AMD Radeon RX 5600 XT Features 2,304 Stream Processors

[btarunr]

AMD's upcoming Radeon RX 5600 XT graphics card features the same exact stream processor count as the $350 RX 5700, according to a leaked specs sheet of a an AIB partner's custom-design graphics card. With a stream processor count of 2,304, it's safe to assume that the RX 5600 XT is based on the same 7 nm "Navi 10" silicon as the RX 5700 series. What set the RX 5600 XT apart from the RX 5700, besides lower clock-speeds, is the memory subsystem, which is severely stripped down. The Radeon RX 5600 XT will be equipped with 6 GB of GDDR6 memory across a 192-bit wide memory interface. What's more, the memory ticks at 12 Gbps, compared to 14 Gbps on the RX 5700 series.

With these specs, the RX 5600 XT has 288 GB/s of memory bandwidth at its disposal, same as NVIDIA's GeForce GTX 1660 Ti. In contrast, with 8 GB of 256-bit GDDR6 running at 14 Gbps, the RX 5700 enjoys 448 GB/s. The specs sheet suggests that AMD has also dialed down the engine clock-speeds (GPU clocks) a bit, with up to 1620 MHz boost, up to 1460 MHz gaming, and 1235 MHz base. With these specs, it's highly likely that the RX 5600 XT outperforms the GTX 1660 Ti and gets close to the RTX 2060. It all boils down to pricing. The RX 5500 XT is a decent GTX 1650-series alternative with a lukewarm price thanks to NVIDIA's aggressive product-stack management by getting its partners to lower prices of the GTX 1660 and GTX 1660 Super. It would be interesting to see if AMD can outfox NVIDIA in the sub-$300 market.



View at TechPowerUp Main Site

 

[Durvelle27]

I was wondering when this was going to get posted as it's been posted on other sites for hours.

 

[Hyderz]

so the perf sits around a gtx 1070ti and 1080 , noice

 

[FordGT90Concept]

But, but, but where is Navi 12? If they wait too long, it's going to have to contend with Ampere instead of Turing. Maybe Ampere has been delayed (referring to Turing existing) and Navi 12 announced last because of yield problems with large chips on 7nm.

 

[ShurikN]

so the perf sits around a gtx 1070ti and 1080 , noice

It would be nice but I doubt it. It's too gimped in terms of core clock and memory bandwidth. Maybe it'll have more overclocking headroom tho.
If they can avoid the 5500XT pricing blunder, and put it at around 250-260 USD mark would be great. That would make the 1660super and 1660Ti a tough buy.

 

[macrobe]

Logically it should have been by 2048 stream processors and 8GB, but they are clearly chasing the price.

 

[Super XP]

This release makes total sense, and a good way to use chips that didn't make it to the RX 5700XT's.
AMD is using up those thousands of Navi 10 dies that didn’t meet the requirements to qualify for the 5700 SKU basically.
Smart Move,

 

[macrobe]

This release makes total sense, and a good way to use chips that didn't make it to the RX 5700XT's.
AMD is using up those thousands of Navi 10 dies that didn’t meet the requirements to qualify for the 5700 SKU basically.
Smart Move,

Nope, the savings come from the 2 memory chips less.

 

[Super XP]

Nope, the savings come from the 2 memory chips less.

I'm only posting what is readily available online.

 

[tfdsaf]

This release makes total sense, and a good way to use chips that didn't make it to the RX 5700XT's.
AMD is using up those thousands of Navi 10 dies that didn’t meet the requirements to qualify for the 5700 SKU basically.
Smart Move,

Not really, since why would they release it with the same amount of shaders as the RX 5700? Wouldn't it have less if its partially defencive? It seems to me this is essentially failed RX 5700XT with 2 less GDDR6 memory chips as additional cost cutting and with 192bit bus that is additional die space, so more profits for AMD.Personally I don't mind this, but only if its great value and for me this would have to be priced at $250 and be faster than the GTX 1660ti.

 

[Super XP]

Not really, since why would they release it with the same amount of shaders as the RX 5700? Wouldn't it have less if its partially defencive? It seems to me this is essentially failed RX 5700XT with 2 less GDDR6 memory chips as additional cost cutting and with 192bit bus that is additional die space, so more profits for AMD.Personally I don't mind this, but only if its great value and for me this would have to be priced at $250 and be faster than the GTX 1660ti.

You could very well be correct. And I agree this GPU needs to be priced properly as you suggest or it won't serve its intended purpose.

 

[Xaled]

but only if its great value and for me this would have to be priced at $250 and be faster than the GTX 1660ti.

I won't be optimistic to that degree. It should either be faster or cheaper. Lisa SU's gpu price policey is just not good and isn'tworking. imo that's why AMD lost market share in last quarter.

 

[Super XP]

I won't be optimistic to that degree. It should either be faster or cheaper. Lisa SU's gpu price policey is just not good and isn'tworking. imo that's why AMD lost market share in last quarter.

I have to admit, the GPU costs are a tad on the high side in relation to the competition. Though I also find Nvidia GPU's overpriced.

But I agree AMD needs to rethink its overall GPU pricing.

 

[Casecutter]

where is Navi 12?

I think AMD/RGT will have the big chip by end of Q2 and then re-name and re-vamp the whole stack, moving these Navi parts down in market segment and price. I understand is Nvidia ampere expected date is about that same time. If Nvidia's big chip is late(r) AMD/RGT has a little window to... "make hay".

Not really, since why would they release it with the same amount of shaders as the RX 5700? Wouldn't it have less if its partially defencive? It seems to me this is essentially failed RX 5700XT with 2 less GDDR6 memory chips as additional cost cutting and with 192bit bus that is additional die space, so more profits for AMD.Personally I don't mind this, but only if its great value and for me this would have to be priced at $250 and be faster than the GTX 1660ti.

I think they surgically cut shader count of RX 5700 across all shader engines that way they maintain the memory bus. On this I understood they lop-off one of the 4 complete shader engines in doing so lose 64-bit of the memory (leading to 192-Bit), and with it a 25% reduction in L2.

 

[lukart]

If AMD plays this card right, Nvidia might have big trouble selling from 250~450$ price range... That must hurt Nvidia as it's the majority of the market.
Promising card indeed.

 

[OC-Ghost]

I dont understand these specs, butchering 5700 market with a slap in the face with slower memory (even 5500 has 14Gbps). And this Asrock is a Challenger OC model with slow/original launch specs?

 

[EarthDog]

I dont understand these specs, butchering 5700 market with a slap in the face with slower memory (even 5500 has 14Gbps). And this Asrock is a Challenger OC model with slow/original launch specs?

??

There is massive gap in both performance and price in the amd product stack, bud. Makes complete sense.

 

[Casecutter]

If the truth be known the actual "performance" difference in the using 12 Gbps, compared to 14 Gbps with 192-Bit, is not that big of loss and it might be cheaper and plentiful memory so AMD/RGT see's it as good compromise to keep the margins inline while keep the $/perf competitive. If AMD/RGT can get it to be between the Vega 56-64 with a $250 MSRP straight out of the gate, they have a great opportunity to real lead in a 1440p card for improving FreeSync panel sales which are now really the upgrade path.

 

[Cheeseball]

With the 5500 XT (8 GB) going for $199 on average now, they need to strictly launch this at $250 for it to be a 1660 Ti/RTX 2060 non-Super competitor to match what the 5700 XT is to the 2070 Super.

 

[Fluffmeister]

Yeah it's a product aimed at a price/performance target, and being the angels that are AMD this thread isn't full of "GIMPED" comments. Win win.

 

[cucker tarlson]

But, but, but where is Navi 12? If they wait too long, it's going to have to contend with Ampere instead of Turing. Maybe Ampere has been delayed (referring to Turing existing) and Navi 12 announced last because of yield problems with large chips on 7nm.

seriously,who cares about big navi on rdna1 at this point,that would only compete with cards that have been out for 1.5 year.should drop now or never.
rdna2 and ampere are coming.

This release makes total sense, and a good way to use chips that didn't make it to the RX 5700XT's.
AMD is using up those thousands of Navi 10 dies that didn’t meet the requirements to qualify for the 5700 SKU basically.
Smart Move,

smart move,lol,this is the same chip with less vram
it makes absolutely no sense for amd to sell a sweetspot 1440p card like 5700 with 6gb vram and a heavily gimped bandwidth,they should sell them as 5700s
the only thing they save here is 2gb vram,if this ends up too close to 5700 price wise nobody's gonna buy it.

There is massive gap in both performance and price in the amd product stack, bud. Makes complete sense.

right,but to sacrifice a better die to gimp the memory and sell it cheaper,man,that's not a good approach.
a smaller die with 8 gigs and better bandwidth would be a lot better

 

[Casecutter]

I suppose it's a more honest "approach" than Nvidia's GTX 970 with gimped memory and then advertising as full 4Gb and selling it at a stellar price.

I don't think RGT/AMD would sacrifice the memory Bus to save on only the cost of 2Gb of memory, there's probably a reason in quickly fusing off one complete shader engine, and then only needing to surgically cut a few stream processors in the remaining 3 engines. Given their New Compute Unit Design the time/effort to sit there and effectively gimp almost completely the shader engine, which probably leaves less of the use of that segments memory controller is easier. It's faster and lowers the chance in destroying the whole chip, just to fuse-off a complete shader engine.

Now sure the lack of 2Gb of "top-shelf" 14Gpbs "expensive" memory (and some are predicting GDDR6 prices to rise) but what if it aids in warding off supply issues, from perhaps price-hikes, while still deliver perf/price other cards won't if GDDR6 issues arise. I think AMD/RGT is making the most of their cramped TSCM production and this will provide a viable competitor and a $250 price point, until all the CPU/GPU 7nm production they have levels out. From Consoles, to Apple parts... to Ryzen 3 AMD/RGT have to juggle what wafer starts go where.

Against Nvidia to be perhaps 30% better than working off older 12nm process TU116; there's the GTX 1660 with 6Gb of GDDR5 from the same 192-Bit. Effectively very much the same vain as what AMD/RGT is doing now, but giving it 12Gpbs GDDR6 effectively doubles the bandwidth. If able to hold a $250 MSRP AMD/RGT verses that TU116 original MSRP it will be just 10% more in price that's a great value. From that understanding... those who paid-up for a GTX 1660 last April got swindled, paying "Top Self" price for process/technologies that were vintage, though computer part aren't like fine wine. Then vying the GTX 1660 Super or TI using GDDR6 (and same 12nm TU116) if the RX 5600 XT is 10-15% faster then it aligns perfectly with those. Then "Top" GTX TU116 parts are more often priced higher than the Super's $230 MSRP, while GTX 1660 Ti price at $270 and up here in the US today. AMD/RGT is in a good spot...

 

[vega22]

Got to assume these chips didn't pass the binning process to be the better chips or had defects that require the drop in mem bandwidth to function.

That being said this card makes sense. The entry cards get a stronger core than they would of otherwise and amd off loads chips that otherwise would of been waste.

Price: performance....time will tell how that stacks up I guess.

Who knows, maybe if these outsell expectations we will be flashing them with mod'd BIOS for full power limits?

 

[danbert2000]

This could potentially be a very energy efficient card, what with the reduced frequencies (and voltages) and fewer memory chips. I'm guessing that AMD had more memory controller defects than shader defects and was able to keep the yield high at the 5700 level of shaders. Or maybe there were some 5700s that couldn't clock as high, or could but with too much voltage, and rather than put out a card that was just a 5700 with lower clocks, they decided to cut a perfectly good memory controller in order to leave out some GDDR6 chips and reduce the VRM complexity by capping the voltage to the chip.

In either case, wide and slow means that the efficiency should be good. And if they are doing this to reuse some failed 5700s rather than just cutting a good chip for market segmentation, then it is a good way to recoup the 7nm fab costs.