[Rumor, Tweaktown] AMD to launch next-gen Navi graphics cards at E3

[Glo.]

What makes people NOT believe that RX 5500 will catch up to GTX 1660 Super?

 

[jpiniero]

What makes people NOT believe that RX 5500 will catch up to GTX 1660 Super?

AMD 's earlier benchmarks. It sounded more like between 580 and 590 performance.

It'll be cheaper though.

 

[alexruiz]

What makes people NOT believe that RX 5500 will catch up to GTX 1660 Super?

Good question.
I do not think it will catch it because of sustained clocks and boost.

I remember reading a a review comparing the "IPC" of an RTX CUDA core to a Navi 10 SP core at the same normalizad clock speed. Navi 10 was already ahead of Turing at the same clock.
Taking parts of comparable specs (RX 5700 and RTX 2070, 2304 CUDA/SP, 144 TMUs, 64 ROPs) the RTX 2070 is still faster, and not because the Turing core is faster, it is because it sustains higher clocks. If we could get both parts at the same clocks and boosts, the RX 5700 would be faster.

Going back to Navi 14 / GTX 166x.
The specs are comparable, but again I think the sustained clocks will be the difference.
I would love to be wrong though

 

[VirtualLarry]

Going back to Navi 14 / GTX 166x.
The specs are comparable, but again I think the sustained clocks will be the difference.
I would love to be wrong though

That can also be due to AIB cooler and VRM choices. Some choose to overbuild, some not. I agree that NV is largely sand-bagging when they set their base clocks, because their cards almost always stay boosted, to some extent.

I think that was what AMD's intent with introducing "game clock" specs, which is supposed to be the average clock that an AMD Navi card boosts to, during gaming sessions.

 

[lifeblood]

My speculation:
Radeon RX5300 (1408 SP, 1448MHz, 4GB part) ~ Radeon RX 570
Radeon RX 5500 (1408 SP, 1670MHz, 4GB part) ~ Radeon RX 590
Radeon RX 5500 XT (1408 SP, 1717MHz, 8GB part) ~ GTX 1660.
I do not think the 1408SP parts will catch the GTX 1660 super. A full die Navi 14 (RX 5500 XTX?) with 1536 SP would catch the 1660 super, and such part would be even faster than the GTX 1660 Ti.

Ok, I'll bite.

My speculation:
Radeon RX5300 (1408 SP, 1448MHz, 4GB part) > Between RX 560 and RX 570
Radeon RX 5500 (1408 SP, 1670MHz, 4GB part) ~ Radeon RX 580
Radeon RX 5500 XT (1536 SP, 1717MHz, 8GB part) ~ Radeon RX 590

I expect to see the 5500 a week before Thanksgiving. I don’t know if we’ll see the 5500 XT this year or next, same with the 5300 cards. I’m still expecting a future RX 5600 (Navi 12?) to fill the $250 price spot but I do not expect to see it until 2020.

 

[lifeblood]

What makes people NOT believe that RX 5500 will catch up to GTX 1660 Super?

The fact they (AMD) compared it with the 1650, not the 1660, and definitely not the 1660 Ti. If it was performing at that level they would be comparing it to the 1660.

 

[Glo.]

The fact they (AMD) compared it with the 1650, not the 1660, and definitely not the 1660 Ti. If it was performing at that level they would be comparing it to the 1660.

Does it perform then on GTX 1650 level?

You see where this logic is flawed?

Can you sell 4 GB GPU in 2019 for the price of GTX 1660? No. Can you sell 4 GB GPU in 2019 for the price of GTX 1650? Yes.

 

[alexruiz]

Ok, I'll bite.

My speculation:
Radeon RX5300 (1408 SP, 1448MHz, 4GB part) > Between RX 560 and RX 570
Radeon RX 5500 (1408 SP, 1670MHz, 4GB part) ~ Radeon RX 580
Radeon RX 5500 XT (1536 SP, 1717MHz, 8GB part) ~ Radeon RX 590

I expect to see the 5500 a week before Thanksgiving. I don’t know if we’ll see the 5500 XT this year or next, same with the 5300 cards. I’m still expecting a future RX 5600 (Navi 12?) to fill the $250 price spot but I do not expect to see it until 2020.

We'll see.
I think you are too low in your estimates, but time will prove someone right... which might be neither of us

 

[lifeblood]

We'll see.
I think you are too low in your estimates, but time will prove someone right... which might be neither of us

I hope I'm wrong and your right.

 

[Glo.]

The same discussion before was that no way RX 5700 XT can be faster than RTX 2070. Then it was that it cannot be more effiecient, than RTX 2070.

Oh how that aged well.

 

[SPBHM]

the 1060S has access to 192bit GDDR6, the 5500 128bit GDDR6, that's probably worth some performance.

 

[Glo.]

So it appears that GTX 1660 Ti Max-Q on average 12-14% faster than RX 5500M, based on MSI Alpha tests.

And we still have to remember that RX 5500M is hampered by the CPU, and GTX 1660 Ti Max-Q was not(Notebookcheck has had Nvidia GPU laptops paired with Intel CPUs). That is not bad.

The most interesting result for me is Overwatch. GTX 1660 Ti Max-Q is just 6% faster.

https://www.notebookcheck.net/AMD-Radeon-RX-5500M-GPU.436556.0.html

A lot of games are showing extremely bad results for RX 5500M(slower than GTX 1650), which in real world should be complete BS, but knowing state of AMD drivers for this GPU - it is quite possible.

 

[Glo.]

This is my opinion what will happen with RX 5500.

RX 5500 4 GB: 150$ - GTX 1660 performance.
RX 5500 XT 8 GB: 199$ - GTX 1660 Super/GTX 1660 Ti performance.

 

[lifeblood]

The same discussion before was that no way RX 5700 XT can be faster than RTX 2070. Then it was that it cannot be more effiecient, than RTX 2070.

Oh how that aged well.

And I'm very happy to say I was quite surprised. AMD came back strong much to my delight. But your now missing the point. 5700 XT beat the RTX 2070, not the RTX 2070 Super. And in line with that same idea, RX 5500 will beat the GTX 1650 but not the 1650 Super, and RX 5500 XT (if it ever gets here) will beat the GTX 1660 but not the GTX 1660 Super.

If you’ve ever studied warfare you would be familiar with something called “Cycle of Operations” or tempo. That’s how fast you are able to operate. Ideally you want to take an action and be done with it and moved on to your next objective while your enemy is still responding to your previous action. That means his action is generally ineffective because the battle is already over. The British in the Falklands had a slow tempo because of cold and other factors, but they still beat the Argentines as their tempo was even slower and they simply got outmaneuvered.

Nvidia has a much faster tempo then AMD. Nvidia had their counter attack in place before AMD got their main attack rolling. First the 2070 & 2060 Supers were released before the 5700 series. Look at Anandtech’s release day benchmarks for the 5700 XT. The 2070 isn’t even listed, only the 2070 Super. Jump forward to today and the 1660 Super is being sold while the 5500 XT has not even been announced. And the 1650 Super sitting there announced but waiting on pricing decision while the retail 5500 has yet to be formally announced/released.

It’s nice to see AMD putting out winning CPU’s, they’ve definitely caught Intel with their pants down. But on the GPU front, while they’re put out excellent products, they’re still, unfortunately, in the second place position. They need to up their game.

 

[Glo.]

And I'm very happy to say I was quite surprised. AMD came back strong much to my delight. But your now missing the point. 5700 XT beat the RTX 2070, not the RTX 2070 Super. And in line with that same idea, RX 5500 will beat the GTX 1650 but not the 1650 Super, and RX 5500 XT (if it ever gets here) will beat the GTX 1660 but not the GTX 1660 Super.

Let me quote exactly what I believe will happen:

This is my opinion what will happen with RX 5500.

RX 5500 4 GB: 150$ - GTX 1660 performance.
RX 5500 XT 8 GB: 199$ - GTX 1660 Super/GTX 1660 Ti performance.

There also will be games in which RX 5500 XT will beat both Nvidia GPUs: GTX 1660 Super and 1660 Ti.

 

[lifeblood]

So it appears that GTX 1660 Ti Max-Q on average 12-14% faster than RX 5500M, based on MSI Alpha tests.

And we still have to remember that RX 5500M is hampered by the CPU, and GTX 1660 Ti Max-Q was not(Notebookcheck has had Nvidia GPU laptops paired with Intel CPUs). That is not bad.

The most interesting result for me is Overwatch. GTX 1660 Ti Max-Q is just 6% faster.

https://www.notebookcheck.net/AMD-Radeon-RX-5500M-GPU.436556.0.html

A lot of games are showing extremely bad results for RX 5500M(slower than GTX 1650), which in real world should be complete BS, but knowing state of AMD drivers for this GPU - it is quite possible.

Honestly I wouldn’t put any weight on those results. Laptops are so variable, each one changes the power, speed, have different cooling. Even if the mobile 5500 got it’s butt kicked up and down the hall I wouldn’t assume the retail 5500 would perform similarly.

The fact Nvidia pushed out the 1650/1660 Supers says far more than the laptop results.

 

[tajoh111]

So it appears that GTX 1660 Ti Max-Q on average 12-14% faster than RX 5500M, based on MSI Alpha tests.

And we still have to remember that RX 5500M is hampered by the CPU, and GTX 1660 Ti Max-Q was not(Notebookcheck has had Nvidia GPU laptops paired with Intel CPUs). That is not bad.

The most interesting result for me is Overwatch. GTX 1660 Ti Max-Q is just 6% faster.

https://www.notebookcheck.net/AMD-Radeon-RX-5500M-GPU.436556.0.html

A lot of games are showing extremely bad results for RX 5500M(slower than GTX 1650), which in real world should be complete BS, but knowing state of AMD drivers for this GPU - it is quite possible.

You just have to look at the specs and you will see your comparison is off by quite a bit.

The Max Q variants of the chips have a lower TDP than regular laptop variants and if we look at the clocks of the max q, we will see they are way further depressed than the rx5500m. 1130-1335mhz for the max q and for the 5500m, 1327-1645mhz. This is reflected by power consumption in the laptops. The 1660 ti max q laptop consumed 30% less power than the 5500m. The closer match is the regular 1660 mobile which has clocks in the 1455 -1590mhz range.

The final nail on the coffin is the TGP of the chips. The GTX 1660 ti is rated at 80watts, the 5500m at 85watts and the max q is rated at 60watts.

If your going to extrapolate from laptop results, you need to take into account how much further clocks can go from their mobile variants.

 

[Veradun]

If your going to extrapolate from laptop results, you need to take into account how much further clocks can go from their mobile variants.

And this applies to AMD cards as well

 

[Glo.]

You just have to look at the specs and you will see your comparison is off by quite a bit.

The Max Q variants of the chips have a lower TDP than regular laptop variants and if we look at the clocks of the max q, we will see they are way further depressed than the rx5500m. 1130-1335mhz for the max q and for the 5500m, 1327-1645mhz. This is reflected by power consumption in the laptops. The 1660 ti max q laptop consumed 30% less power than the 5500m. The closer match is the regular 1660 mobile which has clocks in the 1455 -1590mhz range.

The final nail on the coffin is the TGP of the chips. The GTX 1660 ti is rated at 80watts, the 5500m at 85watts and the max q is rated at 60watts.

If your going to extrapolate from laptop results, you need to take into account how much further clocks can go from their mobile variants.

And what data have you to judge the actual clock speeds of RX 5500M in that specific NoteBook?

None. Notebookcheck is a failure on this front. Their testing methodology is also flawed, simply dividing the minum and maximum FPS in the middle.

 

[Glo.]

RX 5300M - 4 GB GDDR6, 20 CUs.
RX 5500M - 4/8 GB GDDR6, 24 CUs.

Both models are in new MacBook Pro 16.

 

[amrnuke]

RX 5300M - 4 GB GDDR6, 20 CUs.
RX 5500M - 4/8 GB GDDR6, 24 CUs.

Both models are in new MacBook Pro 16.

There are a lot of lines of analysis here. With that 9880 Intel chip at 35W cTDP and 45W TDP, and if it still operates on the USB-C power (100W max) then this makes the power draw of the 5500M interesting to say the least, at 24 CUs.

1) They move away from USB-C charging.
2) They assume Intel chip will have a short boost period, so it'll draw 10-15W most of the time except for short bursts, and as a result they can dedicate more power to GPU when needed, and vice versa - put the GPU on low-power if the CPU needs boost.
3) They can draw extra power temporarily from the battery and throttle when the battery is low.

Probably a lot more ways to handle it, but in any case, the 5500M has a really narrow power window, which bodes, IMO, fairly well for the retail 5500 at some point.

 

[Glo.]

RX 5300M - 4 GB GDDR6, 20 CUs.
RX 5500M - 4/8 GB GDDR6, 24 CUs.

Both models are in new MacBook Pro 16.

The GPUs also have 189 GB/s bandwidth available. 3 GB/s less than Vega 20 Pro had.

 

[jpiniero]

There are a lot of lines of analysis here. With that 9880 Intel chip at 35W cTDP and 45W TDP, and if it still operates on the USB-C power (100W max) then this makes the power draw of the 5500M interesting to say the least, at 24 CUs.

Apple probally ensures that the power doesn't go over 96W, but the GPU could be allowed to use maybe 70W if the CPU isn't doing much. So I don't know if you can gleam too much about the 5500M's performance from just that.

Should add that it now includes a bigger battery (100 wh, up from 83)

 

[Glo.]

Radeon Pro 5500M has up to 1.3 GHz max turbo clock speeds.

Its 50W GPU.

1536 ALUs, 1.3 GHz, 189 GB/s of bandwidth, 4 TFLOPs of compupte power.

 

[soresu]

1536 ALUs, 1.3 GHz, 189 GB/s of bandwidth, 4 TFLOPs of compupte power.

About 5 TFLOPS gaming power from GCN perspective then, not bad for 50W.

I wonder how much juice it would take to match 580 perf....