a ~200W GPU and ~125W CPU).
 
Last edited:
Joined
Feb 8, 2017
Messages
42 (0.04/day)
Are you saying one node advantage is not enough for AMD to equal Nvidia on performance? :roll:
LOL, why are you playing to be dumb?

He is talking about wafers quantity, AMD can't afford a big gpu with large die size and waste precious wafers, when the RX 5700 and 5700xt are very small, in fact they are half the size of Nvidia's equivalent cards, so they are saving money not only on die size, but also amount of gpu's produced.

AMD is literally beating Nvidia's big guns with half the size GPU's, that is extremely impressive.
 
Joined
Sep 29, 2011
Messages
209 (0.07/day)
Location
Ottawa, Canada
System Name Current Rig
Processor AMD Ryzen 7 1700@3.95GHz
Motherboard Asus X370 Crosshair VI
Cooling Arctic Cooling 240mm
Memory 2x8GB DDR4-3200 G.Skill Trident Z RGB
Video Card(s) Gigabyte Windforce R9 290 (bios flashed to 1050MHz core
Storage 1TB SSD
Display(s) 3x22" LG Flatron (eyefinity)
Case Cooler Master Storm Striker
Power Supply Antec True Power 750w
Yeah. The 2060S/2070/2070S represent vastly more compelling options, have additional tech and are even more stable in their drivers.

As to these, they're late, on a newer denser process and are already having their prices limited due to poor competition. Plus if the 5700/5700XT are any indication their thermals won't be brilliant either

What's even sadder is that Nvidia will absolutely obliterate them when they move over to 7nm+ via Ampere. No wonder AMD is accelerating 5nm. They couldn't design an efficient uarch for their lives.
"...are even more stable in their drivers." OK, right there, you're obviously a paid shill. You know it's a bold-faced lie. AMD's drivers are superior to nVidia's, and have been for at least a couple of years. Their Radeon control panel is also dramatically more modern and easy to use.

Thank you very much for your input, nVidia Public Relations. Now actual users will carry on the conversation.
 
Top